bamboo_engine/session_app/

child_completion_coordinator.rs

//! Child-session completion coordinator.
//!
//! Receives terminal child runner notifications from `bamboo-engine`, updates
//! durable parent wait state, and notifies the parent of each terminal outcome.
//! The configured wait policy independently decides when its wait is released.

use std::collections::HashMap;
use std::sync::{Arc, Mutex as StdMutex, OnceLock, RwLock as StdRwLock, Weak};
use std::time::Duration;

use bamboo_domain::poison::PoisonRecover;
use bamboo_domain::{
    AgentHookPoint, HookPayload, HookToolOutcome, SessionActivationPolicy, SessionChildOutcome,
    SessionMessageBody, SessionMessageContent, SessionMessageEnvelope, SessionMessageId,
    SessionMessageKind, SessionMessageSource, SessionProviderMessage,
};

use crate::execution::{
    reserve_runner_core, reserve_session_execution, spawn_session_execution, AgentRunner,
    AgentStatus, ChildCompletion, ChildCompletionHandler, ChildCompletionSource, ReserveOutcome,
    SessionExecutionArgs, SessionExecutionReservation, SessionExecutionReserveOutcome, SpawnJob,
    SpawnScheduler,
};
use crate::runtime::config::{BashResumeHook, GuardianSpawner, BASH_COMPLETION_RESUME_KIND};
use crate::runtime::guardian_state::{
    parse_guardian_verdict, read_guardian_config, read_guardian_state, write_guardian_state,
    GuardianVerdict,
};
use crate::Agent;
use async_trait::async_trait;
use bamboo_agent_core::storage::Storage;
use bamboo_agent_core::tools::ToolExecutor;
use bamboo_agent_core::{
    AgentEvent, BashCompletionInfo, BashCompletionSink, Message, Role, Session, SessionKind,
};
use bamboo_domain::session::runtime_state::{
    AgentRuntimeState, AgentStatusState, ChildWaitPolicy, WaitingForChildrenState,
};
use bamboo_llm::{Config, ProviderModelRouter, ProviderRegistry};
use bamboo_storage::LockedSessionStore;
use chrono::Utc;
use sha2::{Digest, Sha256};
use tokio::sync::{broadcast, RwLock};

use crate::model_config_helper::{
    resolve_fast_model, resolve_gold_config, resolve_provider_routing_key, GOLD_CONFIG_METADATA_KEY,
};
use crate::session_activation::{
    SessionActivationLaunch, SessionActivationReserveOutcome, SessionActivationSpawner,
};
use crate::session_app::child_session::result_projection::{
    child_result_identity, decode_report, fits_child_result_budget, plain_child_report_message,
    project_child_report, unavailable_child_result,
};
use crate::session_app::execute::consume_pending_clarification_resume;
use crate::session_app::provider_model::{persist_model_ref, session_effective_model_ref};
use crate::session_app::resume::{ResumeExecutionPort, ResumeSpawnRequest};
use crate::session_app::types::{ResumeConfigSnapshot, ResumeOutcome};

const AGENT_RUNTIME_STATE_METADATA_KEY: &str = "agent.runtime.state";
const RUNTIME_RESUME_MESSAGE_HIDDEN_KEY: &str = "hidden_from_ui";
const RUNTIME_RESUME_MESSAGE_KIND_KEY: &str = "runtime_kind";
const CHILD_COMPLETION_INLINE_FIELD_BYTES: usize = 48 * 1024;
const CHILD_COMPLETION_OVERSIZE_TAIL_BYTES: usize = 8 * 1024;

fn read_runtime_state(session: &Session) -> AgentRuntimeState {
    session
        .agent_runtime_state
        .clone()
        .or_else(|| {
            session
                .metadata
                .get(AGENT_RUNTIME_STATE_METADATA_KEY)
                .and_then(|raw| serde_json::from_str::<AgentRuntimeState>(raw).ok())
        })
        .unwrap_or_else(|| AgentRuntimeState::new(format!("{}-child-wait", session.id)))
}

fn write_runtime_state(session: &mut Session, runtime_state: &AgentRuntimeState) {
    session.agent_runtime_state = Some(runtime_state.clone());
    if let Ok(serialized) = serde_json::to_string(runtime_state) {
        session
            .metadata
            .insert(AGENT_RUNTIME_STATE_METADATA_KEY.to_string(), serialized);
    }
}

/// Re-read and prepare an activation target under the same per-session
/// persistence lock that commits the generic suspension clear.
///
/// The boolean is false for an unanswered human question or a respected
/// child/Bash wait in the latest durable snapshot; leave activation unreserved.
async fn prepare_session_inbox_activation(
    persistence: &LockedSessionStore,
    session_id: &str,
    interrupt_specific_wait: bool,
) -> std::io::Result<Option<(Session, bool)>> {
    let ready = Arc::new(std::sync::atomic::AtomicBool::new(false));
    let ready_for_mutation = ready.clone();
    let saved = persistence
        .update_runtime_config(session_id, move |latest| {
            // Inbox steering never answers a human question. Check under the
            // same final writer lock, including when an earlier user envelope
            // has already advanced the prefix's interrupt watermark.
            if latest.has_pending_question() {
                return;
            }
            let mut runtime_state = read_runtime_state(latest);
            let specifically_waiting = runtime_state.waiting_for_children.is_some()
                || runtime_state.waiting_for_bash.is_some();
            if specifically_waiting && !interrupt_specific_wait {
                return;
            }
            // Explicit steering interrupts only this reasoning gate. The
            // durable child/Bash wait remains owned so later terminal events
            // still have exactly one coordinator and can authorize their
            // staged outcomes. End-of-run bookkeeping re-suspends if that wait
            // is still present.
            runtime_state.status = AgentStatusState::Idle;
            runtime_state.suspension = None;
            write_runtime_state(latest, &runtime_state);
            latest.metadata.remove("runtime.suspend_reason");
            latest.updated_at = Utc::now();
            ready_for_mutation.store(true, std::sync::atomic::Ordering::Release);
        })
        .await?;
    Ok(saved.map(|session| (session, ready.load(std::sync::atomic::Ordering::Acquire))))
}

fn is_error_like(status: &str) -> bool {
    matches!(status, "error" | "timeout" | "cancelled")
}

/// Terminal child run statuses, as mirrored into the session index.
fn is_terminal_child_status(status: &str) -> bool {
    matches!(
        status,
        "completed" | "error" | "timeout" | "cancelled" | "skipped"
    )
}

fn child_completion_envelope(
    completion: &ChildCompletion,
    wait_registered_at: chrono::DateTime<Utc>,
    result: Option<String>,
    provider_message: &Message,
) -> SessionMessageEnvelope {
    let envelope = build_child_completion_envelope(
        completion,
        wait_registered_at,
        result.as_deref(),
        provider_message,
        false,
    );
    if serde_json::to_vec(&envelope).map_or(true, |bytes| {
        bytes.len() > bamboo_domain::SessionInboxLimits::default().max_payload_bytes
    }) {
        // UTF-8 byte caps do not bound JSON escaping or duplicate fields in
        // ChildOutcome/provider metadata. Fall back to digest-only summaries
        // before admission so an escaped or mixed result/error cannot strand
        // the parent wait at the SessionInbox payload limit.
        return build_child_completion_envelope(
            completion,
            wait_registered_at,
            result.as_deref(),
            provider_message,
            true,
        );
    }
    envelope
}

fn child_completion_id(
    completion: &ChildCompletion,
    wait_registered_at: chrono::DateTime<Utc>,
    result: Option<&str>,
    legacy_wait_identity: bool,
) -> SessionMessageId {
    fn identity(value: Option<&str>) -> serde_json::Value {
        match value {
            Some(value) if value.len() > CHILD_COMPLETION_INLINE_FIELD_BYTES => serde_json::json!({
                "oversized": true, "utf8_bytes": value.len(),
                "sha256": hex::encode(Sha256::digest(value.as_bytes())),
            }),
            value => serde_json::json!(value),
        }
    }
    SessionMessageId::stable(
        "session_child_completion",
        &serde_json::json!({
            "parent_session_id": completion.parent_session_id,
            "child_session_id": completion.child_session_id,
            "status": completion.status,
            "error": identity(completion.error.as_deref()), "result": identity(result),
            "wait_registered_at": (legacy_wait_identity || completion.source.is_none()).then_some(wait_registered_at),
            "terminal_source": completion.source,
        }),
    )
}

fn build_child_completion_envelope(
    completion: &ChildCompletion,
    wait_registered_at: chrono::DateTime<Utc>,
    result: Option<&str>,
    provider_message: &Message,
    compact: bool,
) -> SessionMessageEnvelope {
    fn bounded_terminal_field(
        label: &str,
        child_session_id: &str,
        value: Option<&str>,
        compact: bool,
    ) -> (Option<String>, serde_json::Value) {
        let Some(value) = value else {
            return (None, serde_json::Value::Null);
        };
        let oversized = value.len() > CHILD_COMPLETION_INLINE_FIELD_BYTES;
        let identity = if oversized {
            serde_json::json!({
                "oversized": true,
                "utf8_bytes": value.len(),
                "sha256": hex::encode(Sha256::digest(value.as_bytes())),
            })
        } else {
            serde_json::Value::String(value.to_string())
        };
        if !oversized && !compact {
            return (Some(value.to_string()), identity);
        }
        let digest = hex::encode(Sha256::digest(value.as_bytes()));
        let retrieval_view = if label == "error" {
            "diagnostics"
        } else {
            "result"
        };
        let inspection = serde_json::json!({
            "intent": "inspect", "target": child_session_id, "message": retrieval_view,
        });
        let recovery = if label == "error" {
            format!(
                "Check observed status and recovery diagnostics by calling SubAgent with {inspection}. \
                 Detailed execution errors remain available to the authenticated inspector."
            )
        } else {
            format!(
                "Read it in bounded slices by calling SubAgent with {inspection}; use message=\"messages\" for transcript previews. \
                 Follow returned cursors with a JSON-encoded view/cursor query in message, omitting unused query fields."
            )
        };
        let mut summary = format!(
            "Child {label} could not fit in the durable inline envelope ({} UTF-8 bytes, sha256={digest}). {recovery}",
            value.len()
        );
        if !compact {
            let mut tail_start = value
                .len()
                .saturating_sub(CHILD_COMPLETION_OVERSIZE_TAIL_BYTES);
            while tail_start < value.len() && !value.is_char_boundary(tail_start) {
                tail_start += 1;
            }
            summary.push_str("\n\nBounded tail:\n");
            summary.push_str(&value[tail_start..]);
        }
        (Some(summary), identity)
    }

    let (stored_result, _) =
        bounded_terminal_field("result", &completion.child_session_id, result, compact);
    let (stored_error, _) = bounded_terminal_field(
        "error",
        &completion.child_session_id,
        completion.error.as_deref(),
        compact,
    );
    let mut bounded_provider_message = provider_message.clone();
    let provider_oversized = serde_json::to_vec(provider_message)
        .map(|bytes| bytes.len() > CHILD_COMPLETION_INLINE_FIELD_BYTES)
        .unwrap_or(true);
    if provider_oversized || compact {
        // A short provider presentation can carry an important instruction or
        // guardian verdict. Keep it when only the typed fields' JSON escaping
        // forced a compact envelope.
        let mut content = if provider_oversized {
            format!(
                "Runtime notification: child session `{}` finished with status `{}`.",
                completion.child_session_id, completion.status
            )
        } else {
            bounded_provider_message.content.clone()
        };
        if let Some(result) = stored_result.as_deref() {
            content.push_str("\n\n");
            content.push_str(result);
        }
        if let Some(error) = stored_error.as_deref() {
            content.push_str("\n\n");
            content.push_str(error);
        }
        bounded_provider_message.content = content;
        if provider_oversized {
            bounded_provider_message.content_parts = None;
        }
        // The runtime resume also carries the raw error in metadata. Bounding
        // only its visible content leaves the durable inbox envelope oversized
        // and strands the parent's wait when admission rejects it.
        if compact {
            let original = bounded_provider_message
                .metadata
                .as_ref()
                .and_then(serde_json::Value::as_object);
            let mut metadata = serde_json::Map::new();
            for key in [
                RUNTIME_RESUME_MESSAGE_HIDDEN_KEY,
                RUNTIME_RESUME_MESSAGE_KIND_KEY,
                "child_session_id",
                "child_status",
                "child_final_response_included",
                "child_completion_source",
                "child_completion_source_sha256",
                "guardian_approved",
            ] {
                if let Some(value) = original.and_then(|metadata| metadata.get(key)) {
                    metadata.insert(key.to_string(), value.clone());
                }
            }
            if original.is_some_and(|metadata| metadata.contains_key("child_error")) {
                metadata.insert(
                    "child_error".to_string(),
                    stored_error
                        .as_ref()
                        .map_or(serde_json::Value::Null, |error| {
                            serde_json::Value::String(error.clone())
                        }),
                );
            }
            bounded_provider_message.metadata = Some(serde_json::Value::Object(metadata));
        } else if let Some(child_error) = bounded_provider_message
            .metadata
            .as_mut()
            .and_then(serde_json::Value::as_object_mut)
            .and_then(|metadata| metadata.get_mut("child_error"))
        {
            *child_error = stored_error
                .as_ref()
                .map_or(serde_json::Value::Null, |error| {
                    serde_json::Value::String(error.clone())
                });
        }
    }
    let body = SessionMessageBody::ChildOutcome(SessionChildOutcome {
        child_session_id: completion.child_session_id.clone(),
        status: completion.status.clone(),
        result: stored_result,
        error: stored_error,
        provider_message: Some(session_provider_message(&bounded_provider_message)),
    });
    SessionMessageEnvelope {
        // A committed source keeps one identity across wait release and re-arm.
        // Synthetic recovery completions remain tied to their authorizing wait.
        id: child_completion_id(completion, wait_registered_at, result, false),
        source: SessionMessageSource::Runtime {
            subsystem: "child_completion_coordinator".to_string(),
        },
        target_session_id: completion.parent_session_id.clone(),
        kind: SessionMessageKind::ChildOutcome,
        body,
        created_at: completion.completed_at,
        thread_id: None,
        in_reply_to: None,
        attempt: None,
        correlation_id: Some(format!("child_completion:{}", completion.child_session_id)),
    }
}

fn session_provider_message(message: &Message) -> SessionProviderMessage {
    SessionProviderMessage {
        content: SessionMessageContent {
            text: message.content.clone(),
            parts: message.content_parts.clone().unwrap_or_default(),
            skill_request: None,
        },
        metadata: message
            .metadata
            .as_ref()
            .and_then(serde_json::Value::as_object)
            .cloned()
            .unwrap_or_default(),
        never_compress: message.never_compress,
    }
}

// Repair is a read-model fact, not permission to notify a parent. Only an
// explicitly older launch generation may leave this negative fence behind.
fn repaired_child_requires_source(child: &Session) -> bool {
    match child.metadata.get("runtime.reconciled_terminal_generation") {
        Some(raw) => raw.parse::<u64>().map_or(true, |generation| {
            generation >= child.child_launch_generation()
        }),
        None => child
            .metadata
            .contains_key("runtime.reconciled_activation_fence"),
    }
}

/// Reconstruct the set of completed child session ids for a parent from the
/// session index (the single source of truth), folding in the child whose
/// completion event is being processed so a momentarily-lagging index can never
/// stall the parent's resume.
async fn derive_completed_child_ids(
    storage: &Arc<dyn Storage>,
    parent_session_id: &str,
    just_completed_child_id: &str,
) -> Vec<String> {
    let mut completed: Vec<String> = storage
        .list_child_run_statuses(parent_session_id)
        .await
        .unwrap_or_default()
        .into_iter()
        .filter(|(_, status)| status.as_deref().is_some_and(is_terminal_child_status))
        .map(|(id, _)| id)
        .collect();
    if !completed.iter().any(|id| id == just_completed_child_id) {
        completed.push(just_completed_child_id.to_string());
    }
    completed.sort();
    completed.dedup();
    completed
}

#[cfg(test)]
fn read_config_snapshot(config: &Arc<RwLock<Config>>, cached_config: &StdRwLock<Config>) -> Config {
    if let Ok(config_guard) = config.try_read() {
        let snapshot = config_guard.clone();

        if let Ok(mut cached_guard) = cached_config.try_write() {
            *cached_guard = snapshot.clone();
        }

        snapshot
    } else {
        cached_config
            .try_read()
            .map(|guard| guard.clone())
            .unwrap_or_default()
    }
}

// Bind the explicit Gold model to the account selected in this very resume.
// None means the same selection snapshot was embedder-managed; do not install
// a Host policy in that case. An incomplete managed proof is still Some and
// is rejected by the guard, never treated as an unmanaged selection.
fn make_child_resume_gold_resolver(
    config: Arc<RwLock<Config>>,
    provider_name: String,
    expected: Arc<dyn bamboo_llm::LLMProvider>,
    origin: Option<Arc<bamboo_llm::provider_registry::ManagedProviderOrigin>>,
) -> Option<crate::runtime::config::GoldTargetResolver> {
    let origin = origin?;
    Some(Arc::new(move |model, delegate| {
        bamboo_llm::gold_admission::guarded_gold_target_with_origin(
            config.clone(),
            origin.clone(),
            bamboo_domain::ProviderModelRef::new(provider_name.as_str(), model),
            expected.clone(),
            delegate,
        )
    }))
}

fn make_resume_auxiliary_model_resolver(
    config_handle: Arc<RwLock<Config>>,
    _config_snapshot: Config,
    provider_registry: Arc<ProviderRegistry>,
    provider_name_for_aux: String,
) -> Arc<dyn Fn() -> crate::AuxiliaryModelConfig + Send + Sync> {
    crate::model_areas::live_auxiliary_model_resolver(
        config_handle,
        provider_registry,
        provider_name_for_aux,
    )
}

/// Per-parent async locks that serialize concurrent `on_child_completed`
/// invocations for the same parent session.
///
/// Race eliminated: when `wait_for=Any` and two child sessions complete
/// simultaneously, both invocations load the parent with
/// `waiting_for_children=Some` before either persists the cleared state, so
/// both pass `wait_policy_satisfied`, both clear `waiting_for_children`, add a
/// duplicate resume message, and call `resume_parent` — a double resume.
/// Holding this per-parent `tokio::sync::Mutex` across the load-check-save
/// critical section makes the second caller observe the already-cleared state.
///
/// The inner `std::sync::Mutex` guards only the brief HashMap lookup/insert
/// (no await inside); the per-parent `tokio::sync::Mutex` is the one held
/// across the async critical section. Entries exist only while a holder or
/// waiter owns a `SessionResumeLock`; historical parent IDs are reclaimed.
fn parent_locks() -> &'static std::sync::Mutex<HashMap<String, Arc<tokio::sync::Mutex<()>>>> {
    static LOCKS: OnceLock<std::sync::Mutex<HashMap<String, Arc<tokio::sync::Mutex<()>>>>> =
        OnceLock::new();
    LOCKS.get_or_init(|| std::sync::Mutex::new(HashMap::new()))
}

/// Fetch (or create) the per-session async lock from [`parent_locks`]. Held
/// across the load-check-clear-resume critical section so the three resume
/// sources for one session — child completion, the loop-facing bash **push**
/// ([`BashCompletionSink::on_bash_completed`]), and the bash **backstop** poll
/// ([`ChildCompletionCoordinator::bash_self_resume`]) — can never double-resume.
/// The inner sync `Mutex` guards only the brief map lookup (no await inside).
pub(crate) fn session_resume_lock(session_id: &str) -> SessionResumeLock {
    let mut map = parent_locks().lock().recover_poison();
    let lock = map
        .entry(session_id.to_string())
        .or_insert_with(|| Arc::new(tokio::sync::Mutex::new(())))
        .clone();
    SessionResumeLock {
        session_id: session_id.to_string(),
        lock: Some(lock),
    }
}

/// The lease is constructed before awaiting the mutex, so cancelled waiters
/// also reclaim their registration. Lookup and last-owner removal use the same
/// brief registry lock; two live mutexes can never exist for the same ID.
pub(crate) struct SessionResumeLock {
    session_id: String,
    lock: Option<Arc<tokio::sync::Mutex<()>>>,
}

impl std::ops::Deref for SessionResumeLock {
    type Target = tokio::sync::Mutex<()>;
    fn deref(&self) -> &Self::Target {
        self.lock.as_ref().expect("live resume-lock lease")
    }
}

impl Drop for SessionResumeLock {
    fn drop(&mut self) {
        self.lock.take();
        let mut map = parent_locks().lock().recover_poison();
        if map
            .get(&self.session_id)
            .is_some_and(|lock| Arc::strong_count(lock) == 1)
        {
            map.remove(&self.session_id);
        }
    }
}

fn wait_policy_satisfied(
    policy: ChildWaitPolicy,
    wait_child_ids: &[String],
    completed_child_ids: &[String],
    latest_child_id: &str,
    latest_status: &str,
) -> bool {
    if wait_child_ids.is_empty() {
        return false;
    }

    match policy {
        ChildWaitPolicy::All => wait_child_ids
            .iter()
            .all(|id| completed_child_ids.iter().any(|completed| completed == id)),
        ChildWaitPolicy::Any => completed_child_ids
            .iter()
            .any(|id| wait_child_ids.iter().any(|wait_id| wait_id == id)),
        ChildWaitPolicy::FirstError => {
            // The error short-circuit only counts a completion from a child
            // this wait actually tracks (issue #546): a stray/duplicate
            // completion from an untracked child — e.g. a frozen runner's
            // task waking up after the watchdog already synthesized its
            // timeout, in a later run's wait — must not resume the parent.
            (is_error_like(latest_status) && wait_child_ids.iter().any(|id| id == latest_child_id))
                || wait_child_ids
                    .iter()
                    .all(|id| completed_child_ids.iter().any(|completed| completed == id))
        }
    }
}

/// Extract the child session's last assistant content, if any. Returns `None`
/// when the child produced no assistant message (e.g. errored before the first
/// model response, or only emitted tool messages).
#[cfg(test)]
fn child_final_assistant_text(child: &Session) -> Option<String> {
    child
        .messages
        .iter()
        .rev()
        .find(|message| matches!(message.role, Role::Assistant))
        .map(|message| message.content.clone())
        .filter(|content| !content.trim().is_empty())
}

fn runtime_resume_message(
    completion: &ChildCompletion,
    remaining_children: usize,
    child_final_response: Option<&str>,
) -> Message {
    let mut body = format!(
        "Runtime notification: child session `{}` finished with status `{}`. Remaining child sessions: {}.",
        completion.child_session_id, completion.status, remaining_children
    );

    // Fold the child's full final response back into the parent — no
    // truncation. Sub-agents are first-class agents whose complete conclusion
    // should be available to the parent without an extra SubAgent inspection round
    // trip. The message is left compressible (see `never_compress` below) so a
    // long transcript can still be reclaimed under parent compaction.
    let final_response = child_final_response.map(str::to_string);
    if let Some(response) = final_response.as_deref() {
        body.push_str("\n\nChild final response:\n");
        body.push_str(response);
    } else if let Some(error) = completion.error.as_deref() {
        if !error.is_empty() {
            body.push_str("\n\nChild error:\n");
            body.push_str(error);
        }
    }

    let inspection = serde_json::json!({
        "intent": "inspect", "target": completion.child_session_id, "message": "messages",
    });
    body.push_str(&format!(
        "\n\nResume the parent task using this child result and continue from the previous plan. \
         If you need transcript evidence, call SubAgent with {inspection}. \
         Follow returned cursors with a JSON-encoded view/cursor query in message; use view=\"message\" and message_id inside that query for a selected full message. Omit unused query fields.",
    ));

    let mut message = Message::user(body);
    message.metadata = Some(serde_json::json!({
        RUNTIME_RESUME_MESSAGE_HIDDEN_KEY: true,
        RUNTIME_RESUME_MESSAGE_KIND_KEY: "child_completion_resume",
        "child_session_id": completion.child_session_id,
        "child_status": completion.status,
        "child_error": completion.error,
        "child_final_response_included": final_response.is_some(),
        "child_completion_source": completion.source,
    }));
    // Allow parent-side compaction to reclaim this (now untruncated) message if
    // the parent context grows — important once children nest and fold full
    // results upward. The compact inspection hint preserves recoverability.
    message.never_compress = false;
    message
}

/// A current sealed required-packet completion always uses the bounded route.
/// Unusable reports are unavailable, never a reason to import legacy raw text.
/// Unbound children, Guardian verdicts and resident completions retain their paths.
fn sealed_child_report(
    completion: &ChildCompletion,
    child: &Session,
    parent: &Session,
) -> Option<serde_json::Value> {
    if child.subagent_type().as_deref() == Some("guardian")
        || child
            .metadata
            .get("lifecycle")
            .is_some_and(|v| v == "resident")
    {
        return None;
    }
    let source = completion.source.as_ref()?;
    if !source.matches_completion(completion)
        || ChildCompletionSource::from_committed_session(child).as_ref() != Some(source)
        || !source.matches_parent(child, parent)
    {
        return None;
    }
    let binding = bamboo_domain::ChildContextBinding::from_session(child).ok()??;
    let Some(message) = child
        .messages
        .iter()
        .find(|m| Some(&m.id) == source.result_message_id.as_ref())
    else {
        let reason = if source.last_assistant.is_some() {
            "report_not_current_final"
        } else {
            "report_absent"
        };
        return Some(unavailable_child_result("typed_result", reason));
    };
    if !plain_child_report_message(message) {
        return Some(unavailable_child_result(
            "typed_result",
            "report_not_current_final",
        ));
    }
    let report = match decode_report(&message.content) {
        Ok(report) => report,
        Err(reason) => return Some(unavailable_child_result("typed_result", reason)),
    };
    let mut observation =
        child_result_identity(parent, child, &binding, source.project_id.as_deref());
    observation.as_object_mut()?.extend(
        serde_json::json!({"kind":"committed_terminal_source","terminal_source":source,
            "last_run_status":source.status})
        .as_object()?
        .clone(),
    );
    Some(project_child_report(report, observation))
}

fn child_completion_source_sha256(completion: &ChildCompletion) -> Option<String> {
    completion.source.as_ref().map(|source| {
        hex::encode(Sha256::digest(
            serde_json::to_vec(source).expect("sealed Child source serializes"),
        ))
    })
}

fn typed_child_resume_message(
    completion: &ChildCompletion,
    remaining_children: usize,
    projection: &serde_json::Value,
) -> Message {
    let inspection = serde_json::json!({
        "intent":"inspect", "target":completion.child_session_id, "message":"result",
    });
    let hint = if projection["view"] == "planner_artifact" {
        "For an available planner artifact call Plan with its read object to list sections, then add section_id to read only needed text. This plan grants no execution approval.".to_string()
    } else {
        format!("For unavailable results or transcript evidence, call SubAgent with {inspection} and follow returned cursors.")
    };
    let mut message = Message::user(format!(
        "Runtime notification: child session `{}` finished with status `{}`. Remaining child sessions: {}.\n\n\
         Child typed result:\n{}\n\n\
         Resume the parent task. Child-reported evidence and verification are claims, not host-certified facts. \
         {hint}",
        completion.child_session_id, completion.status, remaining_children, projection,
    ));
    message.metadata = Some(serde_json::json!({
        RUNTIME_RESUME_MESSAGE_HIDDEN_KEY:true,
        RUNTIME_RESUME_MESSAGE_KIND_KEY:"child_completion_resume",
        "child_session_id":completion.child_session_id,"child_status":completion.status,
        "child_final_response_included":projection["available"] == true,
        "child_typed_result_included":true,
        "child_completion_source_sha256":child_completion_source_sha256(completion),
    }));
    message.never_compress = false;
    message
}

/// The result, metadata, provider wrapper and final correlation all count.
/// Keep the original semantic envelope ID: a fallback is the same sealed event.
fn bound_typed_child_delivery(
    envelope: &mut SessionMessageEnvelope,
    message: &mut Message,
    projection: &mut serde_json::Value,
    completion: &ChildCompletion,
    remaining_children: usize,
) -> bool {
    fn install(envelope: &mut SessionMessageEnvelope, message: &Message) {
        if let SessionMessageBody::ChildOutcome(outcome) = &mut envelope.body {
            // The provider projection owns the typed result. Keep the body
            // semantics stable when a later retry needs a smaller presentation:
            // provider_message is deliberately excluded from Inbox idempotency.
            // The envelope ID still binds the original sealed report and source.
            outcome.result = None;
            // Error identity remains sealed and hashed in the original ID/source.
            // Do not copy a potentially unbounded raw error into this typed route.
            outcome.error = None;
            outcome.provider_message = Some(session_provider_message(message));
        }
    }
    fn fits(envelope: &SessionMessageEnvelope, message: &Message) -> bool {
        fits_child_result_budget(envelope)
            && fits_child_result_budget(message)
            && envelope
                .to_provider_message()
                .is_ok_and(|actual| fits_child_result_budget(&actual))
    }
    install(envelope, message);
    if fits(envelope, message) {
        return true;
    }
    *projection = unavailable_child_result("typed_result", "result_budget_exceeded");
    *message = typed_child_resume_message(completion, remaining_children, projection);
    install(envelope, message);
    fits(envelope, message)
}

/// The hidden resume message for a completed **guardian** review: a directive,
/// verdict-tailored note that carries the reviewer's findings straight into the
/// parent (so it can act without a SubAgent inspection), mirroring
/// [`runtime_resume_message`]'s hidden/compressible shape.
fn guardian_resume_message(completion: &ChildCompletion, verdict: &GuardianVerdict) -> Message {
    let mut body = if verdict.approve {
        String::from(
            "Guardian review APPROVED: an independent reviewer verified the work and found no blocking issues. You may finalize the task.",
        )
    } else {
        String::from(
            "Guardian review REJECTED: an independent reviewer found issues. Address every finding below before completing — do NOT declare the task complete until they are resolved.",
        )
    };
    if let Some(summary) = verdict.summary.as_deref().filter(|s| !s.trim().is_empty()) {
        body.push_str("\n\nReviewer summary: ");
        body.push_str(summary);
    }
    if !verdict.findings.is_empty() {
        body.push_str("\n\nFindings:");
        for (idx, finding) in verdict.findings.iter().enumerate() {
            body.push_str(&format!("\n{}. {}", idx + 1, finding));
        }
    }
    let inspection = serde_json::json!({
        "intent": "inspect", "target": completion.child_session_id, "message": "messages",
    });
    body.push_str(&format!(
        "\n\nIf you need guardian transcript evidence, call SubAgent with {inspection}. \
         Follow returned cursors with a JSON-encoded view/cursor query in message, omitting unused query fields.",
    ));

    let mut message = Message::user(body);
    message.metadata = Some(serde_json::json!({
        RUNTIME_RESUME_MESSAGE_HIDDEN_KEY: true,
        RUNTIME_RESUME_MESSAGE_KIND_KEY: "guardian_review_resume",
        "child_session_id": completion.child_session_id,
        "child_status": completion.status,
        "guardian_approved": verdict.approve,
        "child_completion_source": completion.source,
    }));
    message.never_compress = false;
    message
}

/// Observe the existing registered Root surface without retaining its owner.
pub type RootToolSurfaceObserver = Arc<
    dyn Fn() -> futures::future::BoxFuture<'static, Option<Arc<dyn ToolExecutor>>> + Send + Sync,
>;

pub type ReservedRootExecutionAdapter = Arc<
    dyn Fn(
            Arc<Agent>,
            &Session,
            &SessionExecutionReservation,
            Arc<dyn ToolExecutor>,
        ) -> Result<(Arc<Agent>, Arc<dyn ToolExecutor>), bamboo_agent_core::AgentError>
        + Send
        + Sync,
>;

#[derive(Clone)]
pub struct ChildCompletionCoordinator {
    storage: Arc<dyn Storage>,
    persistence: Arc<bamboo_storage::LockedSessionStore>,
    sessions: crate::SessionCache,
    agent_runners: Arc<RwLock<HashMap<String, AgentRunner>>>,
    session_event_senders: Arc<RwLock<HashMap<String, broadcast::Sender<AgentEvent>>>>,
    agent: Arc<Agent>,
    config: Arc<RwLock<Config>>,
    provider_registry: Arc<ProviderRegistry>,
    provider_router: Arc<ProviderModelRouter>,
    app_data_dir: std::path::PathBuf,
    account_feed_inbox: Option<crate::execution::AccountFeedInbox>,
    root_account_sink: Option<Arc<crate::events::AccountEventSink>>,
    root_tools: Arc<RwLock<Option<Arc<dyn ToolExecutor>>>>,
    reserved_root_adapter: Arc<StdRwLock<Option<ReservedRootExecutionAdapter>>>,
    /// Late-bound guardian reviewer spawner, set post-construction by the server
    /// (mirrors `root_tools`). Re-injected into resumed runs so a guardian's
    /// reject→fix verdict can be re-reviewed across the suspend/resume boundary.
    guardian_spawner: Arc<RwLock<Option<Arc<dyn GuardianSpawner>>>>,
    /// Weak late binding avoids the scheduler -> completion handler ->
    /// scheduler ownership cycle while still routing idle child activation
    /// through the canonical placement-aware spawn core.
    spawn_scheduler: Arc<RwLock<Weak<SpawnScheduler>>>,
}

impl ChildCompletionCoordinator {
    pub fn with_root_account_sink(mut self, sink: Arc<crate::events::AccountEventSink>) -> Self {
        self.root_account_sink = Some(sink);
        self
    }

    #[allow(clippy::too_many_arguments)]
    pub fn new(
        storage: Arc<dyn Storage>,
        persistence: Arc<LockedSessionStore>,
        sessions: crate::SessionCache,
        agent_runners: Arc<RwLock<HashMap<String, AgentRunner>>>,
        session_event_senders: Arc<RwLock<HashMap<String, broadcast::Sender<AgentEvent>>>>,
        agent: Arc<Agent>,
        config: Arc<RwLock<Config>>,
        provider_registry: Arc<ProviderRegistry>,
        provider_router: Arc<ProviderModelRouter>,
        app_data_dir: std::path::PathBuf,
        account_feed_inbox: Option<crate::execution::AccountFeedInbox>,
    ) -> Self {
        Self {
            storage,
            persistence,
            sessions,
            agent_runners,
            session_event_senders,
            agent,
            config,
            provider_registry,
            provider_router,
            app_data_dir,
            account_feed_inbox,
            root_account_sink: None,
            root_tools: Arc::new(RwLock::new(None)),
            reserved_root_adapter: Arc::new(StdRwLock::new(None)),
            guardian_spawner: Arc::new(RwLock::new(None)),
            spawn_scheduler: Arc::new(RwLock::new(Weak::new())),
        }
    }

    /// Register a host producer at the existing genuine reserved Root seam.
    /// This is optional; independent workers and embeddings retain None.
    pub fn set_reserved_root_execution_adapter(&self, adapter: ReservedRootExecutionAdapter) {
        *self
            .reserved_root_adapter
            .write()
            .unwrap_or_else(std::sync::PoisonError::into_inner) = Some(adapter);
    }

    pub fn root_tool_surface_observer(&self) -> RootToolSurfaceObserver {
        let source = Arc::downgrade(&self.root_tools);
        Arc::new(move || {
            let source = source.clone();
            Box::pin(async move {
                let source = source.upgrade()?;
                let current = source.read().await.clone();
                current
            })
        })
    }

    pub async fn set_root_tools(&self, tools: Arc<dyn ToolExecutor>) {
        *self.root_tools.write().await = Some(tools);
    }

    pub async fn set_spawn_scheduler(&self, scheduler: &Arc<SpawnScheduler>) {
        *self.spawn_scheduler.write().await = Arc::downgrade(scheduler);
    }

    /// Wire the guardian reviewer spawner (server-provided), so resumed runs can
    /// re-spawn a guardian to re-review a fix after a reject verdict.
    pub async fn set_guardian_spawner(&self, spawner: Arc<dyn GuardianSpawner>) {
        *self.guardian_spawner.write().await = Some(spawner);
    }

    fn build_resume_config(
        &self,
        session: &Session,
        config_snapshot: &Config,
    ) -> ResumeConfigSnapshot {
        crate::session_app::resolution::resolve_resume_config_snapshot_for_execution(
            config_snapshot,
            &self.provider_registry,
            session,
            None,
        )
    }

    /// Drive a parent-resume and return the final [`ResumeOutcome`] so callers
    /// can distinguish a successful spawn (`Started`) from a gate-blocked
    /// attempt (`Completed`). The bash self-resume poll task uses this to
    /// detect the finalize-clobber case — its appended resume message was
    /// reverted by the suspending runner's final `merge_save_runtime`, so the
    /// resume port's `has_pending_user_message` gate fails and nothing spawns —
    /// and retry the clear→append→resume (see [`Self::bash_self_resume`]).
    async fn resume_parent(&self, parent_session_id: String) -> ResumeOutcome {
        for attempt in 0..=5u8 {
            if attempt > 0 {
                tokio::time::sleep(Duration::from_millis(250 * attempt as u64)).await;
            }

            let Some(session) = self.load_session(&parent_session_id).await else {
                tracing::warn!(%parent_session_id, "cannot resume parent after child completion: session not found");
                return ResumeOutcome::NotFound;
            };
            let config_snapshot = self.config.read().await.clone();
            let resume_config = self.build_resume_config(&session, &config_snapshot);
            let outcome =
                crate::session_app::resume::resume_session_execution_with_configured_fallback(
                    self,
                    &parent_session_id,
                    resume_config,
                )
                .await;
            tracing::info!(
                %parent_session_id,
                attempt,
                outcome = outcome.as_str(),
                "child completion requested parent resume"
            );

            if !matches!(outcome, ResumeOutcome::AlreadyRunning { .. }) {
                return outcome;
            }
        }
        // Exhausted the AlreadyRunning retry budget; surface the final state.
        // The wait state was already cleared and the resume message persisted,
        // so nothing event-driven will retry — the child-wait watchdog is the
        // backstop that picks this stranded parent up (it resumes suspended
        // sessions that hold a pending runtime resume message but no runner).
        tracing::error!(
            %parent_session_id,
            "parent resume gave up after AlreadyRunning retry budget; \
             relying on the child-wait watchdog backstop"
        );
        ResumeOutcome::AlreadyRunning {
            run_id: String::new(),
        }
    }

    async fn save_and_cache(&self, session: &mut Session) {
        if let Err(error) = self
            .persistence
            .merge_save_runtime_and_publish(session, |saved, _| {
                self.sessions.insert(
                    saved.id.clone(),
                    Arc::new(crate::SessionSnapshot::new(saved.clone())),
                );
            })
            .await
        {
            tracing::warn!(session_id = %session.id, %error, "failed to persist session");
        }
    }

    async fn reuse_legacy_child_delivery(
        &self,
        envelope: &mut SessionMessageEnvelope,
        completion: &ChildCompletion,
        wait: Option<&WaitingForChildrenState>,
        result: Option<&str>,
    ) -> Result<(), bamboo_domain::SessionInboxError> {
        if let (Some(wait), Some(inbox), Some(_)) =
            (wait, self.agent.session_inbox(), completion.source.as_ref())
        {
            let mut legacy = envelope.clone();
            legacy.id = child_completion_id(completion, wait.registered_at, result, true);
            if inbox.existing_staged_delivery(&legacy).await?.is_some() {
                *envelope = legacy;
            }
        }
        Ok(())
    }

    async fn legacy_child_delivery(
        &self,
        parent: &Session,
        completion: &ChildCompletion,
    ) -> Result<Option<(SessionMessageEnvelope, bool)>, bamboo_domain::SessionInboxError> {
        let (Some(source), Some(inbox)) = (completion.source.as_ref(), self.agent.session_inbox())
        else {
            return Ok(None);
        };
        let source_sha256 = child_completion_source_sha256(completion);
        let source = serde_json::to_value(source).expect("sealed Child source serializes");
        let matches_source = |message: &Message, envelope: &SessionMessageEnvelope| {
            if !matches!(&envelope.source, SessionMessageSource::Runtime { subsystem }
                if subsystem == "child_completion_coordinator")
                || envelope.target_session_id != parent.id
                || !matches!(&envelope.body, SessionMessageBody::ChildOutcome(outcome)
                    if outcome.child_session_id == completion.child_session_id && outcome.status == completion.status)
                || !bamboo_domain::is_matching_session_message(message, envelope)
            {
                return false;
            }
            let Some(metadata) = message.metadata.as_ref() else {
                return false;
            };
            // Old typed deliveries predate source metadata. Only a bounded,
            // available projection carries enough proof to match its generation.
            // Unavailable historical projections without a source are not guessed.
            let typed_source = message
                .content
                .split_once("\n\nChild typed result:\n")
                .and_then(|(_, text)| text.lines().next())
                .filter(|line| line.len() <= 8192)
                .and_then(|line| serde_json::from_str::<serde_json::Value>(line).ok())
                .and_then(|projection| {
                    projection
                        .get("host_observation")?
                        .get("terminal_source")
                        .cloned()
                });
            metadata.get("child_completion_source") == Some(&source)
                || source_sha256.as_ref().is_some_and(|digest| {
                    metadata
                        .get("child_completion_source_sha256")
                        .and_then(serde_json::Value::as_str)
                        == Some(digest.as_str())
                })
                || typed_source.as_ref() == Some(&source)
        };
        for message in &parent.messages {
            let Some(marker) = message
                .metadata
                .as_ref()
                .and_then(|m| m.get("session_message"))
            else {
                continue;
            };
            let Ok(envelope) = serde_json::from_value::<SessionMessageEnvelope>(marker.clone())
            else {
                continue;
            };
            if matches_source(message, &envelope)
                && inbox.was_admitted(&parent.id, &envelope.id).await?
            {
                return Ok(Some((envelope, true)));
            }
        }
        // An old writer may have cleared its wait after staging this exact
        // source, before a reasoning turn committed and ACKed its presentation.
        // Reuse that pending envelope rather than inventing a second event ID.
        for envelope in inbox
            .pending_child_outcomes(&parent.id, &completion.child_session_id)
            .await?
        {
            if envelope
                .to_provider_message()
                .is_ok_and(|message| matches_source(&message, &envelope))
            {
                return Ok(Some((envelope, false)));
            }
        }
        Ok(None)
    }
}

#[async_trait]
impl ChildCompletionHandler for ChildCompletionCoordinator {
    async fn on_child_completed(&self, completion: ChildCompletion) {
        // Terminality guard: a child that reports a NON-terminal status (e.g.
        // "suspended" — awaiting parent approval, its own bash wait, or its own
        // grandchildren) is not done. It must never satisfy the parent's wait:
        // `derive_completed_child_ids` folds the just-reported child in
        // unconditionally, so without this guard a suspending child would
        // resume the parent with a premature "finished with status
        // `suspended`" message. The child will publish a real terminal
        // completion when it later resumes and finishes.
        if !is_terminal_child_status(&completion.status) {
            tracing::info!(
                parent_session_id = %completion.parent_session_id,
                child_session_id = %completion.child_session_id,
                status = %completion.status,
                "non-terminal child status; leaving the parent wait armed"
            );
            return;
        }

        // Acquire the per-session async lock to eliminate the concurrent
        // double-resume race (see `parent_locks` for the full scenario). The
        // inner std::sync::Mutex is released immediately so no sync lock is
        // held across the await that follows.
        let per_parent = session_resume_lock(&completion.parent_session_id);
        let _per_parent_guard = per_parent.lock().await;

        let (parent, should_resume, messenger, child_admission, child_admissions) = loop {
            let Some(mut parent) = self.load_session(&completion.parent_session_id).await else {
                tracing::warn!(
                    parent_session_id = %completion.parent_session_id,
                    child_session_id = %completion.child_session_id,
                    "child completion received for missing parent"
                );
                return;
            };

            let expected_parent = parent.clone();

            // A parent may itself be a child (nested sub-agents): the rest of this
            // handler is kind-agnostic — it operates on `completion.parent_session_id`,
            // inspects that session's own `waiting_for_children` runtime state, and
            // resumes it. (Previously this bailed unless the parent was Root, which
            // silently dropped grandchild completions.)
            let mut runtime_state = read_runtime_state(&parent);

            // Single source of truth: reconstruct the completed-child set from the
            // session index rather than from a denormalized copy on the parent file.
            let mut completed_child_ids = derive_completed_child_ids(
                &self.storage,
                &completion.parent_session_id,
                &completion.child_session_id,
            )
            .await;

            let mut should_resume = false;
            let mut remaining_children = 0usize;
            let active_wait = runtime_state.waiting_for_children.clone();
            let tracked_wait = active_wait.as_ref().is_some_and(|wait| {
                wait.child_session_ids
                    .contains(&completion.child_session_id)
            });
            if !tracked_wait {
                // Registration owns the physical writer while it fresh-checks
                // Child terminality. Our no-wait decision must use that same
                // boundary, so a callback cannot disappear before its arm lands.
                let mut unchanged = parent.clone();
                match self
                    .persistence
                    .compare_exchange_child_wait_and_publish(
                        &parent,
                        &mut unchanged,
                        true,
                        Arc::new(|_| {}),
                    )
                    .await
                {
                    Ok(false) => continue,
                    Ok(true) if completion.source.is_none() => return,
                    Ok(true) => {}
                    Err(error) => {
                        tracing::warn!(parent_session_id = %parent.id, %error,
                            "child completion could not settle no-wait observation");
                        return;
                    }
                }
            }
            // Independent Stores can have stale child-status indexes. Refresh
            // this arm's known children from their durable control planes.
            if let Some(wait) = active_wait.as_ref() {
                for id in &wait.child_session_ids {
                    if id == &completion.child_session_id {
                        continue;
                    }
                    completed_child_ids.retain(|completed| completed != id);
                    if let Ok(Some(child)) = self
                        .storage
                        .load_child_wait_session(&parent, id, false)
                        .await
                    {
                        if completion_child_is_owned(&parent.id, child.parent_session_id.as_deref())
                            && child
                                .last_run_status()
                                .as_deref()
                                .is_some_and(is_terminal_child_status)
                        {
                            if repaired_child_requires_source(&child) {
                                let confirmed = self
                                    .storage
                                    .load_child_wait_session(&parent, id, true)
                                    .await
                                    .ok()
                                    .flatten();
                                if !confirmed.as_ref().is_some_and(|full| {
                                    ChildCompletionSource::from_committed_session(full)
                                        .is_some_and(|source| source.matches_parent(full, &parent))
                                }) {
                                    continue;
                                }
                            }
                            completed_child_ids.push(id.clone());
                        }
                    }
                }
            }
            if let Some(wait) = active_wait.as_ref() {
                remaining_children = wait
                    .child_session_ids
                    .iter()
                    .filter(|id| !completed_child_ids.iter().any(|completed| completed == *id))
                    .count();
                should_resume = tracked_wait
                    && wait_policy_satisfied(
                        wait.wait_for,
                        &wait.child_session_ids,
                        &completed_child_ids,
                        &completion.child_session_id,
                        &completion.status,
                    );
            }

            // READ-SIDE OWNERSHIP GUARD (issue #546): `SubAgent.wait` ids are
            // model-provided and unvalidated, and the watchdog unstrands a wait
            // over a FOREIGN/unknown id by publishing a synthetic completion
            // here. We must resume the parent (so it is not stranded) but MUST
            // NOT fold that foreign session's transcript into the parent — that
            // would be a cross-session disclosure primitive. Decide ownership
            // from the child's OWN parent linkage (control-plane only, no
            // messages loaded), and only load its full content when it is truly
            // this parent's child. An unowned id resumes with the neutral/error
            // message (`runtime_resume_message` falls back to `completion.error`
            // when no child content is supplied).
            let reported_control = match self
                .storage
                .load_child_wait_session(&parent, &completion.child_session_id, false)
                .await
            {
                Ok(child) => child,
                Err(error) => {
                    tracing::warn!(child_session_id = %completion.child_session_id, %error, "Child control plane unavailable; leaving wait armed");
                    return;
                }
            };
            let reported_child_owned = reported_control.as_ref().is_some_and(|child| {
                completion_child_is_owned(
                    &completion.parent_session_id,
                    child.parent_session_id.as_deref(),
                )
            });
            if completion.source.is_none()
                && reported_control
                    .as_ref()
                    .is_some_and(repaired_child_requires_source)
            {
                return;
            }

            // Load the completed child once, ONLY when owned. The guardian
            // branch inspects its subagent_type + final verdict; the generic
            // path folds its final assistant content into the hidden resume
            // message (avoiding an extra `SubAgent.get` round trip after resume).
            let loaded_child = if reported_child_owned {
                match self
                    .storage
                    .load_child_wait_session(&parent, &completion.child_session_id, true)
                    .await
                {
                    Ok(Some(child)) => Some(child),
                    Ok(None) => return,
                    Err(error) => {
                        tracing::warn!(child_session_id = %completion.child_session_id, %error,
                            "owned Child snapshot unavailable; leaving wait armed");
                        return;
                    }
                }
            } else {
                tracing::warn!(
                    parent_session_id = %completion.parent_session_id,
                    child_session_id = %completion.child_session_id,
                    "completion child is not a child of this parent; resuming with a neutral \
                     message and NOT folding its content"
                );
                None
            };

            if completion.source.is_none()
                && loaded_child
                    .as_ref()
                    .is_some_and(repaired_child_requires_source)
            {
                tracing::warn!(child_session_id = %completion.child_session_id, "repaired Child has no parent completion source; leaving wait armed");
                return;
            }

            // An old committed callback cannot count a successor, stage another
            // outcome, or clear a newer wait. Source-less synthetic completions may
            // unblock recovery, but never borrow the latest transcript answer.
            if let Some(source) = completion.source.as_ref() {
                if loaded_child
                    .as_ref()
                    .and_then(ChildCompletionSource::from_committed_session)
                    .as_ref()
                    != Some(source)
                    || !source.matches_completion(&completion)
                    || loaded_child
                        .as_ref()
                        .is_none_or(|child| !source.matches_parent(child, &parent))
                {
                    tracing::warn!(child_session_id = %completion.child_session_id,
                    "stale/unavailable committed child completion source; leaving wait armed");
                    return;
                }
            }
            if !tracked_wait
                && crate::ticket_runtime::is_independent_ticket_child(
                    self.storage.as_ref(),
                    &parent,
                    &completion.child_session_id,
                )
                .await
            {
                return;
            }
            let previous_delivery = match self.legacy_child_delivery(&parent, &completion).await {
                Ok(envelope) => envelope,
                Err(error) => {
                    tracing::warn!(parent_session_id = %parent.id, %error,
                        "cannot verify historical Child delivery; leaving replay recoverable");
                    return;
                }
            };
            let previously_admitted = previous_delivery
                .as_ref()
                .is_some_and(|(_, admitted)| *admitted);
            let child_final_response = loaded_child
                .as_ref()
                .and_then(|child| completion.source.as_ref()?.result(child, &parent));
            // Select the exact provider-facing resume message before durable
            // admission. The typed body carries its content/parts and safe runtime
            // metadata, so the canonical path is semantically identical to the
            // rolling-upgrade transcript fallback.
            let guardian_resume = if completion.source.is_some() || should_resume {
                let reviewed_round = runtime_state.round.current_round;
                loaded_child.as_ref().and_then(|child| {
                if child.subagent_type().as_deref() != Some("guardian") {
                    return None;
                }
                let mut guardian_state = read_guardian_state(&parent)?;
                if guardian_state.guardian_child_id.as_deref()
                    != Some(completion.child_session_id.as_str())
                {
                    // A *different* guardian is legitimately still in flight —
                    // leave its Pending state intact and use the generic resume.
                    tracing::warn!(
                        parent_session_id = %completion.parent_session_id,
                        child_session_id = %completion.child_session_id,
                        expected = ?guardian_state.guardian_child_id,
                        "guardian completion does not match recorded guardian_child_id; using generic resume"
                    );
                    return None;
                }
                if guardian_state.phase.is_reviewed() {
                    // Replaying the same sealed completion must not mutate a
                    // newer reasoning round or repeat the control transition.
                    return guardian_state.last_verdict.as_ref()
                        .map(|verdict| guardian_resume_message(&completion, verdict));
                }
                // This IS the guardian we dispatched, so we MUST advance the
                // phase out of `Pending` — otherwise the next terminal gate's
                // `Pending => return None` would let the run complete unreviewed.
                // A reviewer that errored or produced unparseable output is
                // treated as a SYNTHETIC REJECT (never a silent pass), so the
                // budgeted re-review loop governs the outcome: fail-closed, but
                // still bounded by `max_reviews`.
                let verdict = child_final_response.clone()
                    .and_then(|text| match parse_guardian_verdict(&text) {
                        Ok(verdict) => Some(verdict),
                        Err(error) => {
                            tracing::warn!(
                                child_session_id = %completion.child_session_id,
                                %error,
                                "guardian verdict unparseable; recording a synthetic reject"
                            );
                            None
                        }
                    })
                    .unwrap_or_else(|| {
                        GuardianVerdict::rejected(vec![
                            "The guardian reviewer did not return a usable verdict (it errored or \
                             emitted unparseable output); the work has NOT been independently \
                             verified."
                                .to_string(),
                        ])
                    });
                let approved = verdict.approve;
                let message = guardian_resume_message(&completion, &verdict);
                guardian_state.record_verdict(verdict, reviewed_round);
                write_guardian_state(&mut parent, guardian_state);
                tracing::info!(
                    parent_session_id = %completion.parent_session_id,
                    child_session_id = %completion.child_session_id,
                    approved,
                    "guardian verdict recorded; resuming parent"
                );
                Some(message)
            })
            } else {
                None
            };
            let guardian_control_changed = parent.metadata.get("guardian.state")
                != expected_parent.metadata.get("guardian.state");
            // A permanent receipt suppresses another notification, but it does
            // not prove that the corresponding Guardian verdict was committed.
            if previously_admitted && !should_resume && !guardian_control_changed {
                return;
            }
            let activation_policy = if should_resume || guardian_control_changed {
                SessionActivationPolicy::RespectSpecificWait
            } else {
                SessionActivationPolicy::InterruptSpecificWait
            };
            let mut typed_result = if let Some(child) = loaded_child.as_ref() {
                match super::child_session::planner_artifact::publish_completion(
                    &self.app_data_dir,
                    &completion,
                    child,
                    &parent,
                )
                .await
                {
                    Ok(Some(projection)) => Some(projection),
                    Ok(None) => sealed_child_report(&completion, child, &parent),
                    Err(error) => {
                        tracing::warn!(%error, child_session_id = %child.id,
                            "planner artifact publication failed; leaving wait armed");
                        return;
                    }
                }
            } else {
                None
            };
            let mut resume_message = if let Some(projection) = typed_result.as_mut() {
                let message =
                    typed_child_resume_message(&completion, remaining_children, projection);
                if fits_child_result_budget(&message) {
                    message
                } else {
                    *projection =
                        unavailable_child_result("typed_result", "result_budget_exceeded");
                    typed_child_resume_message(&completion, remaining_children, projection)
                }
            } else {
                guardian_resume.unwrap_or_else(|| {
                    runtime_resume_message(
                        &completion,
                        remaining_children,
                        child_final_response.as_deref(),
                    )
                })
            };
            if typed_result.is_some() && !fits_child_result_budget(&resume_message) {
                tracing::warn!(child_session_id = %completion.child_session_id,
                "typed child resume identity exceeds budget; leaving wait armed");
                return;
            }

            // Notification permission belongs to this exact outcome, not the
            // whole queue prefix or the parent's wait release. A parent may
            // reason on this result while its unfinished wait remains armed.
            let messenger = self.agent.session_messenger().cloned();
            let mut child_admission = if let Some(messenger) = messenger.as_ref() {
                let mut envelope = child_completion_envelope(
                    &completion,
                    active_wait
                        .as_ref()
                        .map_or(completion.completed_at, |wait| wait.registered_at),
                    child_final_response.clone(),
                    &resume_message,
                );
                if let Some(router) = self.agent.activation_router() {
                    let previous_fenced_run = previous_delivery
                        .as_ref()
                        .and_then(|(envelope, _)| envelope.correlation_id.as_deref())
                        .and_then(|value| value.strip_prefix("child_completion_after_run:"));
                    if let Some(run_id) = router
                        .child_completion_successor_run(
                            &completion.parent_session_id,
                            active_wait.is_some() || guardian_control_changed,
                            previous_fenced_run,
                        )
                        .await
                    {
                        envelope.correlation_id =
                            Some(format!("child_completion_after_run:{run_id}"));
                    }
                }
                if let Some(projection) = typed_result.as_mut() {
                    if !bound_typed_child_delivery(
                        &mut envelope,
                        &mut resume_message,
                        projection,
                        &completion,
                        remaining_children,
                    ) {
                        tracing::warn!(child_session_id = %completion.child_session_id,
                        "typed child outcome wrapper exceeds budget; leaving wait armed");
                        return;
                    }
                }
                if let Some((previous, _)) = previous_delivery.as_ref() {
                    envelope = previous.clone();
                } else if let Err(error) = self
                    .reuse_legacy_child_delivery(
                        &mut envelope,
                        &completion,
                        active_wait.as_ref(),
                        child_final_response.as_deref(),
                    )
                    .await
                {
                    tracing::warn!(parent_session_id = %parent.id, %error,
                        "cannot verify existing Child delivery; leaving wait armed");
                    return;
                }
                match messenger.admit(envelope).await {
                    Ok(admission) => {
                        if !should_resume && !guardian_control_changed {
                            let mut observed = expected_parent.clone();
                            match self
                                .persistence
                                .compare_exchange_child_wait_and_publish(
                                    &expected_parent,
                                    &mut observed,
                                    true,
                                    Arc::new(|_| {}),
                                )
                                .await
                            {
                                Ok(false) => continue,
                                Ok(true) => {}
                                Err(error) => {
                                    tracing::warn!(parent_session_id = %parent.id, %error,
                                        "partial Child delivery lost its parent authority; leaving staged outcome inert");
                                    return;
                                }
                            }
                        }
                        // A partial Guardian verdict must commit before any
                        // reasoning owner can consume its presentation. Its
                        // still-armed wait lets the watchdog recover this gap.
                        if should_resume || !guardian_control_changed {
                            if let Err(error) = messenger
                                .prepare_message_activation(&admission, activation_policy)
                                .await
                            {
                                tracing::warn!(parent_session_id = %completion.parent_session_id,
                                    child_session_id = %completion.child_session_id, %error,
                                    "child outcome notification permission failed; leaving wait armed");
                                return;
                            }
                        }
                        Some(admission)
                    }
                    Err(error) => {
                        tracing::warn!(
                            parent_session_id = %completion.parent_session_id,
                            child_session_id = %completion.child_session_id,
                            %error,
                            "child outcome SessionInbox admission failed; leaving parent wait armed"
                        );
                        return;
                    }
                }
            } else {
                None
            };

            let mut child_admissions = child_admission.iter().cloned().collect::<Vec<_>>();

            // The durable Child index can reach terminal state before its
            // completion callback acquires this parent lock. Clearing an already
            // satisfied wait after staging only the current callback would drop
            // every later callback at the no-wait guard above. Stage those exact
            // waited-for terminal siblings first, using the same wait identity and
            // Inbox protocol. Any/FirstError still do not wait for running siblings.
            if should_resume {
                if let (Some(wait), Some(messenger)) = (active_wait.as_ref(), messenger.as_ref()) {
                    let parent_run_id = if let Some(router) = self.agent.activation_router() {
                        router
                            .child_completion_successor_run(
                                &completion.parent_session_id,
                                true,
                                None,
                            )
                            .await
                    } else {
                        None
                    };
                    for child_id in &wait.child_session_ids {
                        if child_id == &completion.child_session_id
                            || !completed_child_ids.contains(child_id)
                        {
                            continue;
                        }
                        // Check lineage before reading transcript content. The
                        // index is parent-scoped, but ownership must still come
                        // from this child's own durable control plane.
                        let owned = match self
                            .storage
                            .load_child_wait_session(&parent, child_id, false)
                            .await
                        {
                            Ok(Some(child)) => completion_child_is_owned(
                                &completion.parent_session_id,
                                child.parent_session_id.as_deref(),
                            ),
                            _ => false,
                        };
                        let sibling = if owned {
                            self.storage
                                .load_child_wait_session(&parent, child_id, true)
                                .await
                        } else {
                            tracing::warn!(%child_id, "terminal sibling ownership unavailable; leaving parent wait armed");
                            return;
                        };
                        let sibling = match sibling {
                            Ok(Some(child))
                                if completion_child_is_owned(
                                    &completion.parent_session_id,
                                    child.parent_session_id.as_deref(),
                                ) && child
                                    .last_run_status()
                                    .as_deref()
                                    .is_some_and(is_terminal_child_status) =>
                            {
                                child
                            }
                            _ => {
                                tracing::warn!(%child_id, "terminal sibling snapshot unavailable; leaving parent wait armed");
                                return;
                            }
                        };
                        if (ChildCompletionSource::has_source_record(&sibling)
                            || repaired_child_requires_source(&sibling))
                            && ChildCompletionSource::from_committed_session(&sibling).is_none()
                        {
                            tracing::warn!(%child_id, "terminal sibling source is invalid; leaving wait armed");
                            return;
                        }
                        let sibling_completion = ChildCompletion {
                            parent_session_id: completion.parent_session_id.clone(),
                            child_session_id: child_id.clone(),
                            status: sibling.last_run_status().expect("terminal status checked"),
                            error: sibling.last_run_error(),
                            completed_at: sibling.updated_at,
                            source: ChildCompletionSource::from_committed_session(&sibling),
                        };
                        if sibling_completion
                            .source
                            .as_ref()
                            .is_some_and(|source| !source.matches_parent(&sibling, &parent))
                        {
                            tracing::warn!(%child_id, "terminal sibling parent context is stale; leaving wait armed");
                            return;
                        }
                        let result = sibling_completion
                            .source
                            .as_ref()
                            .and_then(|source| source.result(&sibling, &parent));
                        let mut typed_result =
                            match super::child_session::planner_artifact::publish_completion(
                                &self.app_data_dir,
                                &sibling_completion,
                                &sibling,
                                &parent,
                            )
                            .await
                            {
                                Ok(Some(projection)) => Some(projection),
                                Ok(None) => {
                                    sealed_child_report(&sibling_completion, &sibling, &parent)
                                }
                                Err(error) => {
                                    tracing::warn!(%error, %child_id,
                                    "sibling planner artifact publication failed; leaving wait armed");
                                    return;
                                }
                            };
                        let mut presentation = if let Some(projection) = typed_result.as_ref() {
                            typed_child_resume_message(
                                &sibling_completion,
                                remaining_children,
                                projection,
                            )
                        } else {
                            runtime_resume_message(
                                &sibling_completion,
                                remaining_children,
                                result.as_deref(),
                            )
                        };
                        let mut envelope = child_completion_envelope(
                            &sibling_completion,
                            wait.registered_at,
                            result.clone(),
                            &presentation,
                        );
                        if let Some(run_id) = parent_run_id.as_ref() {
                            envelope.correlation_id =
                                Some(format!("child_completion_after_run:{run_id}"));
                        }
                        if let Some(projection) = typed_result.as_mut() {
                            if !bound_typed_child_delivery(
                                &mut envelope,
                                &mut presentation,
                                projection,
                                &sibling_completion,
                                remaining_children,
                            ) {
                                tracing::warn!(%child_id, "typed terminal sibling wrapper exceeds budget; leaving wait armed");
                                return;
                            }
                        }
                        let previous_delivery = match self
                            .legacy_child_delivery(&parent, &sibling_completion)
                            .await
                        {
                            Ok(delivery) => delivery,
                            Err(error) => {
                                tracing::warn!(%child_id, %error,
                                    "cannot verify historical sibling delivery; leaving wait armed");
                                return;
                            }
                        };
                        if let Some((previous, _)) = previous_delivery {
                            envelope = previous;
                        } else if let Err(error) = self
                            .reuse_legacy_child_delivery(
                                &mut envelope,
                                &sibling_completion,
                                Some(wait),
                                result.as_deref(),
                            )
                            .await
                        {
                            tracing::warn!(%child_id, %error,
                                "cannot verify existing sibling delivery; leaving wait armed");
                            return;
                        }
                        match messenger.admit(envelope).await {
                            Ok(admission) => {
                                if let Err(error) = messenger
                                    .prepare_message_activation(
                                        &admission,
                                        SessionActivationPolicy::RespectSpecificWait,
                                    )
                                    .await
                                {
                                    tracing::warn!(%child_id, %error,
                                        "terminal sibling notification permission failed; leaving wait armed");
                                    return;
                                }
                                child_admissions.push(admission.clone());
                                // Keep every permission separately, while one
                                // request coalesces the batch's activation.
                                if child_admission.as_ref().is_none_or(|current| {
                                    admission.delivery.generation > current.delivery.generation
                                }) {
                                    child_admission = Some(admission);
                                }
                            }
                            Err(error) => {
                                tracing::warn!(%child_id, %error, "terminal sibling admission failed; leaving parent wait armed");
                                return;
                            }
                        }
                    }
                }
            }

            // Partial notifications do not own status/suspension writes. In
            // particular, do not stamp Suspended over a parent already reasoning
            // on an earlier result. The original wait is left byte-for-byte intact.
            if !should_resume && !guardian_control_changed && child_admission.is_some() {
                break (parent, true, messenger, child_admission, child_admissions);
            }

            if should_resume {
                runtime_state.waiting_for_children = None;
                runtime_state.status = AgentStatusState::Idle;
                runtime_state.suspension = None;
                parent.metadata.remove("runtime.suspend_reason");

                if child_admission.is_none() {
                    // Rolling-upgrade fallback only. The canonical path keeps the
                    // child outcome solely in SessionInbox until the next safe
                    // reasoning boundary.
                    parent.add_message(resume_message);
                }
            }

            parent.updated_at = Utc::now();
            write_runtime_state(&mut parent, &runtime_state);
            // Canonical ChildOutcome admission owns the provider-visible message;
            // this transaction only changes the parent's wait/control state. A
            // concurrent Root turn may already have advanced its model-context
            // ledger, so use the control-plane writer, which adopts that durable
            // ledger under the session lock. The rolling-upgrade fallback above
            // appends a message and still requires an append-safe checkpoint.
            let runtime_only = child_admission.is_some();
            let sessions = self.sessions.clone();
            let history = expected_parent.clone();
            let save = self
                .persistence
                .compare_exchange_child_wait_and_publish(
                    &expected_parent,
                    &mut parent,
                    runtime_only,
                    Arc::new(move |saved| {
                        if runtime_only {
                            if let Some(cached) = sessions.get(&saved.id) {
                                cached.update(|current| {
                                    if current.created_at != saved.created_at {
                                        *current = saved.clone();
                                        current.messages = history.messages.clone();
                                        current.provider_transcript =
                                            history.provider_transcript.clone();
                                        current
                                            .runtime_metadata
                                            .get_or_insert_with(Default::default)
                                            .session_inbox_admission =
                                            history.session_inbox_admission().cloned();
                                        return;
                                    }

                                    let messages = current.messages.clone();
                                    let provider_transcript = current.provider_transcript.clone();
                                    let admission =
                                        current.runtime_metadata.as_ref().and_then(|metadata| {
                                            metadata.session_inbox_admission.clone()
                                        });
                                    *current = saved.clone();
                                    current.messages = messages;
                                    current.provider_transcript = provider_transcript;
                                    current
                                        .runtime_metadata
                                        .get_or_insert_with(Default::default)
                                        .session_inbox_admission = admission;
                                });
                            } else {
                                let mut snapshot = saved.clone();
                                snapshot.messages = history.messages.clone();
                                snapshot.provider_transcript = history.provider_transcript.clone();
                                snapshot
                                    .runtime_metadata
                                    .get_or_insert_with(Default::default)
                                    .session_inbox_admission = history
                                    .runtime_metadata
                                    .as_ref()
                                    .and_then(|metadata| metadata.session_inbox_admission.clone());
                                sessions.insert(
                                    snapshot.id.clone(),
                                    Arc::new(crate::SessionSnapshot::new(snapshot)),
                                );
                            }
                        } else {
                            sessions.insert(
                                saved.id.clone(),
                                Arc::new(crate::SessionSnapshot::new(saved.clone())),
                            );
                        }
                    }),
                )
                .await;
            match save {
                Ok(false) => {
                    // Membership or policy changed, even if first-arm tag/time
                    // stayed the same. Reload and recompute; never replay a clear.
                    continue;
                }
                Ok(true) => {
                    break (
                        parent,
                        should_resume || (guardian_control_changed && !previously_admitted),
                        messenger,
                        child_admission,
                        child_admissions,
                    )
                }
                Err(error) => {
                    tracing::warn!(parent_session_id = %completion.parent_session_id,
                    child_session_id = %completion.child_session_id, %error,
                    "child outcome is durable but parent wait transition failed; leaving activation deferred");
                    return;
                }
            }
        };

        // Capture before releasing the per-parent lock so the borrow checker
        // is satisfied; `resume_parent` has its own retry loop and should not
        // hold the per-parent lock (it would block other completions for the
        // same parent, and the state is already durably settled above).
        let resume_parent_id = parent.id.clone();
        drop(_per_parent_guard);

        if should_resume {
            if let (Some(messenger), Some(admission)) = (messenger, child_admission) {
                // Control transitions have now committed. Upgrade each exact
                // outcome, including a lower generation below an ACKed sibling,
                // so a concurrently rearmed wait cannot suppress this batch.
                for outcome in &child_admissions {
                    if let Err(error) = messenger
                        .prepare_message_activation(
                            outcome,
                            SessionActivationPolicy::InterruptSpecificWait,
                        )
                        .await
                    {
                        tracing::warn!(parent_session_id = %resume_parent_id, %error,
                            "child control transition committed but notification permission failed");
                        return;
                    }
                }
                if let Err(error) = messenger.activate_prepared(&admission).await {
                    tracing::warn!(
                        parent_session_id = %resume_parent_id,
                        %error,
                        "child outcome and wait transition are durable but activation failed"
                    );
                }
            } else {
                self.resume_parent(resume_parent_id).await;
            }
        }
    }
}

#[async_trait]
impl ResumeExecutionPort for ChildCompletionCoordinator {
    async fn load_session(&self, session_id: &str) -> Option<Session> {
        match self.storage.load_session(session_id).await {
            Ok(Some(session)) => Some(session),
            Ok(None) => self
                .sessions
                .get(session_id)
                .map(|e| e.value().clone())
                .map(|arc| arc.read().clone()),
            Err(error) => {
                tracing::warn!(%session_id, %error, "failed to load session from storage");
                self.sessions
                    .get(session_id)
                    .map(|e| e.value().clone())
                    .map(|arc| arc.read().clone())
            }
        }
    }

    async fn save_and_cache_session(&self, session: &mut Session) {
        self.save_and_cache(session).await;
    }

    async fn reserve_session_execution(
        &self,
        session_id: &str,
        event_sender: &broadcast::Sender<AgentEvent>,
    ) -> SessionExecutionReserveOutcome {
        reserve_session_execution(
            &self.agent,
            &self.agent_runners,
            &self.session_event_senders,
            session_id,
            event_sender,
        )
        .await
    }

    async fn get_or_create_event_sender(&self, session_id: &str) -> broadcast::Sender<AgentEvent> {
        crate::execution::session_events::get_or_create_event_sender(
            &self.session_event_senders,
            session_id,
        )
        .await
    }

    fn dispatch_resume_execution(
        &self,
        request: ResumeSpawnRequest,
    ) -> Result<(), ResumeSpawnRequest> {
        let owner = self.clone();
        tokio::spawn(async move {
            ResumeExecutionPort::spawn_resume_execution(&owner, request).await;
        });
        Ok(())
    }

    fn dispatch_configured_resume_execution(
        &self,
        request: ResumeSpawnRequest,
    ) -> Result<(), ResumeSpawnRequest> {
        let owner = self.clone();
        tokio::spawn(async move {
            ResumeExecutionPort::spawn_configured_resume_execution(&owner, request).await;
        });
        Ok(())
    }

    async fn spawn_resume_execution(&self, request: ResumeSpawnRequest) {
        self.spawn_resume_with_origin(request, false).await;
    }

    async fn spawn_configured_resume_execution(&self, request: ResumeSpawnRequest) {
        self.spawn_resume_with_origin(request, true).await;
    }
}

impl ChildCompletionCoordinator {
    async fn spawn_resume_with_origin(&self, request: ResumeSpawnRequest, from_config: bool) {
        let ResumeSpawnRequest {
            session_id,
            mut session,
            mut execution_reservation,
            event_sender,
            config,
        } = request;
        if let Err(error) = execution_reservation.ensure_registered().await {
            tracing::warn!(
                %session_id,
                run_id = %execution_reservation.run_id(),
                %error,
                "cannot resume after child completion without exact router ownership"
            );
            return;
        }

        if self
            .agent
            .persistence()
            .root_actor_execution_required(&session)
            && self.root_account_sink.is_none()
        {
            tracing::warn!(%session_id, "Root resume has no actual account sink; rejecting before Actor claim");
            execution_reservation.abandon().await;
            return;
        }
        if let Err(error) = execution_reservation
            .bind_root_actor(&self.agent, &session)
            .await
        {
            tracing::warn!(%session_id, %error, "Root coordinator resume Actor binding rejected");
            execution_reservation.abandon().await;
            return;
        }
        let root_publication = execution_reservation.root_actor_writer().map(|owner| {
            crate::events::RootActorEventPublication::new(
                self.storage.clone(),
                self.root_account_sink
                    .clone()
                    .expect("required Root actual account sink checked before claim"),
                owner,
            )
        });

        let Some(root_tools) = self.root_tools.read().await.clone() else {
            tracing::error!(%session_id, "cannot resume parent after child completion: root tool surface is not initialized");
            return;
        };

        let adapter = self
            .reserved_root_adapter
            .read()
            .unwrap_or_else(std::sync::PoisonError::into_inner)
            .clone();
        let (agent, root_tools) = if let Some(adapter) = adapter {
            // A registered Root producer freezes its finite Source here.
            // Resolve the existing execution context first, while the exact
            // Actor reservation is held. Otherwise an initially unassigned
            // Root acquires its default Workspace in the first round after
            // Source binding, and correctly fails its own scope fence before
            // reaching the provider. This grants no Workspace change to a
            // bound Source; later refreshes still validate the frozen scope.
            if session.kind == SessionKind::Root {
                if let Err(error) = self
                    .agent
                    .prepare_external_session_for_execution(&mut session)
                    .await
                {
                    tracing::error!(%session_id, %error, "Root resume context preparation rejected before Source binding");
                    execution_reservation.abandon().await;
                    return;
                }
            }
            match adapter(
                self.agent.clone(),
                &session,
                &execution_reservation,
                root_tools,
            ) {
                Ok(bound) => bound,
                Err(error) => {
                    tracing::error!(%session_id, %error, "registered Root execution producer rejected resume");
                    execution_reservation.abandon().await;
                    return;
                }
            }
        } else {
            (self.agent.clone(), root_tools)
        };

        let config_snapshot = self.config.read().await.clone();
        let model = session.model.clone();
        let session_model_ref = session_effective_model_ref(&session);
        let requested_provider = session_model_ref
            .as_ref()
            .map(|model_ref| model_ref.provider.as_str())
            .unwrap_or(config.provider_name.as_str());
        let resolved_provider_name = match resolve_provider_routing_key(
            &config_snapshot,
            requested_provider,
            &self.provider_registry,
        ) {
            Ok(provider) => provider,
            Err(error) => {
                tracing::error!(
                    session_id = %session_id,
                    provider = requested_provider,
                    %error,
                    "child-completion resume provider is unavailable; refusing to fall back"
                );
                execution_reservation.abandon().await;
                return;
            }
        };
        let (selected_provider, managed_origin) = if let Some(mut model_ref) = session_model_ref {
            model_ref.provider = resolved_provider_name.clone();
            persist_model_ref(&mut session, &model_ref);
            match self.provider_router.route_with_origin(&model_ref) {
                Ok(selection) => selection,
                Err(error) => {
                    tracing::error!(
                        session_id = %session_id,
                        provider = %model_ref.provider,
                        model = %model_ref.model,
                        %error,
                        "child-completion resume provider routing failed closed"
                    );
                    execution_reservation.abandon().await;
                    return;
                }
            }
        } else {
            match self
                .provider_registry
                .get_with_origin(&resolved_provider_name)
            {
                Some(selection) => selection,
                None => {
                    tracing::error!(
                        session_id = %session_id,
                        provider = %resolved_provider_name,
                        "child-completion resume provider disappeared after resolution; refusing to fall back"
                    );
                    execution_reservation.abandon().await;
                    return;
                }
            }
        };
        let legacy_vision_provider_source =
            bamboo_llm::provider_registry::LegacyVisionProviderSource::capture(
                &config_snapshot,
                &resolved_provider_name,
                selected_provider.clone(),
                managed_origin.clone(),
            )
            .ok()
            .map(Arc::new);
        let agent = match make_child_resume_gold_resolver(
            self.config.clone(),
            resolved_provider_name.clone(),
            selected_provider.clone(),
            managed_origin,
        ) {
            Some(resolver) => Arc::new(
                agent.with_child_resume_gold_target_resolver(selected_provider.clone(), resolver),
            ),
            None => agent,
        };
        let agent = if from_config {
            Arc::new(agent.with_config_derived_image_fallback())
        } else {
            agent
        };
        let agent =
            Arc::new(agent.with_legacy_vision_provider_source(legacy_vision_provider_source));
        let resolved_fast_provider = resolve_fast_model(
            &config_snapshot,
            &resolved_provider_name,
            &self.provider_registry,
        )
        .map(|model| model.provider);
        let reasoning_effort = session.reasoning_effort;
        let reasoning_effort_source = session
            .metadata
            .get("reasoning_effort_source")
            .cloned()
            .unwrap_or_default();
        let gold_config = resolve_gold_config(
            &config_snapshot,
            session
                .metadata
                .get(GOLD_CONFIG_METADATA_KEY)
                .map(String::as_str),
        )
        .or(config.gold_config.clone());

        let (mpsc_tx, _forwarder, history_commit_barrier) =
            crate::execution::create_event_forwarder_with_root_actor(
                session_id.clone(),
                execution_reservation.run_id().to_string(),
                event_sender,
                self.agent_runners.clone(),
                self.account_feed_inbox.clone(),
                root_publication,
            );

        let auxiliary_model_resolver = make_resume_auxiliary_model_resolver(
            self.config.clone(),
            config_snapshot.clone(),
            self.provider_registry.clone(),
            resolved_provider_name.clone(),
        );
        let model_roster = crate::ModelRoster {
            model: Some(model),
            provider_name: Some(resolved_provider_name),
            provider_type: config.provider_type.clone(),
            fast: crate::RoleModel::from_parts(config.fast_model, resolved_fast_provider),
            background: crate::RoleModel::from_parts(
                config.background_model,
                config.background_model_provider,
            ),
            summarization: crate::RoleModel::from_parts(
                config.summarization_model,
                config.summarization_model_provider,
            ),
        };

        // Re-inject guardian state on resume so a reject→fix verdict can be
        // re-reviewed: config from the session (persisted at first spawn),
        // spawner from the coordinator-held handle. Absent guardian config this
        // stays `None`, and the approve→complete path is unchanged.
        let guardian_config = read_guardian_config(&session);
        let guardian_spawner = self.guardian_spawner.read().await.clone();

        consume_pending_clarification_resume(&mut session);
        spawn_session_execution(SessionExecutionArgs {
            agent,
            session_id,
            session,
            execution_reservation,
            tools_override: Some(root_tools),
            provider_override: Some(selected_provider),
            model_roster,
            reasoning_effort,
            reasoning_effort_source,
            auxiliary_model_resolver: Some(auxiliary_model_resolver),
            // Resumed child runs keep the spawn-time disabled snapshot (#136 lives
            // on the long-running main agent path; children are short-lived).
            disabled_filter_resolver: None,
            disabled_tools: Some(config.disabled_tools),
            disabled_skill_ids: Some(config.disabled_skill_ids),
            selected_skill_ids: None,
            selected_skill_mode: None,
            mpsc_tx,
            history_commit_barrier,
            image_fallback: config.image_fallback,
            gold_config,
            guardian_config,
            guardian_spawner,
            bash_resume_hook: {
                let hook: Arc<dyn BashResumeHook> = Arc::new(self.clone());
                Some(hook)
            },
            bash_completion_sink: {
                // Resumed runs keep the push wired too, so a background shell
                // launched after resume still notifies the loop.
                let sink: Arc<dyn BashCompletionSink> = Arc::new(self.clone());
                Some(sink)
            },
            app_data_dir: Some(self.app_data_dir.clone()),
            // Resume does not carry a fresh per-request override; the
            // config-level default (issue #221) still applies.
            run_budget: None,
            runners: self.agent_runners.clone(),
            sessions_cache: self.sessions.clone(),
            on_complete: None,
            // A resumed session that is itself a CHILD (nested sub-agents)
            // must publish its terminal completion so ITS parent is woken
            // in turn (issue #546).
            child_completion_handler: Some(Arc::new(self.clone())),
        });
    }
}

/// Real SessionInbox activation adapter. It reserves through the exact same
/// runner registry as every existing resume path, but deliberately bypasses
/// `has_pending_user_message`: the typed envelope is still in the durable inbox
/// and will be admitted by the loop's first safe turn boundary.
#[async_trait]
impl SessionActivationSpawner for ChildCompletionCoordinator {
    async fn reserve_activation(
        &self,
        target_session_id: &str,
        _inbox_generation: u64,
    ) -> Result<SessionActivationReserveOutcome, bamboo_domain::SessionActivationError> {
        let Some(inbox) = self.agent.session_inbox() else {
            return Err(bamboo_domain::SessionActivationError::Internal(
                "agent runtime has no SessionInbox".to_string(),
            ));
        };
        let backlog = inbox
            .inspect(target_session_id)
            .await
            .map_err(|error| bamboo_domain::SessionActivationError::Internal(error.to_string()))?;
        if !backlog.activation_pending() {
            return Ok(SessionActivationReserveOutcome::NoWork);
        }
        // Load and mutate under the persistence lock. A coordinator that armed
        // a child/Bash wait immediately before this activation therefore wins.
        let prepared = prepare_session_inbox_activation(
            &self.persistence,
            target_session_id,
            backlog.interrupt_pending(),
        )
        .await
        .map_err(|error| {
            bamboo_domain::SessionActivationError::Internal(format!(
                "persist resumable SessionInbox target: {error}"
            ))
        })?;
        let Some((session, ready)) = prepared else {
            return Ok(SessionActivationReserveOutcome::NotFound);
        };
        if !ready {
            tracing::info!(
                session_id = target_session_id,
                "SessionInbox backlog is activation-eligible but a specific durable wait remains armed"
            );
            return Ok(SessionActivationReserveOutcome::NoWork);
        }

        enum LaunchPlan {
            Root(Box<ResumeConfigSnapshot>),
            Child {
                scheduler: Arc<SpawnScheduler>,
                parent_session_id: String,
                model: String,
                disabled_tools: Option<Vec<String>>,
            },
        }

        // Derive every execution/security input from the latest locked session
        // returned above, never from a stale pre-lock snapshot.
        let launch_plan = match session.kind {
            SessionKind::Root => {
                if self.root_tools.read().await.is_none() {
                    return Err(bamboo_domain::SessionActivationError::Internal(
                        "root tool surface is not initialized".to_string(),
                    ));
                }
                let config_snapshot = self.config.read().await.clone();
                LaunchPlan::Root(Box::new(
                    self.build_resume_config(&session, &config_snapshot),
                ))
            }
            SessionKind::Child => {
                let scheduler = self.spawn_scheduler.read().await.upgrade().ok_or_else(|| {
                    bamboo_domain::SessionActivationError::Internal(
                        "child spawn scheduler is not initialized".to_string(),
                    )
                })?;
                let parent_session_id = session
                    .parent_session_id
                    .as_deref()
                    .map(str::trim)
                    .filter(|id| !id.is_empty())
                    .map(ToOwned::to_owned)
                    .ok_or_else(|| {
                        bamboo_domain::SessionActivationError::Internal(format!(
                            "child SessionInbox target {target_session_id} has no parent owner"
                        ))
                    })?;
                let parent = self
                    .storage
                    .load_session(&parent_session_id)
                    .await
                    .map_err(|error| {
                        bamboo_domain::SessionActivationError::Internal(format!(
                            "load parent owner {parent_session_id}: {error}"
                        ))
                    })?
                    .ok_or_else(|| {
                        bamboo_domain::SessionActivationError::Internal(format!(
                            "child SessionInbox parent owner {parent_session_id} disappeared"
                        ))
                    })?;
                let child_root = if session.root_session_id.trim().is_empty() {
                    parent_session_id.as_str()
                } else {
                    session.root_session_id.as_str()
                };
                let parent_root = if parent.root_session_id.trim().is_empty() {
                    parent.id.as_str()
                } else {
                    parent.root_session_id.as_str()
                };
                if child_root != parent_root {
                    return Err(bamboo_domain::SessionActivationError::Internal(format!(
                        "child SessionInbox target {target_session_id} does not share its parent owner's root"
                    )));
                }
                let model = if session.model.trim().is_empty() {
                    parent.model.clone()
                } else {
                    session.model.clone()
                };
                if model.trim().is_empty() {
                    return Err(bamboo_domain::SessionActivationError::Internal(format!(
                        "child SessionInbox target {target_session_id} has no executable model"
                    )));
                }
                let disabled_tools = match session.metadata.get("disabled_tools") {
                    None => None,
                    Some(raw) => {
                        let tools = serde_json::from_str::<std::collections::BTreeSet<String>>(raw)
                            .map_err(|error| {
                                bamboo_domain::SessionActivationError::Internal(format!(
                                    "child SessionInbox target {target_session_id} has malformed disabled_tools: {error}"
                                ))
                            })?;
                        (!tools.is_empty()).then(|| tools.into_iter().collect())
                    }
                };
                LaunchPlan::Child {
                    scheduler,
                    parent_session_id,
                    model,
                    disabled_tools,
                }
            }
        };
        let event_sender =
            ResumeExecutionPort::get_or_create_event_sender(self, target_session_id).await;

        let reservation = match reserve_runner_core(
            &self.agent_runners,
            &self.session_event_senders,
            target_session_id,
            &event_sender,
        )
        .await
        {
            ReserveOutcome::Reserved(reservation) => reservation,
            ReserveOutcome::AlreadyRunning(run_id) => {
                return Ok(SessionActivationReserveOutcome::AlreadyRunning { run_id });
            }
        };
        let run_id = reservation.run_id.clone();
        let launch = match launch_plan {
            LaunchPlan::Root(config) => {
                let mut execution_reservation =
                    SessionExecutionReservation::from_activation_placeholder(
                        target_session_id,
                        reservation,
                        self.agent
                            .activation_router()
                            .expect("SessionInbox activation requires an activation router")
                            .clone(),
                        self.agent_runners.clone(),
                    );
                // Busy must reach the router before it commits a launched
                // generation. Its existing recovery backoff can then retry the
                // durable input when the old Host's actual lease expires.
                if self
                    .agent
                    .persistence()
                    .root_actor_execution_required(&session)
                    && self.root_account_sink.is_none()
                {
                    execution_reservation
                        .rollback_unpublished_activation()
                        .await;
                    return Err(bamboo_domain::SessionActivationError::Internal(
                        "Root activation has no actual account sink".into(),
                    ));
                }
                if let Err(error) = execution_reservation
                    .bind_root_actor(&self.agent, &session)
                    .await
                {
                    execution_reservation
                        .rollback_unpublished_activation()
                        .await;
                    return Err(bamboo_domain::SessionActivationError::Internal(format!(
                        "Root Actor reservation rejected: {error}"
                    )));
                }
                // Launch and rollback share one exact RAII reservation. Dropping
                // an unlaunched SessionActivationLaunch cannot race a raw slot
                // removal against the reservation's router-placeholder cleanup.
                let reservation_cell = Arc::new(StdMutex::new(Some(execution_reservation)));
                let launch_reservation = reservation_cell.clone();
                let rollback_reservation = reservation_cell;
                let coordinator = self.clone();
                let launch_sessions = self.sessions.clone();
                let launch_session_id = session.id.clone();
                let launch_session = session.clone();
                let request_session_id = target_session_id.to_string();
                SessionActivationLaunch::new_with_async_rollback(
                    run_id,
                    move || {
                        let mut execution_reservation = launch_reservation
                            .lock()
                            .unwrap_or_else(std::sync::PoisonError::into_inner)
                            .take()
                            .expect("activation reservation launches or rolls back exactly once");
                        execution_reservation.mark_activation_published();
                        // The router publishes the exact owner before invoking
                        // this closure, so only now may the prepared snapshot
                        // replace the shared cache entry.
                        if execution_reservation.root_actor_writer().is_none() {
                            launch_sessions.insert(
                                launch_session_id,
                                Arc::new(crate::SessionSnapshot::new(launch_session)),
                            );
                        }
                        let request = ResumeSpawnRequest {
                            session_id: request_session_id,
                            session,
                            execution_reservation,
                            event_sender,
                            config: *config,
                        };
                        tokio::spawn(async move {
                            ResumeExecutionPort::spawn_configured_resume_execution(
                                &coordinator,
                                request,
                            )
                            .await;
                        });
                    },
                    move || async move {
                        let reservation = rollback_reservation
                            .lock()
                            .unwrap_or_else(std::sync::PoisonError::into_inner)
                            .take();
                        if let Some(reservation) = reservation {
                            reservation.rollback_unpublished_activation().await;
                        }
                    },
                )
            }
            LaunchPlan::Child {
                scheduler,
                parent_session_id,
                model,
                disabled_tools,
            } => {
                let execution_reservation =
                    SessionExecutionReservation::from_activation_placeholder(
                        target_session_id,
                        reservation,
                        self.agent
                            .activation_router()
                            .expect("SessionInbox activation requires an activation router")
                            .clone(),
                        self.agent_runners.clone(),
                    );
                let reservation_cell = Arc::new(StdMutex::new(Some(execution_reservation)));
                let launch_reservation = reservation_cell.clone();
                let rollback_reservation = reservation_cell;
                let job = SpawnJob {
                    parent_session_id,
                    child_session_id: target_session_id.to_string(),
                    model,
                    disabled_tools,
                };
                let launch_sessions = self.sessions.clone();
                let launch_session_id = session.id.clone();
                let launch_session = session;
                SessionActivationLaunch::new_with_async_rollback(
                    run_id,
                    move || {
                        let mut execution_reservation = launch_reservation
                            .lock()
                            .unwrap_or_else(std::sync::PoisonError::into_inner)
                            .take()
                            .expect("activation reservation launches or rolls back exactly once");
                        execution_reservation.mark_activation_published();
                        // As with root activation, publish the prepared cache
                        // snapshot only after router ownership commits.
                        launch_sessions.insert(
                            launch_session_id,
                            Arc::new(crate::SessionSnapshot::new(launch_session)),
                        );
                        // Dropping a JoinHandle detaches the task. The captured
                        // combined reservation remains RAII-protected if the
                        // task is later aborted or unwinds during setup.
                        drop(scheduler.launch_reserved(job, execution_reservation));
                    },
                    move || async move {
                        let reservation = rollback_reservation
                            .lock()
                            .unwrap_or_else(std::sync::PoisonError::into_inner)
                            .take();
                        if let Some(reservation) = reservation {
                            reservation.rollback_unpublished_activation().await;
                        }
                    },
                )
            }
        };
        Ok(SessionActivationReserveOutcome::Reserved(launch))
    }
}

/// Hidden resume message for a bash-completion self-resume (issue #84 Phase 2b).
/// Mirrors [`runtime_resume_message`]'s hidden/compressible shape so the resume
/// port's `has_pending_user_message` gate is satisfied.
///
/// `timed_out` selects the wording: the normal path (all shells finished)
/// announces completion; the deadline path (the 6h+10m wait ceiling was hit with
/// shells STILL running) must NOT claim the shells completed — it says they may
/// still be running so the model verifies with BashOutput instead of assuming
/// success on a false premise.
fn bash_completion_resume_message(bash_ids: &[String], timed_out: bool) -> Message {
    let body = if timed_out {
        format!(
            "Runtime notification: the background-Bash wait ceiling was reached while one or more \
             shell(s) ({}) may still be running. The session is being resumed so it is not \
             stranded; verify their actual status with BashOutput before assuming completion.",
            bash_ids.join(", ")
        )
    } else {
        format!(
            "Runtime notification: all background Bash shell(s) ({}) have completed. \
             Review their output with BashOutput and resume the task from where you left off.",
            bash_ids.join(", ")
        )
    };
    let mut message = Message::user(body);
    message.metadata = Some(serde_json::json!({
        RUNTIME_RESUME_MESSAGE_HIDDEN_KEY: true,
        RUNTIME_RESUME_MESSAGE_KIND_KEY: BASH_COMPLETION_RESUME_KIND,
    }));
    message.never_compress = false;
    message
}

/// Decide whether the bash self-resume should retry its clear→append→resume
/// sequence after a resume attempt returned `outcome`, given that the persisted
/// bash wait is (`true`) / is not (`false`) still set on reload.
///
/// Retry **only** when the resume did NOT spawn (`Completed` — no pending user
/// message, i.e. our resume message was dropped — or `AlreadyRunning`) AND the
/// persisted bash wait is still set: the signature of the finalize-clobber, where
/// the suspending runner's one-shot final `merge_save_runtime` lands after our
/// save and reverts `waiting_for_bash=Some` while dropping our resume message, so
/// `has_pending_user_message` fails and nothing spawns. `Started` (resume fired)
/// and `NotFound` (session gone) never retry. Pure helper so the clobber
/// detection is unit-testable in isolation from async I/O.
fn bash_resume_should_retry(outcome: &ResumeOutcome, persisted_waiting_for_bash: bool) -> bool {
    match outcome {
        ResumeOutcome::Started { .. } | ResumeOutcome::NotFound => false,
        ResumeOutcome::Completed | ResumeOutcome::AlreadyRunning { .. } => {
            persisted_waiting_for_bash
        }
    }
}

/// Whether a background-shell completion push should **resume** the owning loop
/// (vs merely enqueue an injection). Resume only when the loop is actually
/// suspended on a bash wait AND every shell it was waiting on has now finished —
/// resuming while other waited shells are still running would drop them back into
/// a foreground turn prematurely. The last shell to finish (or the backstop)
/// drives the resume; earlier ones enqueue their notice. Pure so the invariant is
/// unit-testable in isolation.
fn bash_completion_should_resume(
    loop_suspended_on_bash: bool,
    all_waited_shells_done: bool,
) -> bool {
    loop_suspended_on_bash && all_waited_shells_done
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum BashCompletionDeliveryPlan {
    /// Preserve the durable wait and do not reserve a successor. The last
    /// sibling completion (or the wait backstop) activates the accumulated
    /// inbox in order.
    DurableOnly,
    /// The loop is live/not waiting; notify its current owner after admission.
    Activate,
    /// The last waited shell finished: clear the wait durably, then activate.
    ClearWaitThenActivate,
}

fn bash_completion_delivery_plan(
    loop_suspended_on_bash: bool,
    all_waited_shells_done: bool,
) -> BashCompletionDeliveryPlan {
    if bash_completion_should_resume(loop_suspended_on_bash, all_waited_shells_done) {
        BashCompletionDeliveryPlan::ClearWaitThenActivate
    } else if loop_suspended_on_bash {
        BashCompletionDeliveryPlan::DurableOnly
    } else {
        BashCompletionDeliveryPlan::Activate
    }
}

fn background_bash_post_tool_payload(info: &BashCompletionInfo) -> HookPayload {
    let success = info.status == "completed" && info.exit_code == Some(0);
    let response = serde_json::json!({
        "bash_id": info.bash_id,
        "command": info.command,
        "exit_code": info.exit_code,
        "status": info.status,
        "output_tail": info.output_tail,
    });
    HookPayload::ToolResult {
        tool_name: "Bash".to_string(),
        tool_call_id: info.bash_id.clone(),
        outcome: HookToolOutcome {
            success,
            result: success.then(|| response.to_string()),
            error: (!success).then(|| response.to_string()),
            needs_human: false,
            duration_ms: 0,
        },
    }
}

fn append_background_bash_hook_feedback(info: &mut BashCompletionInfo, feedback: Vec<String>) {
    let feedback = feedback
        .into_iter()
        .map(|text| text.trim().to_string())
        .filter(|text| !text.is_empty())
        .collect::<Vec<_>>();
    if feedback.is_empty() {
        return;
    }
    if !info.output_tail.is_empty() {
        info.output_tail.push_str("\n\n");
    }
    info.output_tail.push_str("<post_tool_use_feedback>\n");
    info.output_tail.push_str(&feedback.join("\n"));
    info.output_tail.push_str("\n</post_tool_use_feedback>");
}

async fn run_background_bash_post_tool_hooks(
    config: &bamboo_config::LifecycleHooksConfig,
    fallback_cwd: Option<std::path::PathBuf>,
    session: &Session,
    info: &mut BashCompletionInfo,
) -> bool {
    let runner = crate::HookRunner::new().with_lifecycle_config(config, fallback_cwd);
    if !runner.has_hooks_for(AgentHookPoint::AfterToolExecution) {
        return false;
    }

    let mut runtime_state = session
        .agent_runtime_state
        .clone()
        .unwrap_or_else(|| AgentRuntimeState::new(&session.id));
    let outcome = runner
        .run_observer_hooks(
            AgentHookPoint::AfterToolExecution,
            &background_bash_post_tool_payload(info),
            session,
            &mut runtime_state,
            None,
        )
        .await;
    append_background_bash_hook_feedback(info, outcome.injected_contexts);
    true
}

/// Apply the bash-resume state transition to a loaded session **in place**: clear
/// the `waiting_for_bash` wait, mark the runtime Idle, drop the suspension +
/// `runtime.suspend_reason`, and append `resume_message`. Returns `false` (a
/// no-op) when the session was not actually waiting on bash — the double-resume
/// guard shared by the push and the backstop. Pure (no I/O) so both
/// [`ChildCompletionCoordinator::perform_bash_resume`] and unit tests exercise the
/// exact same transition.
fn apply_bash_resume_transition(session: &mut Session, resume_message: &Message) -> bool {
    let mut runtime_state = read_runtime_state(session);
    if runtime_state.waiting_for_bash.is_none() {
        return false;
    }
    runtime_state.waiting_for_bash = None;
    runtime_state.status = AgentStatusState::Idle;
    runtime_state.suspension = None;
    write_runtime_state(session, &runtime_state);
    session.metadata.remove("runtime.suspend_reason");
    session.metadata.remove("runtime.background_exec_id");
    session.add_message(resume_message.clone());
    true
}

/// Bash self-resume support (issue #84 Phase 2b; push follow-up).
impl ChildCompletionCoordinator {
    /// **Backstop** for a session suspended on `waiting_for_bash`. The primary,
    /// event-driven wake is the loop-facing push
    /// ([`BashCompletionSink::on_bash_completed`] → [`Self::deliver_bash_completion`]):
    /// the shell's completion task fires it the instant the process exits, and it
    /// resumes the loop directly. This task exists ONLY to catch a **lost push** —
    /// the completion landing in the window before the suspend was persisted (so
    /// the push saw no `waiting_for_bash` and only queued an injection), or a
    /// configuration with no sink wired — and to honour the wait ceiling.
    ///
    /// So it is deliberately NOT a hot spin: a coarse backoff (1 s → 30 s) that
    /// **yields to the push**. In the happy path the push has already cleared
    /// `waiting_for_bash` before the first check fires, so this returns after one
    /// cheap load with no registry polling at all. It only performs a resume when
    /// the shell(s) have finished but the loop is somehow still suspended, or the
    /// 6 h wait ceiling is reached.
    async fn bash_self_resume(&self, session_id: String, bash_ids: Vec<String>) {
        let mut delay = Duration::from_secs(1);
        let max_delay = Duration::from_secs(30);
        // Hard ceiling: the wait lease (6 h) + the registry GC TTL (5 min) +
        // margin. After this the shells are gone from the registry regardless,
        // so force-resume to avoid stranding the session on a GC edge case.
        let max_poll = Duration::from_secs(6 * 3600 + 600);
        let deadline = tokio::time::Instant::now() + max_poll;

        loop {
            tokio::time::sleep(delay).await;

            let Some(session) = self.load_session(&session_id).await else {
                tracing::info!(%session_id, "bash self-resume backstop: session gone; nothing to do");
                return;
            };
            let Some(wait) = read_runtime_state(&session).waiting_for_bash else {
                // The push (or another path) already resumed. This is the common
                // case — the backstop yields silently after a single load.
                return;
            };
            if wait.bash_ids != bash_ids {
                // A newer Human turn may have installed a different wait.
                return;
            }

            let still_running = bash_ids.iter().any(|id| {
                bamboo_tools::tools::bash_runtime::get_shell(id).is_some_and(|shell| {
                    shell.session_id.as_deref() == Some(session_id.as_str())
                        && shell.status() == "running"
                })
            });
            let timed_out = tokio::time::Instant::now() >= deadline;
            if session.metadata.contains_key("runtime.background_exec_id")
                && self.runner_is_running(&session_id).await
            {
                delay = (delay * 2).min(max_delay);
                continue;
            }
            if !still_running || timed_out {
                // The shell(s) finished but the loop is still suspended → the push
                // was lost (pre-persist window / no sink), or the ceiling hit.
                // Resume under the shared per-session lock so we never race the
                // push or a concurrent child-completion resume.
                let guard = session_resume_lock(&session_id);
                let _held = guard.lock().await;
                tracing::warn!(
                    %session_id,
                    shell_count = bash_ids.len(),
                    timed_out,
                    "bash self-resume backstop engaged (push lost or wait ceiling reached)"
                );
                self.perform_bash_resume(
                    &session_id,
                    if session.metadata.contains_key("runtime.background_exec_id") {
                        exec_wait_resume_message(&bash_ids, timed_out)
                    } else {
                        bash_completion_resume_message(&bash_ids, timed_out)
                    },
                    &bash_ids,
                )
                .await;
                return;
            }

            delay = (delay * 2).min(max_delay);
        }
    }

    /// Clear a session's `waiting_for_bash` state, append `resume_message`, and
    /// drive the parent resume — the shared clear→append→resume used by BOTH the
    /// event-driven push ([`Self::deliver_bash_completion`]) and the backstop poll
    /// ([`Self::bash_self_resume`]).
    ///
    /// **The caller MUST hold the [`session_resume_lock`] for `session_id`** so the
    /// load-check-clear-resume critical section is serialized against every other
    /// resume source (no double resume). No-op when the persisted wait was already
    /// cleared (another source handled it first).
    ///
    /// The clear→append→resume is a **bounded retry loop** that closes the
    /// finalize-clobber strand. The suspending runner's `finalize_task_context`
    /// runs a full `save_runtime_session` (same `merge_save_runtime`, which
    /// overwrites the whole `messages` array) that can land AFTER our save,
    /// reverting `waiting_for_bash=Some` and dropping our resume message, so
    /// `has_pending_user_message` fails and `resume_parent` returns `Completed`
    /// without spawning. We detect that (persisted wait still set after a
    /// non-`Started` outcome) and re-clear/re-append/re-resume. It converges
    /// because the runner's finalize persist is one-shot: once landed, our retry's
    /// save is the last writer, the message sticks, and resume fires.
    async fn perform_bash_resume(
        &self,
        session_id: &str,
        resume_message: Message,
        expected_ids: &[String],
    ) {
        let retry_backoff = Duration::from_millis(200);
        const MAX_RESUME_ATTEMPTS: u8 = 5;
        for attempt in 0..MAX_RESUME_ATTEMPTS {
            if attempt > 0 {
                tokio::time::sleep(retry_backoff).await;
            }

            let Some(mut session) = self.load_session(session_id).await else {
                tracing::warn!(%session_id, "bash resume: session not found; nothing to resume");
                return;
            };
            if read_runtime_state(&session)
                .waiting_for_bash
                .as_ref()
                .is_none_or(|wait| wait.bash_ids != expected_ids)
            {
                return;
            }

            // A Human may already be talking in the same session. Let that
            // owner receive the durable result at its normal inbox boundary.
            if session.metadata.contains_key("runtime.background_exec_id")
                && self.runner_is_running(session_id).await
            {
                continue;
            }

            if !apply_bash_resume_transition(&mut session, &resume_message) {
                // Double-resume guard: the wait was already cleared by another
                // source (the push, the backstop, or a user-driven resume). Do
                // not append a duplicate message or request a redundant resume.
                tracing::info!(
                    %session_id, attempt,
                    "bash resume: persisted bash wait already cleared; nothing to resume"
                );
                return;
            }
            session.updated_at = Utc::now();
            self.save_and_cache(&mut session).await;
            tracing::info!(
                %session_id, attempt,
                "bash resume: cleared bash wait and appended resume message"
            );

            let outcome = self.resume_parent(session_id.to_string()).await;
            match outcome {
                ResumeOutcome::Started { .. } => {
                    tracing::info!(%session_id, attempt, "bash resume: resume fired");
                    return;
                }
                ResumeOutcome::NotFound => {
                    tracing::warn!(%session_id, "bash resume: session vanished during resume");
                    return;
                }
                _ => {
                    // Completed (no pending user message ⇒ our resume message was
                    // dropped by the runner's finalize persist) or AlreadyRunning.
                    // Decide via the persisted bash wait: still set ⇒
                    // finalize-clobber ⇒ retry; cleared ⇒ the session is being
                    // handled (by us or a concurrent resume) ⇒ stop.
                    let clobbered = match self.load_session(session_id).await {
                        Some(reloaded) => read_runtime_state(&reloaded).waiting_for_bash.is_some(),
                        None => {
                            tracing::warn!(%session_id, "bash resume: session vanished after resume");
                            return;
                        }
                    };
                    if bash_resume_should_retry(&outcome, clobbered) {
                        tracing::warn!(
                            %session_id, attempt,
                            outcome = outcome.as_str(),
                            "bash resume: persisted wait still set after resume (finalize-clobber); retrying"
                        );
                        continue;
                    }
                    tracing::info!(
                        %session_id, attempt,
                        outcome = outcome.as_str(),
                        "bash resume: wait cleared and resume handled; stopping"
                    );
                    return;
                }
            }
        }

        tracing::warn!(
            %session_id,
            attempts = MAX_RESUME_ATTEMPTS,
            "bash resume: exhausted clobber-retry budget without confirming resume; giving up"
        );
    }
}

impl BashResumeHook for ChildCompletionCoordinator {
    fn arrange_bash_self_resume(&self, session_id: String, bash_ids: Vec<String>) {
        let coordinator = Arc::new(self.clone());
        tokio::spawn(async move {
            coordinator.bash_self_resume(session_id, bash_ids).await;
        });
    }

    fn supports_background_exec(&self) -> bool {
        self.agent
            .session_messenger()
            .is_some_and(|messenger| messenger.supports_runtime_target_lifetime())
    }

    fn arrange_exec_self_resume(
        &self,
        job: bamboo_tools::tools::exec_command::BackgroundExecJob,
        session_created_at: chrono::DateTime<Utc>,
    ) {
        let session_id = job.session_id().to_owned();
        let ids = vec![job.id().to_owned()];
        let coordinator = self.clone();
        tokio::spawn(async move {
            let info = job.completion().await;
            drop(job);
            coordinator
                .deliver_shell_completion_for_lifetime(
                    info,
                    ShellCompletionKind::Exec,
                    Some(session_created_at),
                )
                .await;
        });
        self.arrange_bash_self_resume(session_id, ids);
    }
}

#[derive(Clone, Copy, PartialEq, Eq)]
enum ShellCompletionKind {
    Bash,
    Exec,
}

async fn clear_completed_exec_wait<P>(
    persistence: &LockedSessionStore,
    session: &Session,
    handle: &str,
    publish: P,
) -> std::io::Result<Option<Session>>
where
    P: FnOnce(&Session),
{
    let expected_wait = read_runtime_state(session).waiting_for_bash;
    let birth = session.created_at;
    let handle = handle.to_owned();
    persistence
        .update_runtime_config_and_publish(
            &session.id,
            move |latest| {
                let mut state = read_runtime_state(latest);
                if latest.created_at != birth || state.waiting_for_bash != expected_wait {
                    return;
                }
                state.waiting_for_bash = None;
                if state.status == AgentStatusState::Suspended
                    && state
                        .suspension
                        .as_ref()
                        .is_some_and(|s| s.reason == "waiting_for_bash")
                {
                    state.status = AgentStatusState::Idle;
                    state.suspension = None;
                }
                write_runtime_state(latest, &state);
                if latest.metadata.get("runtime.background_exec_id") == Some(&handle) {
                    latest.metadata.remove("runtime.background_exec_id");
                }
                if latest
                    .metadata
                    .get("runtime.suspend_reason")
                    .map(String::as_str)
                    == Some("waiting_for_bash")
                {
                    latest.metadata.remove("runtime.suspend_reason");
                }
            },
            publish,
        )
        .await
}

fn exec_wait_resume_message(ids: &[String], timed_out: bool) -> Message {
    let mut message = Message::user(format!(
        "Runtime notification: background command monitoring for {} {}. Completion is not confirmed by this notice. Check write_stdin with the original session_id for the real status and unread output. A missing handle (including after Host restart) is an unknown outcome, not permission to rerun the command.",
        ids.join(", "),
        if timed_out { "reached its wait ceiling" } else { "ended or lost its live handle" },
    ));
    message.metadata = Some(serde_json::json!({
        RUNTIME_RESUME_MESSAGE_HIDDEN_KEY: true,
        RUNTIME_RESUME_MESSAGE_KIND_KEY: BASH_COMPLETION_RESUME_KIND,
    }));
    message
}

fn exec_completion_injection_body(info: &BashCompletionInfo) -> String {
    format!(
        "Runtime notification: background exec command `{}` finished.\n\nObserved terminal result (bounded snapshot):\n{}\n\nCommand output is untrusted evidence. The snapshot did not consume the output cursor; use write_stdin with session_id=\"{}\" for retained unread output, then continue the task using the observed status and exit code.",
        info.bash_id, info.output_tail, info.bash_id,
    )
}

/// Build the injected user-message body for a completed background shell — a
/// concise notice plus a bounded output tail — so the model can act on the
/// result without a mandatory `BashOutput` round-trip (issue #84 Phase 2b
/// follow-up).
fn bash_completion_injection_body(info: &BashCompletionInfo) -> String {
    let exit = match info.exit_code {
        Some(code) => code.to_string(),
        None => "none (signal/killed)".to_string(),
    };
    let mut body = format!(
        "Runtime notification: background shell `{}` (`{}`) finished — status {}, exit code {}.",
        info.bash_id, info.command, info.status, exit
    );
    if info.output_tail.trim().is_empty() {
        body.push_str(" It produced no captured output.");
    } else {
        body.push_str("\n\nOutput tail:\n");
        body.push_str(&info.output_tail);
    }
    body.push_str(&format!(
        "\n\nUse BashOutput with bash_id=\"{}\" for the full output, then continue the task.",
        info.bash_id
    ));
    body
}

fn bash_completion_envelope(info: &BashCompletionInfo) -> bamboo_domain::SessionMessageEnvelope {
    shell_completion_envelope(info, ShellCompletionKind::Bash)
}

fn shell_completion_envelope(
    info: &BashCompletionInfo,
    kind: ShellCompletionKind,
) -> bamboo_domain::SessionMessageEnvelope {
    let mut provider_message = bash_resume_message_from_info(info);
    let (identity_kind, instruction, content, data) = if kind == ShellCompletionKind::Exec {
        provider_message.content = exec_completion_injection_body(info);
        (
            "background_exec",
            "background_exec_completed",
            provider_message.content.clone(),
            serde_json::json!({
                "session_id": info.session_id,
                "exec_session_id": info.bash_id,
                "status": info.status,
                "exit_code": info.exit_code,
                "result": info.output_tail,
            }),
        )
    } else {
        (
            "background_bash",
            "background_bash_completed",
            bash_completion_injection_body(info),
            serde_json::json!({
            "session_id": info.session_id,
            "bash_id": info.bash_id,
            "command": info.command,
            "exit_code": info.exit_code,
            "status": info.status,
            "output_tail": info.output_tail,
            }),
        )
    };
    // The id covers the immutable completion snapshot. An exact delivery retry
    // (whose transport timestamp may differ) is idempotent; a changed status,
    // output tail, command, or exit code is a distinct correction rather than
    // silently reusing one id with different semantics.
    let identity = data.clone();
    bamboo_domain::SessionMessageEnvelope {
        id: bamboo_domain::SessionMessageId::stable(
            &format!("{identity_kind}_completion"),
            &identity,
        ),
        source: bamboo_domain::SessionMessageSource::Runtime {
            subsystem: identity_kind.to_string(),
        },
        target_session_id: info.session_id.clone(),
        kind: bamboo_domain::SessionMessageKind::RuntimeInstruction,
        body: bamboo_domain::SessionMessageBody::RuntimeInstruction(
            bamboo_domain::SessionRuntimeInstruction {
                instruction: instruction.to_string(),
                content: Some(bamboo_domain::SessionMessageContent::text(content)),
                data: Some(data),
                provider_message: Some(session_provider_message(&provider_message)),
            },
        ),
        created_at: Utc::now(),
        thread_id: None,
        in_reply_to: None,
        attempt: None,
        correlation_id: Some(info.bash_id.clone()),
    }
}

/// Enqueue a completed shell's summary as a pending injected message on the
/// owning session for the rolling-upgrade polling backstop. New push delivery
/// uses [`SessionMessenger`]; this helper remains only so an older process that
/// has not wired the typed delivery plane can still resume persisted sessions.
/// Race-safe: `update_runtime_config` loads and saves under the per-session lock.
async fn enqueue_bash_completion_injection(
    persistence: &LockedSessionStore,
    info: &BashCompletionInfo,
) -> std::io::Result<Option<Session>> {
    let body = bash_completion_injection_body(info);
    let queued = serde_json::json!({
        "content": body,
        "created_at": Utc::now(),
    });
    persistence
        .update_runtime_config(&info.session_id, move |session| {
            let mut pending = session.pending_injected_messages().unwrap_or_default();
            pending.push(queued);
            session.set_pending_injected_messages(pending);
        })
        .await
}

/// Build the hidden, compressible resume message for a completed background
/// shell — the same rich notice body used for a live-loop injection
/// ([`bash_completion_injection_body`]), but tagged as a resume message so it
/// satisfies the `has_pending_user_message` gate that lets a suspended session
/// spawn. This is what the **push** appends when it wakes a suspended loop, so
/// the model gets the shell's status + output tail in one shot without a
/// separate `BashOutput` round-trip.
fn bash_resume_message_from_info(info: &BashCompletionInfo) -> Message {
    let mut message = Message::user(bash_completion_injection_body(info));
    message.metadata = Some(serde_json::json!({
        RUNTIME_RESUME_MESSAGE_HIDDEN_KEY: true,
        RUNTIME_RESUME_MESSAGE_KIND_KEY: BASH_COMPLETION_RESUME_KIND,
    }));
    message.never_compress = false;
    message
}

impl ChildCompletionCoordinator {
    /// Loop-facing delivery of a completed background shell. Two paths, chosen
    /// under the per-session resume lock so we never race the backstop poll or a
    /// concurrent child-completion resume:
    ///
    /// - **Suspended loop** (the model ended its turn with the shell running, so
    ///   `waiting_for_bash` is set) AND every waited shell has now finished →
    ///   **resume the loop directly**, event-driven, appending the rich completion
    ///   notice as the resume message. This is the push's whole point: no polling.
    /// - Every notice is then delivered through the same typed SessionMessenger
    ///   as peer/user steering. The durable inbox and activation router decide
    ///   whether to notify the current owner or reserve one successor.
    async fn deliver_bash_completion(&self, info: BashCompletionInfo) {
        self.deliver_shell_completion(info, ShellCompletionKind::Bash)
            .await;
    }

    async fn deliver_shell_completion(&self, info: BashCompletionInfo, kind: ShellCompletionKind) {
        self.deliver_shell_completion_for_lifetime(info, kind, None)
            .await;
    }

    async fn deliver_shell_completion_for_lifetime(
        &self,
        mut info: BashCompletionInfo,
        kind: ShellCompletionKind,
        expected_birth: Option<chrono::DateTime<Utc>>,
    ) {
        // A background shell completes outside the originating tool call, so
        // fire its PostToolUse seam here before routing the completion into the
        // next round/resume message. Do not hold the resume lock while an
        // arbitrary command hook runs: its configured timeout must never block
        // the independent liveness backstop.
        if let Some(hook_session) = self
            .load_session(&info.session_id)
            .await
            .filter(|_| kind == ShellCompletionKind::Bash)
        {
            let config_snapshot = self.config.read().await.clone();
            let _ = run_background_bash_post_tool_hooks(
                &config_snapshot.lifecycle_hooks,
                Some(self.app_data_dir.clone()),
                &hook_session,
                &mut info,
            )
            .await;
        }

        let guard = session_resume_lock(&info.session_id);
        let _held = guard.lock().await;

        let Some(session) = self.load_session(&info.session_id).await else {
            tracing::warn!(
                session_id = %info.session_id,
                bash_id = %info.bash_id,
                "background bash completion: owning session not found; nothing to notify"
            );
            return;
        };

        if expected_birth.is_some_and(|birth| session.created_at != birth) {
            // A reused display id cannot redirect an old command's result into
            // a new Session lifetime.
            return;
        }
        let runtime = read_runtime_state(&session);
        let waiting = runtime.waiting_for_bash.as_ref().is_some_and(|wait| {
            kind == ShellCompletionKind::Bash || wait.bash_ids.contains(&info.bash_id)
        });
        // The producer flips the shell's `running` flag false BEFORE firing this
        // push, so a now-empty per-session registry means every shell the loop was
        // waiting on has finished — safe to resume. If OTHER waited shells are
        // still running, fall through to the enqueue path and let the last one (or
        // the backstop) drive the resume.
        let all_shells_done = runtime.waiting_for_bash.as_ref().is_none_or(|wait| {
            wait.bash_ids.iter().all(|id| {
                id == &info.bash_id
                    || bamboo_tools::tools::bash_runtime::get_shell(id).is_none_or(|shell| {
                        shell.session_id.as_deref() == Some(info.session_id.as_str())
                            && shell.status() != "running"
                    })
            })
        });
        let delivery_plan = bash_completion_delivery_plan(waiting, all_shells_done);

        // Stage 1: make the completion durable BEFORE clearing the wait. The
        // activation is intentionally deferred until the wait-state mutation
        // below is durably committed; otherwise a successor can observe the old
        // suspension, while clear-before-deliver can lose the only wake on
        // crash.
        let messenger = self.agent.session_messenger().cloned();
        let admission = match messenger.as_ref() {
            Some(messenger) => match if kind == ShellCompletionKind::Exec {
                messenger
                    .admit_runtime_for_target(
                        shell_completion_envelope(&info, kind),
                        expected_birth.unwrap_or(session.created_at),
                    )
                    .await
            } else {
                messenger.admit(bash_completion_envelope(&info)).await
            } {
                Ok(admission) => Some(admission),
                Err(error) => {
                    tracing::warn!(
                        session_id = %info.session_id,
                        bash_id = %info.bash_id,
                        %error,
                        "background bash completion: durable SessionInbox admission failed; leaving wait armed"
                    );
                    return;
                }
            },
            None => {
                tracing::warn!(
                    session_id = %info.session_id,
                    bash_id = %info.bash_id,
                    "SessionMessenger unavailable; using compatibility injection before clearing wait"
                );
                match enqueue_bash_completion_injection(&self.persistence, &info).await {
                    Ok(Some(_)) => None,
                    Ok(None) => {
                        tracing::warn!(
                            session_id = %info.session_id,
                            "background bash compatibility target disappeared"
                        );
                        return;
                    }
                    Err(error) => {
                        tracing::warn!(
                            session_id = %info.session_id,
                            %error,
                            "background bash compatibility admission failed; leaving wait armed"
                        );
                        return;
                    }
                }
            }
        };

        if delivery_plan == BashCompletionDeliveryPlan::DurableOnly {
            tracing::info!(
                session_id = %info.session_id,
                bash_id = %info.bash_id,
                "background bash completion is durable; sibling shells still run, so wait remains armed"
            );
            return;
        }

        if let (Some(messenger), Some(admission)) = (messenger.as_ref(), admission.as_ref()) {
            let prepared = if kind == ShellCompletionKind::Exec {
                messenger
                    .prepare_message_activation(
                        admission,
                        SessionActivationPolicy::RespectSpecificWait,
                    )
                    .await
            } else {
                messenger.prepare_activation(admission).await
            };
            if let Err(error) = prepared {
                tracing::warn!(
                    session_id = %info.session_id,
                    bash_id = %info.bash_id,
                    %error,
                    "background bash completion activation watermark failed; leaving wait armed"
                );
                return;
            }
        }

        if delivery_plan == BashCompletionDeliveryPlan::ClearWaitThenActivate {
            tracing::info!(
                session_id = %info.session_id,
                bash_id = %info.bash_id,
                status = %info.status,
                "background bash completion: push-resuming suspended loop (event-driven)"
            );
            if kind == ShellCompletionKind::Exec {
                // The Human may have started a new foreground turn since the
                // handoff. Clear only this exact wait under the persistence
                // lock; never save the earlier message/status snapshot over
                // that active conversation.
                match clear_completed_exec_wait(
                    &self.persistence,
                    &session,
                    &info.bash_id,
                    |saved| {
                        self.sessions.insert(
                            saved.id.clone(),
                            Arc::new(crate::SessionSnapshot::new(saved.clone())),
                        );
                    },
                )
                .await
                {
                    Ok(Some(_)) => {}
                    Ok(None) => return,
                    Err(error) => {
                        tracing::warn!(session_id = %info.session_id, %error, "background exec completion is durable but wait clear failed");
                        return;
                    }
                }
            } else {
                let mut resumable = session.clone();
                let mut runtime_state = read_runtime_state(&resumable);
                runtime_state.waiting_for_bash = None;
                runtime_state.status = AgentStatusState::Idle;
                runtime_state.suspension = None;
                write_runtime_state(&mut resumable, &runtime_state);
                resumable.metadata.remove("runtime.suspend_reason");
                resumable.metadata.remove("runtime.background_exec_id");
                resumable.updated_at = Utc::now();
                if let Err(error) = self.persistence.merge_save_runtime(&mut resumable).await {
                    tracing::warn!(
                        session_id = %info.session_id,
                        %error,
                        "background bash completion is durable but wait-state clear failed; leaving activation to the wait backstop"
                    );
                    return;
                }
                self.sessions.insert(
                    resumable.id.clone(),
                    Arc::new(crate::SessionSnapshot::new(resumable)),
                );
            }
        }

        // Stage 3: activation is allowed only after the durable wait-state
        // clear. A crash before here leaves either the wait backstop or startup
        // inbox reconciliation able to recover.
        let (Some(messenger), Some(admission)) = (messenger, admission) else {
            if delivery_plan == BashCompletionDeliveryPlan::ClearWaitThenActivate {
                let _ = self.resume_parent(info.session_id.clone()).await;
            }
            return;
        };
        match messenger.activate_prepared(&admission).await {
            Ok(receipt) => tracing::info!(
                session_id = %info.session_id,
                bash_id = %info.bash_id,
                status = %info.status,
                waiting,
                generation = receipt.delivery.generation,
                activation = ?receipt.activation,
                "background bash completion delivered through SessionMessenger"
            ),
            Err(error) => tracing::warn!(
                session_id = %info.session_id,
                bash_id = %info.bash_id,
                %error,
                "background bash completion: SessionMessenger delivery failed"
            ),
        }
    }
}

impl BashCompletionSink for ChildCompletionCoordinator {
    fn on_bash_completed(&self, info: BashCompletionInfo) {
        // Best-effort, off the shell's completion-poll task: hand the delivery to
        // a detached task so the producer is never blocked (mirrors
        // `arrange_bash_self_resume`).
        let coordinator = Arc::new(self.clone());
        tokio::spawn(async move {
            coordinator.deliver_bash_completion(info).await;
        });
    }
}

// ---------------------------------------------------------------------------
// Child-wait watchdog (issue #546)
// ---------------------------------------------------------------------------

/// How often the child-wait watchdog sweeps suspended sessions.
const CHILD_WAIT_SWEEP_INTERVAL_SECS: u64 = 30;
/// Leave a freshly registered wait alone for this long so the event-driven
/// completion push always gets the first shot (and just-enqueued spawn jobs
/// have time to persist their running marker).
const CHILD_WAIT_REGISTRATION_GRACE_SECS: i64 = 60;
/// A waited child with NO live runner and a non-terminal index status is
/// declared dead once its control-plane has been quiet for this long.
const DEAD_CHILD_GRACE_SECS: i64 = 120;
/// Slack on top of the per-child liveness policy before a Running-but-frozen
/// runner entry (dead task) is force-finalized by the sweeper. The per-child
/// watchdog cancels at `max_idle`/`max_total` and the child then publishes its
/// own timeout; an entry frozen this far PAST those limits proves that
/// machinery is dead.
const STALE_RUNNER_SLACK_SECS: i64 = 600;

/// Whether a waited child's index status means the sweeper must consider it
/// DEAD when nothing is driving it: non-terminal and not legitimately
/// suspended. `None` = never ran (created-but-never-started, or a spawn that
/// was lost before its running marker persisted).
fn is_dead_child_candidate_status(status: Option<&str>) -> bool {
    match status {
        // "suspended" children wait on a human / their own children / bash —
        // their wake has its own driver; never declare them dead here.
        Some(status) => !is_terminal_child_status(status) && status != "suspended",
        None => true,
    }
}

/// Whether a reported completion's child id is genuinely a child of the parent
/// it claims (issue #546 read-side disclosure guard). `SubAgent.wait` ids are
/// model-provided, and the watchdog resolves an unowned id by publishing a
/// synthetic completion so the parent is unstranded — but the parent must NEVER
/// receive a FOREIGN session's content folded into its transcript.
/// `child_parent_linkage` is that session's own `parent_session_id` (`None`
/// when the session does not exist). Pure so the rule is unit-testable.
fn completion_child_is_owned(reported_parent: &str, child_parent_linkage: Option<&str>) -> bool {
    child_parent_linkage == Some(reported_parent)
}

fn child_wait_watchdog_resume_message(body: String) -> Message {
    let mut message = Message::user(body);
    message.metadata = Some(serde_json::json!({
        RUNTIME_RESUME_MESSAGE_HIDDEN_KEY: true,
        RUNTIME_RESUME_MESSAGE_KIND_KEY: "child_wait_watchdog_resume",
    }));
    message.never_compress = false;
    message
}

fn empty_child_wait_message() -> Message {
    let inspection = serde_json::json!({"intent": "inspect", "message": "tree"});
    child_wait_watchdog_resume_message(format!(
        "Runtime notification: this session was suspended waiting for child sessions, but the \
         wait tracked no children (internal inconsistency). The session has been resumed; use \
         SubAgent with {inspection} to inspect child state and continue the task.",
    ))
}

fn child_wait_lease_expired_message(child_ids: &[String]) -> Message {
    let inspection = serde_json::json!({"intent": "inspect", "message": "tree"});
    child_wait_watchdog_resume_message(format!(
        "Runtime notification: the wait lease for child session(s) [{}] expired before they all \
         reported completion. They were NOT cancelled and may still be running or already \
         finished — verify their actual status by calling SubAgent with {inspection} before assuming \
         anything, then continue the task.",
        child_ids.join(", ")
    ))
}

/// Child-wait watchdog (issue #546): the heartbeat backstop for parents
/// suspended on `waiting_for_children`.
///
/// The primary wake is the event-driven completion push (child terminal →
/// [`ChildCompletionHandler::on_child_completed`] → resume). This sweeper
/// exists because ANY break in that chain — a panicked child task, a dead
/// spawn scheduler, a process restart, a clobbered/exhausted resume, a wait
/// registered over an already-terminal child — previously stranded the parent
/// forever. It mirrors the bash backstop's philosophy: coarse, cheap, yields
/// to the push, and only acts when the durable state proves nothing else can.
///
/// All wake decisions funnel through the SAME machinery the push uses
/// (synthetic/replayed completions → `on_child_completed`, per-parent
/// serialization via [`session_resume_lock`]), so there is exactly one resume
/// implementation.
impl ChildCompletionCoordinator {
    /// Spawn the watchdog: one boot-time reconciliation pass, then a sweep
    /// every [`CHILD_WAIT_SWEEP_INTERVAL_SECS`]. Call once at server startup.
    pub fn spawn_child_wait_watchdog(self: &Arc<Self>) {
        let coordinator = Arc::clone(self);
        tokio::spawn(async move {
            use futures::FutureExt;
            if std::panic::AssertUnwindSafe(coordinator.reconcile_orphans_at_boot())
                .catch_unwind()
                .await
                .is_err()
            {
                tracing::error!("child-wait watchdog: boot reconciliation panicked");
            }
            let mut ticker = tokio::time::interval(std::time::Duration::from_secs(
                CHILD_WAIT_SWEEP_INTERVAL_SECS,
            ));
            ticker.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
            // Skip the immediate tick (boot reconciliation just ran).
            ticker.tick().await;
            loop {
                ticker.tick().await;
                if std::panic::AssertUnwindSafe(coordinator.sweep_child_waits())
                    .catch_unwind()
                    .await
                    .is_err()
                {
                    tracing::error!("child-wait watchdog: sweep panicked; continuing");
                }
            }
        });
    }

    /// One-shot inspection uses a fixed boot cutoff. A missing local runner
    /// does not prove a remote Worker stopped; Storage selects exact trusted
    /// terminal facts or revokes an expired logical lease under its guards.
    async fn reconcile_orphans_at_boot(&self) {
        let cutoff = Utc::now();

        // (1) Inspect old running candidates. The shared paced index sweep also
        // repairs already-terminal records; this legacy entry remains narrow.
        let running = self
            .storage
            .list_sessions_by_run_status("running")
            .await
            .unwrap_or_default();
        for (child_id, parent_id) in running {
            let Some(parent_id) = parent_id else { continue };
            if self.runner_is_running(&child_id).await {
                continue;
            }
            let Some(control_plane) = self.load_control_plane(&child_id).await else {
                continue;
            };
            // A child that started AFTER boot is alive by definition — the
            // cutoff guards the tiny window where a fresh spawn races this scan.
            if control_plane.updated_at >= cutoff {
                continue;
            }
            tracing::warn!(
                child_session_id = %child_id,
                parent_session_id = %parent_id,
                "boot reconciliation: inspecting exact Child terminal authority"
            );
            self.synthesize_child_completion(
                &parent_id,
                &child_id,
                "error",
                Some(
                    "orphaned by server restart: the process died while this child session \
                     was running"
                        .to_string(),
                ),
                cutoff,
            )
            .await;
        }

        // (2) Sessions suspended on `waiting_for_bash`: their backstop poll was
        // an in-memory task that died with the process. Re-arm it — with the
        // shell registry empty after a restart it resumes on its first check.
        let suspended = self
            .storage
            .list_sessions_by_run_status("suspended")
            .await
            .unwrap_or_default();
        for (session_id, _) in suspended {
            let Some(control_plane) = self.load_control_plane(&session_id).await else {
                continue;
            };
            if control_plane
                .metadata
                .get("runtime.suspend_reason")
                .map(String::as_str)
                != Some("waiting_for_bash")
            {
                continue;
            }
            if let Some(wait) = read_runtime_state(&control_plane).waiting_for_bash {
                tracing::warn!(
                    %session_id,
                    "boot reconciliation: re-arming bash self-resume backstop lost in restart"
                );
                let coordinator = self.clone();
                tokio::spawn(async move {
                    coordinator
                        .bash_self_resume(session_id, wait.bash_ids)
                        .await;
                });
            }
        }
    }

    async fn runner_is_running(&self, session_id: &str) -> bool {
        let runners = self.agent_runners.read().await;
        runners
            .get(session_id)
            .is_some_and(|runner| matches!(runner.status, AgentStatus::Running))
    }

    async fn load_control_plane(&self, session_id: &str) -> Option<Session> {
        match self.storage.load_runtime_control_plane(session_id).await {
            Ok(session) => session,
            Err(error) => {
                tracing::warn!(
                    %session_id,
                    %error,
                    "child-wait watchdog: failed to load session control plane"
                );
                None
            }
        }
    }

    /// One sweep: inspect every session whose index status is `suspended`.
    /// Cheap in the common case — the candidate list is small and each check is
    /// a sidecar (control-plane) load.
    async fn sweep_child_waits(&self) {
        let suspended = match self.storage.list_sessions_by_run_status("suspended").await {
            Ok(entries) => entries,
            Err(error) => {
                tracing::warn!(%error, "child-wait watchdog: failed to list suspended sessions");
                return;
            }
        };
        for (session_id, _) in suspended {
            self.sweep_one_suspended_session(&session_id).await;
        }
    }

    async fn sweep_one_suspended_session(&self, session_id: &str) {
        // A live runner means the session already resumed (the index status
        // only advances at its next terminal) — nothing to do.
        if self.runner_is_running(session_id).await {
            return;
        }
        let Some(session) = self.load_control_plane(session_id).await else {
            return;
        };
        let runtime_state = read_runtime_state(&session);
        let suspend_reason = session
            .metadata
            .get("runtime.suspend_reason")
            .map(String::as_str)
            .unwrap_or_default()
            .to_string();
        match (
            suspend_reason.as_str(),
            runtime_state.waiting_for_children.clone(),
        ) {
            // Human-gated or bash-owned waits: not ours to time out.
            ("waiting_for_bash", _)
            | ("awaiting_clarification", _)
            | ("awaiting_parent_approval", _) => {}
            (_, Some(wait)) => self.sweep_child_wait(session_id, wait).await,
            // Suspended with NO armed wait: either the coordinator cleared the
            // wait but its resume never spawned, or state is half-cleared.
            ("waiting_for_children", None) | ("", None) => {
                self.rescue_stranded_resume(session_id).await;
            }
            _ => {}
        }
    }

    /// Evaluate one armed child wait against reality and act on what the
    /// durable state proves: dead children are synthesized terminal, an
    /// already-satisfied wait gets its lost wake replayed, an expired lease or
    /// empty wait force-resumes the parent.
    async fn sweep_child_wait(&self, parent_session_id: &str, wait: WaitingForChildrenState) {
        let now = Utc::now();

        if wait.child_session_ids.is_empty() {
            tracing::warn!(
                %parent_session_id,
                "child-wait watchdog: wait armed over an empty child set; force-resuming"
            );
            self.force_resume_child_wait(parent_session_id, empty_child_wait_message())
                .await;
            return;
        }

        // The 6h lease (previously written but never read). Expiry does not
        // kill children — child runners own child liveness; the parent is
        // resumed with a verify-don't-assume note.
        if wait.timeout_at.is_some_and(|deadline| now >= deadline) {
            tracing::warn!(
                %parent_session_id,
                "child-wait watchdog: wait lease expired; force-resuming parent"
            );
            self.force_resume_child_wait(
                parent_session_id,
                child_wait_lease_expired_message(&wait.child_session_ids),
            )
            .await;
            return;
        }

        // Yield to the event-driven push on fresh waits.
        if now.signed_duration_since(wait.registered_at).num_seconds()
            < CHILD_WAIT_REGISTRATION_GRACE_SECS
        {
            return;
        }

        let statuses: HashMap<String, Option<String>> = self
            .storage
            .list_child_run_statuses(parent_session_id)
            .await
            .unwrap_or_default()
            .into_iter()
            .collect();

        struct DeadChild {
            child_id: String,
            status: String,
            reason: String,
            /// `false` = the id does not name a child of THIS parent (wait ids
            /// are model-provided and unvalidated): publish the wake but never
            /// persist onto / cancel / finalize the named session.
            owned: bool,
        }

        let mut terminal: Vec<(String, String)> = Vec::new();
        let mut dead: Vec<DeadChild> = Vec::new();
        for child_id in &wait.child_session_ids {
            let status = statuses.get(child_id).and_then(|status| status.as_deref());
            if let Some(status) = status {
                if is_terminal_child_status(status) {
                    terminal.push((child_id.clone(), status.to_string()));
                    continue;
                }
            }
            if !is_dead_child_candidate_status(status) {
                continue;
            }

            // Ownership check BEFORE anything destructive: `SubAgent.wait` ids
            // are model-provided and unvalidated, so an id absent from the
            // parent-scoped status map may name a REAL session in another tree
            // (a grandchild, a foreign root). Such a session must never be
            // mutated, cancelled, or finalized here — but the parent's bogus
            // wait entry still needs a synthetic completion to clear it.
            let control_plane = self.load_control_plane(child_id).await;
            let owned = control_plane
                .as_ref()
                .is_some_and(|cp| cp.parent_session_id.as_deref() == Some(parent_session_id));
            if !owned {
                dead.push(DeadChild {
                    child_id: child_id.clone(),
                    status: "error".to_string(),
                    reason: if control_plane.is_some() {
                        "waited-on session id is not a child of this session; clearing it \
                         from the wait without touching that session"
                            .to_string()
                    } else {
                        "waited-on child session does not exist".to_string()
                    },
                    owned: false,
                });
                continue;
            }

            let runner = { self.agent_runners.read().await.get(child_id).cloned() };
            match runner {
                Some(runner) if matches!(runner.status, AgentStatus::Running) => {
                    // A live-looking runner: only intervene when it is frozen
                    // far PAST the per-child liveness limits — which proves the
                    // per-child watchdog machinery itself is dead (task
                    // panicked or lost), because it would have cancelled and
                    // published a timeout long before.
                    let last_activity = runner.last_activity_at().unwrap_or(runner.started_at);
                    let idle_secs = now.signed_duration_since(last_activity).num_seconds();
                    let total_secs = now.signed_duration_since(runner.started_at).num_seconds();
                    let policy = match &control_plane {
                        Some(child) => {
                            crate::runtime::execution::spawn::watchdog_policy_for_session(child)
                        }
                        None => Default::default(),
                    };
                    let idle_limit = policy.max_idle_secs.saturating_add(STALE_RUNNER_SLACK_SECS);
                    let total_limit = policy
                        .max_total_secs
                        .saturating_add(STALE_RUNNER_SLACK_SECS);
                    if idle_secs >= idle_limit || total_secs >= total_limit {
                        runner.cancel_token.cancel();
                        dead.push(DeadChild {
                            child_id: child_id.clone(),
                            status: "timeout".to_string(),
                            reason: format!(
                                "child runner stalled: no events for {idle_secs}s \
                                 (limit {idle_limit}s including watchdog slack); \
                                 force-finalized by the child-wait watchdog"
                            ),
                            owned: true,
                        });
                    }
                }
                _ => {
                    // Nothing is driving this child, yet its index status will
                    // never advance by itself. The grace covers the enqueue →
                    // running-marker window and slow spawn queues.
                    let quiet_secs = control_plane
                        .as_ref()
                        .map(|child| now.signed_duration_since(child.updated_at).num_seconds())
                        .unwrap_or(i64::MAX);
                    if quiet_secs >= DEAD_CHILD_GRACE_SECS {
                        dead.push(DeadChild {
                            child_id: child_id.clone(),
                            status: "error".to_string(),
                            reason: format!(
                                "child runner lost (crashed task, dropped spawn job, or \
                                 process restart): index status {status:?} with no live \
                                 runner driving it"
                            ),
                            owned: true,
                        });
                    }
                }
            }
        }

        if !dead.is_empty() {
            for entry in dead {
                tracing::warn!(
                    %parent_session_id,
                    child_session_id = %entry.child_id,
                    status = %entry.status,
                    reason = %entry.reason,
                    owned = entry.owned,
                    "child-wait watchdog: synthesizing terminal completion for dead child"
                );
                if entry.owned {
                    self.synthesize_child_completion(
                        parent_session_id,
                        &entry.child_id,
                        &entry.status,
                        Some(entry.reason),
                        Utc::now(),
                    )
                    .await;
                } else {
                    // Foreign / nonexistent id: wake the parent only.
                    self.publish_synthetic_completion(
                        parent_session_id,
                        &entry.child_id,
                        &entry.status,
                        Some(entry.reason),
                    )
                    .await;
                }
            }
            // The publishes above re-evaluate the wait policy themselves.
            return;
        }

        // A lost callback must not hide one ready result behind an unfinished
        // All wait. Replay every observed terminal generation through the same
        // source/ownership checks and permanent Inbox receipts as the push path.
        // Error-like outcomes go first so FirstError releases promptly.
        terminal.sort_by_key(|(_, status)| !is_error_like(status));
        for (child_id, status) in terminal {
            tracing::debug!(
                %parent_session_id,
                child_session_id = %child_id,
                "child-wait watchdog: replaying terminal child outcome"
            );
            let control_plane = self.load_control_plane(&child_id).await;
            let mut completion = ChildCompletion {
                parent_session_id: parent_session_id.into(),
                child_session_id: child_id.clone(),
                status,
                error: control_plane.as_ref().and_then(Session::last_run_error),
                completed_at: Utc::now(),
                source: None,
            };
            // Lost callbacks replay the committed source, not a synthetic
            // latest answer. Check ownership before loading any transcript.
            if control_plane
                .as_ref()
                .is_some_and(|child| child.parent_session_id.as_deref() == Some(parent_session_id))
            {
                let Ok(Some(child)) = self.storage.load_session(&child_id).await else {
                    continue;
                };
                completion.source = ChildCompletionSource::from_committed_session(&child)
                    .filter(|source| source.matches_completion(&completion));
                if (ChildCompletionSource::has_source_record(&child)
                    || repaired_child_requires_source(&child))
                    && completion.source.is_none()
                {
                    tracing::warn!(%child_id,
                            "unavailable committed replay source; leaving wait armed");
                    continue;
                }
                completion.completed_at = child.updated_at;
            }
            self.publish_completion(completion).await;
        }
    }

    /// Repair only the selected Child's logical state. The caller's requested
    /// status, local runner absence and cancel ACK grant no terminal authority,
    /// Worker-stop proof or parent delivery permission. Existing published
    /// sources retain their normal replay path; a repair snapshot is no source.
    async fn synthesize_child_completion(
        &self,
        parent_session_id: &str,
        child_session_id: &str,
        status: &str,
        error: Option<String>,
        before: chrono::DateTime<Utc>,
    ) {
        let Ok(Some(child)) = self.storage.load_session(child_session_id).await else {
            return;
        };
        if child.parent_session_id.as_deref() != Some(parent_session_id) {
            return;
        }
        let cache = self.sessions.clone();
        let selected = child.clone();
        if let Err(repair_error) = self
            .storage
            .reconcile_session_runtime(
                &child,
                before,
                Some("Child activation lease expired without confirmed Host completion"),
                Arc::new(move |saved| {
                    crate::SessionRepository::publish_reconciled_child(&cache, &selected, saved)
                }),
            )
            .await
        {
            tracing::warn!(%child_session_id, requested_status = status, ?error, %repair_error,
                "Child terminal repair unconfirmed; no completion or Worker-stop proof published");
        }
    }

    async fn publish_synthetic_completion(
        &self,
        parent_session_id: &str,
        child_session_id: &str,
        status: &str,
        error: Option<String>,
    ) {
        self.publish_completion(ChildCompletion {
            parent_session_id: parent_session_id.into(),
            child_session_id: child_session_id.into(),
            status: status.into(),
            error,
            completed_at: Utc::now(),
            source: None,
        })
        .await;
    }

    async fn publish_completion(&self, completion: ChildCompletion) {
        let publisher =
            crate::runtime::execution::session_events::ReplayableSessionEventPublisher::new(
                self.agent_runners.clone(),
                self.session_event_senders.clone(),
                self.account_feed_inbox.clone(),
            );
        let handler: Arc<dyn ChildCompletionHandler> = Arc::new(self.clone());
        crate::runtime::execution::spawn::publish_child_completion(
            &publisher,
            Some(handler),
            completion,
        )
        .await;
    }

    /// A parent whose wait was already cleared (resume message appended) but
    /// whose resume never spawned — retry-budget exhaustion, root-tools not
    /// yet initialized, or a restart between clear and spawn. Detected by: no
    /// live runner, no armed wait, and a pending hidden runtime resume message
    /// as the LAST message. Resume is all that's left to do.
    async fn rescue_stranded_resume(&self, session_id: &str) {
        let Some(session) = self.load_session(session_id).await else {
            return;
        };
        let pending_runtime_resume = session.messages.last().is_some_and(|message| {
            matches!(message.role, Role::User)
                && message
                    .metadata
                    .as_ref()
                    .is_some_and(|meta| meta.get(RUNTIME_RESUME_MESSAGE_KIND_KEY).is_some())
        });
        if !pending_runtime_resume {
            return;
        }
        tracing::warn!(
            %session_id,
            "child-wait watchdog: stranded resume detected (wait cleared, resume never \
             spawned); resuming"
        );
        self.resume_parent(session_id.to_string()).await;
    }

    /// Clear the parent's child wait, append `resume_message`, and drive the
    /// resume — with a bounded clobber-retry mirroring
    /// [`Self::perform_bash_resume`]: a suspending runner's one-shot finalize
    /// save can land after ours and revert the wait while dropping the
    /// message; we detect the re-armed wait and re-clear.
    async fn force_resume_child_wait(&self, session_id: &str, resume_message: Message) {
        const MAX_ATTEMPTS: u8 = 5;
        let lock = session_resume_lock(session_id);
        for attempt in 0..MAX_ATTEMPTS {
            if attempt > 0 {
                tokio::time::sleep(Duration::from_millis(200)).await;
            }
            {
                let _held = lock.lock().await;
                let Some(mut session) = self.load_session(session_id).await else {
                    return;
                };
                let mut runtime_state = read_runtime_state(&session);
                if runtime_state.waiting_for_children.is_none() {
                    // Another source already resumed this parent.
                    return;
                }
                runtime_state.waiting_for_children = None;
                runtime_state.status = AgentStatusState::Idle;
                runtime_state.suspension = None;
                write_runtime_state(&mut session, &runtime_state);
                session.metadata.remove("runtime.suspend_reason");
                session.add_message(resume_message.clone());
                session.updated_at = Utc::now();
                self.save_and_cache(&mut session).await;
            }
            let outcome = self.resume_parent(session_id.to_string()).await;
            match outcome {
                ResumeOutcome::Started { .. } | ResumeOutcome::NotFound => return,
                ResumeOutcome::Completed | ResumeOutcome::AlreadyRunning { .. } => {
                    // Only retry when the persisted wait was clobbered back to
                    // armed; if it stayed cleared with the message intact, the
                    // next sweep's stranded-resume rescue finishes the job.
                    let clobbered = self
                        .load_session(session_id)
                        .await
                        .map(|session| read_runtime_state(&session).waiting_for_children.is_some())
                        .unwrap_or(false);
                    if !clobbered {
                        return;
                    }
                }
            }
        }
        tracing::error!(
            %session_id,
            "child-wait watchdog: force-resume exhausted its clobber-retry budget"
        );
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    include!("child_completion_gold_origin_tests.rs");
    include!("child_completion_background_tests.rs");
    include!("child_completion_exec_tests.rs");
    use bamboo_domain::session::runtime_state::SuspensionState;

    // Exercise the exact auxiliary resolver installed by child-completion
    // resume. This fixture fixes the provider publication boundary; it does
    // not itself run a full child/parent session.
    #[tokio::test]
    async fn child_resume_legacy_memory_rejects_cached_route_during_publication() {
        use bamboo_llm::LLMError;
        use futures::StreamExt;
        use serde_json::json;
        use wiremock::matchers::{method, path};
        use wiremock::{Mock, MockServer, ResponseTemplate};

        async fn dispatch_background(
            models: crate::AuxiliaryModelConfig,
        ) -> Result<bool, LLMError> {
            match (
                models.background_model_name,
                models.background_model_provider,
            ) {
                (None, None) => Ok(false),
                (Some(model), Some(provider)) => {
                    let mut stream = provider
                        .chat_stream(
                            &[Message::user("synthetic memory query")],
                            &[],
                            Some(32),
                            &model,
                        )
                        .await?;
                    while let Some(chunk) = stream.next().await {
                        chunk?;
                    }
                    Ok(true)
                }
                _ => panic!("background name and provider must remain paired"),
            }
        }

        let mock = MockServer::start().await;
        Mock::given(method("POST"))
            .and(path("/chat/completions"))
            .respond_with(
                ResponseTemplate::new(200)
                    .insert_header("content-type", "text/event-stream")
                    .set_body_string(concat!(
                        "data: {\"choices\":[{\"index\":0,\"delta\":{\"content\":\"ok\"}}]}\n\n",
                        "data: [DONE]\n\n",
                    )),
            )
            .mount(&mock)
            .await;
        let home = tempfile::tempdir().expect("isolated test home");
        let mut baseline = Config::default();
        baseline.provider = "account-a".into();
        baseline.default_provider_instance = Some("account-a".into());
        baseline.defaults = None;
        baseline.http_proxy = "http://127.0.0.1:9".into();
        baseline.https_proxy.clear();
        baseline.proxy_auth = None;
        baseline
            .memory_mut()
            .get_or_insert_with(Default::default)
            .background_model = Some("memory-only".into());
        baseline.provider_instances.insert(
            "account-a".into(),
            serde_json::from_value(json!({
                "provider_type":"openai", "enabled":true,
                "api_key":"synthetic-child-only", "base_url":mock.uri(), "model":"chat-a"
            }))
            .unwrap(),
        );
        let registry = Arc::new(
            ProviderRegistry::from_config(&baseline, home.path().to_path_buf())
                .await
                .expect("initial managed generation"),
        );
        let live = Arc::new(RwLock::new(baseline.clone()));
        let resolver = make_resume_auxiliary_model_resolver(
            live.clone(),
            baseline.clone(),
            registry.clone(),
            "account-a".into(),
        );
        let initial = resolver();
        assert_eq!(
            initial.background_model_name.as_deref(),
            Some("memory-only")
        );
        assert!(Arc::ptr_eq(
            initial.background_model_provider.as_ref().unwrap(),
            &registry.get("account-a").unwrap()
        ));

        let mut denied = baseline;
        denied
            .provider_instances
            .get_mut("account-a")
            .unwrap()
            .extra
            .insert("runtime_models".into(), json!(["chat-a"]));
        let denied_registry = ProviderRegistry::from_config(&denied, home.path().to_path_buf())
            .await
            .expect("denied managed generation");
        let mut writer = live.write().await;
        *writer = denied.clone();
        let blocked = resolver();
        let blocked_name = blocked.background_model_name.clone();
        let blocked_has_provider = blocked.background_model_provider.is_some();
        registry.replace_with(denied_registry);
        drop(writer);
        let blocked_dispatched =
            tokio::time::timeout(Duration::from_secs(5), dispatch_background(blocked))
                .await
                .expect("blocked dispatch should finish")
                .expect("loopback response");
        let blocked_requests = mock.received_requests().await.unwrap().len();
        let after = resolver();
        let after_name = after.background_model_name.clone();
        let after_dispatched =
            tokio::time::timeout(Duration::from_secs(5), dispatch_background(after))
                .await
                .expect("denied dispatch should finish")
                .expect("loopback response");
        let after_requests = mock.received_requests().await.unwrap().len();

        let mut allowed = denied;
        allowed
            .provider_instances
            .get_mut("account-a")
            .unwrap()
            .extra
            .insert("runtime_models".into(), json!(["chat-a", "memory-only"]));
        let allowed_registry = ProviderRegistry::from_config(&allowed, home.path().to_path_buf())
            .await
            .expect("allowed managed generation");
        let mut writer = live.write().await;
        *writer = allowed;
        registry.replace_with(allowed_registry);
        drop(writer);
        let restored = resolver();
        let restored_name = restored.background_model_name.clone();
        let same_account = restored
            .background_model_provider
            .as_ref()
            .is_some_and(|handle| Arc::ptr_eq(handle, &registry.get("account-a").unwrap()));
        let restored_dispatched =
            tokio::time::timeout(Duration::from_secs(5), dispatch_background(restored))
                .await
                .expect("restored dispatch should finish")
                .expect("loopback response");
        let requests = mock.received_requests().await.unwrap();
        let new_requests = requests.len() - after_requests;
        let final_model = requests.last().map(|request| {
            let body: serde_json::Value = serde_json::from_slice(&request.body).unwrap();
            body["model"].as_str().unwrap().to_string()
        });
        assert_eq!(
            (
                (
                    blocked_name,
                    blocked_has_provider,
                    blocked_dispatched,
                    blocked_requests
                ),
                (after_name, after_dispatched, after_requests),
                (
                    restored_name,
                    same_account,
                    restored_dispatched,
                    new_requests,
                    final_model
                )
            ),
            (
                (None, false, false, 0),
                (None, false, 0),
                (
                    Some("memory-only".to_string()),
                    true,
                    true,
                    1,
                    Some("memory-only".to_string())
                )
            ),
        );
    }

    #[test]
    fn skill_request_child_provider_constructor_keeps_data_absent_and_rejects_injection() {
        let provider = Message::user("rendered child outcome");
        let presentation = session_provider_message(&provider);
        assert_eq!(presentation.content.skill_request, None);
        assert_eq!(presentation.content.text, provider.content);
        let mut envelope = SessionMessageEnvelope::user_input("parent", "unused");
        envelope.source = SessionMessageSource::Runtime {
            subsystem: "child_completion_coordinator".into(),
        };
        envelope.kind = SessionMessageKind::ChildOutcome;
        envelope.body = SessionMessageBody::ChildOutcome(SessionChildOutcome {
            child_session_id: "child".into(),
            status: "completed".into(),
            result: Some("typed result".into()),
            error: None,
            provider_message: Some(presentation),
        });
        envelope.validate().unwrap();
        assert_eq!(
            envelope.to_provider_message().unwrap().content,
            "rendered child outcome"
        );
        if let SessionMessageBody::ChildOutcome(outcome) = &mut envelope.body {
            outcome
                .provider_message
                .as_mut()
                .unwrap()
                .content
                .skill_request = Some(bamboo_domain::SessionSkillRequest {
                mode: None,
                selections: vec![bamboo_domain::SessionSkillSelection {
                    id: "review".into(),
                    source: "builtin".into(),
                    revision: 7,
                    args: serde_json::Value::Null,
                }],
            });
        }
        assert_eq!(
            envelope.validate(),
            Err(bamboo_domain::SessionMessageValidationError::KindSourceBodyMismatch)
        );
    }

    #[tokio::test]
    async fn cancelled_resume_waiters_do_not_retain_historical_parent_ids() {
        for index in 0..512 {
            let id = format!("resume-lock-reclaim-{index}");
            let owner = session_resume_lock(&id);
            let held = owner.lock().await;
            let waiter = session_resume_lock(&id);
            let mut waiting = Box::pin(waiter.lock());
            assert!(futures::poll!(waiting.as_mut()).is_pending());
            drop(held);
            drop(owner);
            drop(waiting);
            drop(waiter);
            assert!(!parent_locks().lock().recover_poison().contains_key(&id));
        }
    }

    #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
    async fn resume_lock_reclamation_preserves_exclusive_parent_wake_ownership() {
        use std::sync::atomic::{AtomicUsize, Ordering};
        let active = Arc::new(AtomicUsize::new(0));
        let mut tasks = Vec::new();
        for _ in 0..32 {
            let active = active.clone();
            tasks.push(tokio::spawn(async move {
                for _ in 0..16 {
                    let lease = session_resume_lock("resume-lock-exclusive");
                    let _guard = lease.lock().await;
                    assert_eq!(active.fetch_add(1, Ordering::SeqCst), 0);
                    tokio::task::yield_now().await;
                    assert_eq!(active.fetch_sub(1, Ordering::SeqCst), 1);
                }
            }));
        }
        for task in tasks {
            task.await.unwrap();
        }
        assert!(!parent_locks()
            .lock()
            .recover_poison()
            .contains_key("resume-lock-exclusive"));
    }
    use bamboo_agent_core::Message;
    use bamboo_domain::SessionInboxPort;
    use futures::stream;
    use std::sync::atomic::{AtomicUsize, Ordering};

    struct FailingRuntimeSaveStorage {
        inner: Arc<bamboo_storage::SessionStoreV2>,
    }

    #[async_trait]
    impl Storage for FailingRuntimeSaveStorage {
        async fn save_session(&self, session: &Session) -> std::io::Result<()> {
            self.inner.save_session(session).await
        }

        async fn load_session(&self, id: &str) -> std::io::Result<Option<Session>> {
            self.inner.load_session(id).await
        }

        async fn delete_session(&self, id: &str) -> std::io::Result<bool> {
            self.inner.delete_session(id).await
        }

        async fn load_runtime_control_plane(&self, id: &str) -> std::io::Result<Option<Session>> {
            self.inner.load_runtime_control_plane(id).await
        }

        async fn save_runtime_state(&self, _session: &Session) -> std::io::Result<()> {
            Err(std::io::Error::other("injected sidecar save failure"))
        }
    }

    struct EmptyTools;

    #[async_trait]
    impl bamboo_agent_core::tools::ToolExecutor for EmptyTools {
        async fn execute(
            &self,
            _call: &bamboo_agent_core::tools::ToolCall,
        ) -> Result<bamboo_agent_core::tools::ToolResult, bamboo_agent_core::tools::ToolError>
        {
            Err(bamboo_agent_core::tools::ToolError::NotFound(
                "no tools".to_string(),
            ))
        }

        fn list_tools(&self) -> Vec<bamboo_agent_core::tools::ToolSchema> {
            Vec::new()
        }
    }

    struct CompletedTestProvider;

    #[async_trait]
    impl bamboo_llm::LLMProvider for CompletedTestProvider {
        async fn chat_stream(
            &self,
            _messages: &[Message],
            _tools: &[bamboo_agent_core::tools::ToolSchema],
            _max_output_tokens: Option<u32>,
            _model: &str,
        ) -> Result<bamboo_llm::LLMStream, bamboo_llm::LLMError> {
            let chunks: Vec<bamboo_llm::provider::Result<bamboo_llm::LLMChunk>> = vec![
                Ok(bamboo_llm::LLMChunk::Token("done".to_string())),
                Ok(bamboo_llm::LLMChunk::Done),
            ];
            Ok(Box::pin(stream::iter(chunks)))
        }
    }

    struct CountingActivationSpawner {
        reservations: Arc<AtomicUsize>,
        launches: Arc<AtomicUsize>,
    }

    #[async_trait]
    impl SessionActivationSpawner for CountingActivationSpawner {
        async fn reserve_activation(
            &self,
            target_session_id: &str,
            inbox_generation: u64,
        ) -> Result<SessionActivationReserveOutcome, bamboo_domain::SessionActivationError>
        {
            self.reservations.fetch_add(1, Ordering::SeqCst);
            let launches = self.launches.clone();
            Ok(SessionActivationReserveOutcome::Reserved(
                SessionActivationLaunch::new(
                    format!("{target_session_id}-{inbox_generation}"),
                    move || {
                        launches.fetch_add(1, Ordering::SeqCst);
                    },
                ),
            ))
        }
    }

    async fn save_sourced_test_child(store: &bamboo_storage::SessionStoreV2, child: &Session) {
        let mut saved = child.clone();
        ChildCompletionSource::prepare(
            &mut saved,
            &format!("test-run-{}", child.id),
            &Default::default(),
        );
        store.save_session(&saved).await.unwrap();
    }

    async fn repaired_unactivated_test_child(
        store: &bamboo_storage::SessionStoreV2,
        parent: &Session,
        id: &str,
    ) -> Session {
        let mut child = Session::new_child_of(id, parent, "model", "Child");
        child.agent_runtime_state = Some(AgentRuntimeState::new("unactivated-host-pipeline"));
        child.set_last_run_status("running");
        child.add_message(Message::assistant(
            "UNSEALED_REPAIR_MUST_NOT_BE_IMPORTED",
            None,
        ));
        store.save_session(&child).await.unwrap();
        store
            .begin_host_run(&child, "unactivated-host-A")
            .await
            .unwrap();
        let mut terminal = store.load_session(id).await.unwrap().unwrap();
        terminal.set_last_run_status("cancelled");
        // An actual legacy Host Main commit with the audited typed split.
        store.save_session(&terminal).await.unwrap();
        let expected = store.load_session(id).await.unwrap().unwrap();
        let repaired = store
            .reconcile_session_runtime(&expected, Utc::now(), None, Arc::new(|_| {}))
            .await
            .unwrap()
            .unwrap();
        assert_eq!(
            repaired.agent_runtime_state.as_ref().unwrap().status,
            AgentStatusState::Cancelled
        );
        assert_eq!(
            repaired
                .metadata
                .get("runtime.reconciled_terminal_generation")
                .map(String::as_str),
            Some("0")
        );
        assert!(!repaired
            .metadata
            .contains_key("runtime.reconciled_activation_fence"));
        assert!(ChildCompletionSource::from_committed_session(&repaired).is_none());
        repaired
    }

    #[tokio::test]
    async fn recovered_source_less_sibling_never_satisfies_all_or_stages_with_legal_short_circuit()
    {
        for policy in [
            ChildWaitPolicy::All,
            ChildWaitPolicy::Any,
            ChildWaitPolicy::FirstError,
        ] {
            let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
            save_waiting_test_parent(
                &store,
                "repair-parent",
                vec!["repair-A".into(), "sealed-B".into()],
                policy,
            )
            .await;
            let parent = store.load_session("repair-parent").await.unwrap().unwrap();
            let repaired = repaired_unactivated_test_child(&store, &parent, "repair-A").await;
            // A direct legacy-looking callback cannot borrow the repaired fact.
            ChildCompletionHandler::on_child_completed(
                coordinator.as_ref(),
                ChildCompletion {
                    parent_session_id: parent.id.clone(),
                    child_session_id: repaired.id.clone(),
                    status: "cancelled".into(),
                    error: None,
                    completed_at: repaired.updated_at,
                    source: None,
                },
            )
            .await;
            assert_eq!(inbox.inspect(&parent.id).await.unwrap().pending, 0);
            let b = save_completion_test_child(&store, &parent.id, "sealed-B", "error").await;
            ChildCompletionHandler::on_child_completed(coordinator.as_ref(), b).await;
            let claims = inbox.claim(&parent.id, 8).await.unwrap();
            // Preserve the existing incremental sealed-B notification, even
            // while All still waits. A contributes neither count nor outcome.
            assert_eq!(claims.len(), 1);
            assert!(
                matches!(&claims[0].envelope.body, SessionMessageBody::ChildOutcome(outcome) if outcome.child_session_id == "sealed-B")
            );
            assert_eq!(launches.load(Ordering::SeqCst), 1);
            let current = store.load_session(&parent.id).await.unwrap().unwrap();
            assert_eq!(
                read_runtime_state(&current).waiting_for_children.is_some(),
                policy == ChildWaitPolicy::All
            );
            assert!(!serde_json::to_string(&current)
                .unwrap()
                .contains("UNSEALED_REPAIR_MUST_NOT_BE_IMPORTED"));
            for raw in ["bad", "1"] {
                let mut invalid = repaired.clone();
                invalid
                    .metadata
                    .insert("runtime.reconciled_terminal_generation".into(), raw.into());
                assert!(repaired_child_requires_source(&invalid));
            }
        }
    }

    #[tokio::test]
    async fn recovered_child_read_failure_cannot_become_neutral_parent_permission() {
        let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
        save_waiting_test_parent(
            &store,
            "fault-parent",
            vec!["fault-child".into()],
            ChildWaitPolicy::All,
        )
        .await;
        let parent = store.load_session("fault-parent").await.unwrap().unwrap();
        let child = repaired_unactivated_test_child(&store, &parent, "fault-child").await;
        let dir = store
            .bamboo_home_dir()
            .join(store.resolve_rel_path(&child.id).await.unwrap());
        let main = dir.join("session.json");
        let held = dir.join("held-main.json");
        tokio::fs::rename(&main, &held).await.unwrap();
        tokio::fs::create_dir(&main).await.unwrap();
        let callback = ChildCompletion {
            parent_session_id: parent.id.clone(),
            child_session_id: child.id,
            status: "cancelled".into(),
            error: None,
            completed_at: child.updated_at,
            source: None,
        };
        ChildCompletionHandler::on_child_completed(coordinator.as_ref(), callback.clone()).await;
        assert!(inbox.claim(&parent.id, 8).await.unwrap().is_empty());
        assert_eq!(launches.load(Ordering::SeqCst), 0);
        assert!(
            read_runtime_state(&store.load_session(&parent.id).await.unwrap().unwrap())
                .waiting_for_children
                .is_some()
        );
        tokio::fs::remove_dir(&main).await.unwrap();
        tokio::fs::rename(&held, &main).await.unwrap();
        ChildCompletionHandler::on_child_completed(coordinator.as_ref(), callback).await;
        assert!(inbox.claim(&parent.id, 8).await.unwrap().is_empty());
        assert_eq!(launches.load(Ordering::SeqCst), 0);
    }

    #[tokio::test]
    async fn recovered_generation_does_not_block_exact_later_queued_cancel() {
        let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
        let mut parent = Session::new("cancel-parent", "model");
        store.save_session(&parent).await.unwrap();
        let mut child = repaired_unactivated_test_child(&store, &parent, "cancel-child").await;
        child.advance_child_launch_generation().unwrap();
        child.add_message(Message::user("queued B input"));
        child.set_last_run_status("pending");
        store.save_session(&child).await.unwrap();
        child = store.load_session(&child.id).await.unwrap().unwrap();
        child.cancel_child_launch_generation();
        child.set_last_run_status("cancelled");
        store.save_session(&child).await.unwrap();
        child = store.load_session(&child.id).await.unwrap().unwrap();
        assert!(!child.host_run_matches_launch());
        assert!(!repaired_child_requires_source(&child));
        let mut runtime = AgentRuntimeState::new("fresh-wait");
        runtime.status = AgentStatusState::Suspended;
        runtime.waiting_for_children = Some(WaitingForChildrenState::for_children(
            vec![child.id.clone()],
            ChildWaitPolicy::All,
            Utc::now(),
        ));
        write_runtime_state(&mut parent, &runtime);
        store.save_session(&parent).await.unwrap();
        ChildCompletionHandler::on_child_completed(
            coordinator.as_ref(),
            ChildCompletion {
                parent_session_id: parent.id.clone(),
                child_session_id: child.id,
                status: "cancelled".into(),
                error: None,
                completed_at: child.updated_at,
                source: None,
            },
        )
        .await;
        assert!(
            read_runtime_state(&store.load_session(&parent.id).await.unwrap().unwrap())
                .waiting_for_children
                .is_none()
        );
        assert_eq!(inbox.claim(&parent.id, 8).await.unwrap().len(), 1);
        assert_eq!(launches.load(Ordering::SeqCst), 1);
    }

    async fn test_child_source(
        store: &bamboo_storage::SessionStoreV2,
        id: &str,
    ) -> Option<ChildCompletionSource> {
        ChildCompletionSource::from_committed_session(
            &store.load_session(id).await.unwrap().unwrap(),
        )
    }

    fn typed_test_report() -> serde_json::Value {
        serde_json::json!({"version":1,"outcome":"blocked","summary":"Quoted \"claim\" \\ 用户🪷\n",
            "reported_evidence":[{"description":"reported only","reference":"/never/read/reported-ref","sha256":null}],
            "reported_verification":[{"check":"fixture","reported_status":"passed","details":"a model claim"}],
            "proposals":["Try the next step"],"blockers":["Need approval"],"open_decisions":["Which route?"]})
    }

    fn typed_test_child(parent: &Session, id: &str, report: &str) -> Session {
        let mut child = Session::new_child(id, &parent.id, "model", "Child");
        let packet = bamboo_domain::ChildContextPacket {
            version: 1,
            objective: "bounded task".into(),
            constraints: vec![],
            acceptance: vec!["report only".into()],
            non_goals: vec![],
            necessary_user_instructions: vec![],
            recorded_decisions: vec![],
            source_user_message_ids: vec![parent.messages[0].id.clone()],
            background_message_ids: vec![],
        };
        let resolved = packet.resolve(parent, "bounded task").unwrap();
        let binding = bamboo_domain::ChildContextBinding::new(
            parent,
            id,
            resolved.required_brief.clone(),
            resolved,
        )
        .unwrap();
        binding.install(&mut child).unwrap();
        child.add_message(binding.assignment_message());
        child.add_message(Message::assistant(report, None));
        child.set_last_run_status("completed");
        ChildCompletionSource::prepare(&mut child, "sealed-report-run", &Default::default());
        child
    }

    fn typed_test_completion(child: &Session) -> ChildCompletion {
        ChildCompletion {
            parent_session_id: child.parent_session_id.clone().unwrap(),
            child_session_id: child.id.clone(),
            status: child.last_run_status().unwrap(),
            error: child.last_run_error(),
            completed_at: child.updated_at,
            source: ChildCompletionSource::from_committed_session(child),
        }
    }

    fn planner_test_child(parent: &Session, id: &str) -> Session {
        let mut report = typed_test_report();
        report["plan_markdown"] = serde_json::json!("# Plan\n## Selected\nPLANNER_SECTION_ONLY\n## Appendix\nLARGE_PLANNER_BODY_MUST_STAY_OUT_OF_RESUME");
        let mut child = typed_test_child(parent, id, &report.to_string());
        child.set_subagent_type("planner");
        child.metadata.insert(
            super::super::child_session::planner_artifact::CONTRACT_KEY.into(),
            "1".into(),
        );
        child
            .agent_runtime_state
            .get_or_insert_with(Default::default)
            .read_only = true;
        ChildCompletionSource::prepare(&mut child, "native-planner-report", &Default::default());
        child
    }

    #[cfg(unix)]
    #[tokio::test]
    async fn planner_artifact_completion_publishes_before_durable_resume_and_deduplicates() {
        let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
        save_waiting_test_parent(
            &store,
            "plan-parent",
            vec!["plan-child".into()],
            ChildWaitPolicy::All,
        )
        .await;
        let mut parent = store.load_session("plan-parent").await.unwrap().unwrap();
        parent.add_message(Message::user("Original selected source"));
        store.save_session(&parent).await.unwrap();
        let child = planner_test_child(&parent, "plan-child");
        store.save_session(&child).await.unwrap();
        let completion = typed_test_completion(&child);
        coordinator.on_child_completed(completion.clone()).await;
        coordinator.on_child_completed(completion).await;
        let claims = inbox.claim(&parent.id, 8).await.unwrap();
        assert_eq!(claims.len(), 1);
        assert_eq!(launches.load(Ordering::SeqCst), 1);
        let message = claims[0].envelope.to_provider_message().unwrap();
        assert!(message.content.contains("planner_artifact"));
        assert!(!message
            .content
            .contains("LARGE_PLANNER_BODY_MUST_STAY_OUT_OF_RESUME"));
        assert!(fits_child_result_budget(&claims[0].envelope));
        let raw = message
            .content
            .split_once("Child typed result:\n")
            .unwrap()
            .1
            .lines()
            .next()
            .unwrap();
        let projection: serde_json::Value = serde_json::from_str(raw).unwrap();
        assert_eq!(projection["execution_approved"], false);
        let path = store
            .bamboo_home_dir()
            .join(projection["file"].as_str().unwrap());
        assert!(std::fs::read_to_string(path)
            .unwrap()
            .contains("PLANNER_SECTION_ONLY"));
        let restored = store.load_session(&parent.id).await.unwrap().unwrap();
        assert_eq!(restored.kind, parent.kind);
        assert_eq!(restored.root_thinking_mode(), parent.root_thinking_mode());
        assert_eq!(
            restored
                .agent_runtime_state
                .as_ref()
                .unwrap()
                .permission_mode,
            parent.agent_runtime_state.as_ref().unwrap().permission_mode
        );
    }

    #[cfg(unix)]
    #[tokio::test]
    async fn planner_artifact_terminal_sibling_recovery_publishes_the_same_bounded_reference() {
        let (_temp, store, inbox, coordinator, _, _) = completion_inbox_fixture().await;
        save_waiting_test_parent(
            &store,
            "plan-recovery",
            vec!["plan-sibling".into(), "trigger-child".into()],
            ChildWaitPolicy::All,
        )
        .await;
        let mut parent = store.load_session("plan-recovery").await.unwrap().unwrap();
        parent.add_message(Message::user("Original selected source"));
        store.save_session(&parent).await.unwrap();
        let child = planner_test_child(&parent, "plan-sibling");
        store.save_session(&child).await.unwrap();
        let completion =
            save_completion_test_child(&store, &parent.id, "trigger-child", "completed").await;
        coordinator.on_child_completed(completion).await;
        let claims = inbox.claim(&parent.id, 8).await.unwrap();
        assert_eq!(claims.len(), 2, "both terminal outcomes remain durable");
        let message = claims
            .iter()
            .map(|c| c.envelope.to_provider_message().unwrap())
            .find(|m| m.content.contains("planner_artifact"))
            .unwrap();
        assert!(!message
            .content
            .contains("LARGE_PLANNER_BODY_MUST_STAY_OUT_OF_RESUME"));
        let raw = message
            .content
            .split_once("Child typed result:\n")
            .unwrap()
            .1
            .lines()
            .next()
            .unwrap();
        let projection: serde_json::Value = serde_json::from_str(raw).unwrap();
        assert!(store
            .bamboo_home_dir()
            .join(projection["file"].as_str().unwrap())
            .is_file());
    }

    #[tokio::test]
    async fn planner_artifact_real_boot_recovery_keeps_unconfirmed_wait_and_report_private() {
        use bamboo_domain::{ActorActivationClaim, ActorDirectoryPort};
        for mode in ["unknown", "expired", "unactivated-host"] {
            let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
            save_waiting_test_parent(
                &store,
                "lost-plan-parent",
                vec!["lost-plan-child".into()],
                ChildWaitPolicy::All,
            )
            .await;
            let mut parent = store
                .load_session("lost-plan-parent")
                .await
                .unwrap()
                .unwrap();
            parent.add_message(Message::user("Original selected source"));
            store.save_session(&parent).await.unwrap();
            if mode == "unactivated-host" {
                repaired_unactivated_test_child(&store, &parent, "lost-plan-child").await;
                coordinator.reconcile_orphans_at_boot().await;
                coordinator.reconcile_orphans_at_boot().await;
                assert!(inbox.claim(&parent.id, 8).await.unwrap().is_empty());
                assert_eq!(launches.load(Ordering::SeqCst), 0);
                assert!(read_runtime_state(
                    &store.load_session(&parent.id).await.unwrap().unwrap()
                )
                .waiting_for_children
                .is_some());
                continue;
            }
            let expired_actor = mode == "expired";
            let mut child = planner_test_child(&parent, "lost-plan-child");
            // Real first activation requires the creator's exact parent birth;
            // the report-only legacy helper intentionally does not supply it.
            child.parent_created_at = Some(parent.created_at);
            child.set_last_run_status("running");
            child
                .agent_runtime_state
                .get_or_insert_with(|| {
                    bamboo_domain::AgentRuntimeState::new("lost-worker-pipeline")
                })
                .status = bamboo_domain::AgentStatusState::Running;
            child.messages.last_mut().unwrap().content =
                "UNSEALED_REPORT_MUST_NOT_BE_IMPORTED".into();
            ChildCompletionSource::prepare(&mut child, "lost-worker", &Default::default());
            child.updated_at = Utc::now() - chrono::Duration::minutes(5);
            store.save_session(&child).await.unwrap();
            if expired_actor {
                let now = Utc::now();
                let activation = store
                    .claim_activation(&ActorActivationClaim {
                        actor_id: child.id.clone(),
                        run_id: "lost-worker".into(),
                        lease_owner: "synthetic-owner".into(),
                        lease_expires_at: now + chrono::Duration::seconds(2),
                        inbox_generation: 0,
                        placement_ref: None,
                        now,
                    })
                    .await
                    .unwrap();
                store
                    .start_activation(&activation.fence(), now)
                    .await
                    .unwrap();
                tokio::time::sleep(Duration::from_millis(2100)).await;
            }
            // Actual boot path twice: no Actor is unknown; an expired exact Actor
            // can close Child authority, without granting a parent resume source.
            coordinator.reconcile_orphans_at_boot().await;
            coordinator.reconcile_orphans_at_boot().await;
            let claims = inbox.claim(&parent.id, 8).await.unwrap();
            assert!(
                claims.is_empty(),
                "source-less recovery is not parent delivery permission"
            );
            assert_eq!(launches.load(Ordering::SeqCst), 0);
            let waiting = store.load_session(&parent.id).await.unwrap().unwrap();
            assert!(waiting
                .agent_runtime_state
                .as_ref()
                .unwrap()
                .waiting_for_children
                .is_some());
            let child = store.load_session(&child.id).await.unwrap().unwrap();
            assert_eq!(
                child.last_run_status().as_deref(),
                Some(if expired_actor { "error" } else { "running" })
            );
            assert_eq!(
                child.agent_runtime_state.as_ref().unwrap().status,
                if expired_actor {
                    bamboo_domain::AgentStatusState::Failed
                } else {
                    bamboo_domain::AgentStatusState::Running
                }
            );
            assert!(ChildCompletionSource::from_committed_session(&child).is_none());
            assert!(!store.bamboo_home_dir().join("plan/delegated").exists());
        }
    }

    fn unusable_typed_test_child(
        parent: &Session,
        id: &str,
        case: &str,
    ) -> (Session, &'static str) {
        let mut child = typed_test_child(parent, id, &typed_test_report().to_string());
        let mut prior = Default::default();
        let reason = match case {
            "prose" => {
                child.messages.last_mut().unwrap().content = "UNUSABLE_RAW_PROSE".repeat(64);
                "report_malformed"
            }
            "malformed" => {
                child.messages.last_mut().unwrap().content =
                    "{\"version\":1,UNUSABLE_RAW_JSON".into();
                "report_malformed"
            }
            "oversized" => {
                child
                    .messages
                    .last_mut()
                    .unwrap()
                    .content
                    .push_str(&" ".repeat(8193));
                "result_budget_exceeded"
            }
            "multipart" => {
                child.messages.last_mut().unwrap().content_parts = Some(
                    serde_json::from_value(
                        serde_json::json!([{"type":"text","text":"UNUSABLE_RAW_PART"}]),
                    )
                    .unwrap(),
                );
                "report_not_current_final"
            }
            "commentary" => {
                child.messages.last_mut().unwrap().phase =
                    Some(bamboo_domain::MessagePhase::Commentary);
                "report_not_current_final"
            }
            "blank" => {
                child.messages.last_mut().unwrap().content = "  ".into();
                "report_not_current_final"
            }
            "old_final" => {
                prior = child.messages.iter().map(|m| m.id.clone()).collect();
                "report_not_current_final"
            }
            "error" => {
                prior = child.messages.iter().map(|m| m.id.clone()).collect();
                child.set_last_run_status("error");
                child.set_last_run_error("UNUSABLE_RAW_ERROR\"\\\n".repeat(4096));
                "report_not_current_final"
            }
            "absent" => {
                child.messages.pop();
                "report_absent"
            }
            _ => panic!("unknown unusable report case"),
        };
        ChildCompletionSource::prepare(&mut child, "unusable-report-run", &prior);
        (child, reason)
    }

    #[test]
    fn sealed_unusable_reports_keep_current_source_boundary_and_escape_budget() {
        let mut parent = Session::new("unavailable-parent", "model");
        parent.add_message(Message::user("required source"));
        for case in [
            "prose",
            "malformed",
            "oversized",
            "multipart",
            "commentary",
            "blank",
            "old_final",
            "error",
            "absent",
        ] {
            let (child, reason) = unusable_typed_test_child(&parent, "unavailable-child", case);
            let completion = typed_test_completion(&child);
            assert!(completion.source.is_some(), "valid seal: {case}");
            let mut projection = sealed_child_report(&completion, &child, &parent).unwrap();
            assert_eq!(
                projection,
                unavailable_child_result("typed_result", reason),
                "{case}"
            );
            let mut message = typed_child_resume_message(&completion, 0, &projection);
            let raw = completion.source.as_ref().unwrap().result(&child, &parent);
            let mut envelope =
                child_completion_envelope(&completion, Utc::now(), raw.clone(), &message);
            assert!(bound_typed_child_delivery(
                &mut envelope,
                &mut message,
                &mut projection,
                &completion,
                0
            ));
            assert!(fits_child_result_budget(&envelope));
            assert!(fits_child_result_budget(&message));
            assert!(fits_child_result_budget(
                &envelope.to_provider_message().unwrap()
            ));
            assert!(message.content.contains("SubAgent"));
            assert_eq!(
                message.metadata.as_ref().unwrap()["child_final_response_included"],
                false
            );
            assert!(projection.get("child_report").is_none());
            assert!(projection.get("host_observation").is_none());
            let SessionMessageBody::ChildOutcome(outcome) = envelope.body else {
                panic!("child outcome")
            };
            assert!(outcome.result.is_none());
            assert!(outcome.error.is_none());
            if let Some(raw) = raw {
                assert!(!message.content.contains(&raw), "no raw report: {case}");
            }
            // A replaced or invalid source never turns into an unavailable
            // completion that could clear a newer wait.
            let mut stale = child.clone();
            stale
                .messages
                .last_mut()
                .unwrap()
                .content
                .push_str("changed");
            assert!(sealed_child_report(&completion, &stale, &parent).is_none());
        }
    }

    #[test]
    fn unavailable_delivery_counts_escaped_wrappers_at_exact_limit() {
        let mut parent = Session::new("unavailable-parent", "model");
        parent.add_message(Message::user("required source"));
        let (child, _) = unusable_typed_test_child(&parent, "unavailable-child", "oversized");
        let completion = typed_test_completion(&child);
        let mut projection = sealed_child_report(&completion, &child, &parent).unwrap();
        let mut message = typed_child_resume_message(&completion, 0, &projection);
        let mut base = child_completion_envelope(&completion, Utc::now(), None, &message);
        assert!(bound_typed_child_delivery(
            &mut base,
            &mut message,
            &mut projection,
            &completion,
            0
        ));
        let bytes = serde_json::to_vec(&base).unwrap().len().max(
            serde_json::to_vec(&base.to_provider_message().unwrap())
                .unwrap()
                .len(),
        );
        for extra in [8192 - bytes, 8193 - bytes] {
            let mut envelope = base.clone();
            // Quotes count twice in the actual escaped wrappers.
            let padding = format!("{}{}", "\"".repeat(extra / 2), "x".repeat(extra % 2));
            envelope.correlation_id.as_mut().unwrap().push_str(&padding);
            let mut projection = projection.clone();
            let mut message = message.clone();
            assert_eq!(
                bound_typed_child_delivery(
                    &mut envelope,
                    &mut message,
                    &mut projection,
                    &completion,
                    0
                ),
                bytes + extra == 8192
            );
            assert_eq!(
                serde_json::to_vec(&envelope).unwrap().len().max(
                    serde_json::to_vec(&envelope.to_provider_message().unwrap())
                        .unwrap()
                        .len()
                ),
                bytes + extra
            );
            assert!(fits_child_result_budget(&message));
        }
    }

    #[test]
    fn sealed_typed_projection_keeps_host_source_distinct_from_all_reported_claims() {
        let mut parent = Session::new("typed-parent", "model");
        parent.add_message(Message::user("required user constraint"));
        let report = typed_test_report();
        let child = typed_test_child(&parent, "typed-child", &report.to_string());
        let completion = typed_test_completion(&child);
        let projection = sealed_child_report(&completion, &child, &parent).unwrap();
        assert_eq!(projection["child_report"], report);
        assert_eq!(
            projection["host_observation"]["terminal_source"],
            serde_json::to_value(&completion.source).unwrap()
        );
        assert_eq!(
            projection["host_observation"]["kind"],
            "committed_terminal_source"
        );
        assert_eq!(
            projection["host_observation"]["last_run_status"],
            "completed"
        );
        assert_eq!(projection["child_report"]["outcome"], "blocked");
        assert_eq!(
            projection["child_report"]["reported_verification"][0]["reported_status"],
            "passed"
        );
        assert!(projection["host_observation"].get("verified").is_none());
        for excluded in ["guardian", "resident", "legacy", "missing_source"] {
            let mut excluded_child = child.clone();
            match excluded {
                "guardian" => {
                    excluded_child
                        .metadata
                        .insert("subagent_type".into(), "guardian".into());
                }
                "resident" => {
                    excluded_child
                        .metadata
                        .insert("lifecycle".into(), "resident".into());
                }
                "legacy" => {
                    excluded_child =
                        Session::new_child("typed-child", &parent.id, "model", "Child");
                    excluded_child.add_message(Message::assistant(report.to_string(), None));
                    excluded_child.set_last_run_status("completed");
                }
                _ => {}
            }
            ChildCompletionSource::prepare(
                &mut excluded_child,
                "excluded-run",
                &Default::default(),
            );
            let mut excluded_completion = typed_test_completion(&excluded_child);
            if excluded == "missing_source" {
                excluded_completion.source = None;
            }
            assert!(
                sealed_child_report(&excluded_completion, &excluded_child, &parent).is_none(),
                "{excluded}"
            );
        }
    }

    #[tokio::test]
    async fn typed_delivery_counts_complete_escaped_envelope_and_final_correlation() {
        let mut parent = Session::new("typed-parent", "model");
        parent.add_message(Message::user("required source"));
        let child = typed_test_child(&parent, "typed-child", &typed_test_report().to_string());
        let completion = typed_test_completion(&child);
        let original = sealed_child_report(&completion, &child, &parent).unwrap();
        let presentation = typed_child_resume_message(&completion, 0, &original);
        let mut base = child_completion_envelope(
            &completion,
            Utc::now(),
            completion.source.as_ref().unwrap().result(&child, &parent),
            &presentation,
        );
        let mut projection = original.clone();
        let mut message = presentation.clone();
        assert!(bound_typed_child_delivery(
            &mut base,
            &mut message,
            &mut projection,
            &completion,
            0
        ));
        assert_eq!(projection, original);
        let temp = tempfile::tempdir().unwrap();
        let store = Arc::new(
            bamboo_storage::SessionStoreV2::new(temp.path().to_path_buf())
                .await
                .unwrap(),
        );
        store.save_session(&parent).await.unwrap();
        let inbox = bamboo_storage::FileSessionInbox::new(store, Default::default());
        let first = inbox.deliver(&base).await.unwrap();
        let stable_id = base.id.clone();
        let base_bytes = serde_json::to_vec(&base.to_provider_message().unwrap())
            .unwrap()
            .len();
        assert!(base_bytes < 8192);
        let correlation = base.correlation_id.as_ref().unwrap().clone();
        for (extra, available) in [(8192 - base_bytes, true), (8193 - base_bytes, false)] {
            let mut envelope = base.clone();
            envelope.correlation_id = Some(format!("{correlation}{}", "x".repeat(extra)));
            assert_eq!(
                serde_json::to_vec(&envelope.to_provider_message().unwrap())
                    .unwrap()
                    .len(),
                base_bytes + extra
            );
            let mut message = presentation.clone();
            let mut projection = original.clone();
            assert!(bound_typed_child_delivery(
                &mut envelope,
                &mut message,
                &mut projection,
                &completion,
                0
            ));
            assert_eq!(projection["available"], available);
            assert!(fits_child_result_budget(&envelope));
            assert!(fits_child_result_budget(&message));
            assert!(fits_child_result_budget(
                &envelope.to_provider_message().unwrap()
            ));
            assert_eq!(
                envelope.idempotency_semantics(),
                base.idempotency_semantics()
            );
            assert_eq!(envelope.id, stable_id);
            assert_eq!(
                inbox.deliver(&envelope).await.unwrap(),
                first,
                "presentation overflow cannot change delivery semantics"
            );
            if !available {
                assert_eq!(projection["reason"], "result_budget_exceeded");
                assert!(projection.get("child_report").is_none());
                assert!(message.content.contains("SubAgent"));
                assert!(!message.content.contains("Bounded tail"));
            }
        }
        let mut large = typed_test_report();
        large["summary"] = serde_json::json!("\"\\\n".repeat(500));
        large["proposals"] = serde_json::json!(["\"\\\n".repeat(250)]);
        let child = typed_test_child(&parent, "escaped-child", &large.to_string());
        assert!(decode_report(&large.to_string()).is_ok());
        let completion = typed_test_completion(&child);
        let mut projection = sealed_child_report(&completion, &child, &parent).unwrap();
        let mut message = typed_child_resume_message(&completion, 0, &projection);
        let mut envelope =
            child_completion_envelope(&completion, Utc::now(), Some(large.to_string()), &message);
        assert!(serde_json::to_vec(&envelope).unwrap().len() > 8192);
        assert!(bound_typed_child_delivery(
            &mut envelope,
            &mut message,
            &mut projection,
            &completion,
            0
        ));
        assert_eq!(projection["available"], false);
        assert!(fits_child_result_budget(&envelope));
        assert!(fits_child_result_budget(&message));
        assert!(fits_child_result_budget(
            &envelope.to_provider_message().unwrap()
        ));
    }

    #[tokio::test]
    async fn typed_source_mutations_reject_before_admission_and_wait_clear() {
        for mutation in ["birth", "assignment", "input", "result", "parent_source"] {
            let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
            save_waiting_test_parent(
                &store,
                "typed-parent",
                vec!["typed-child".into()],
                ChildWaitPolicy::All,
            )
            .await;
            let mut parent = store.load_session("typed-parent").await.unwrap().unwrap();
            parent.add_message(Message::user("required source"));
            store.save_session(&parent).await.unwrap();
            let mut child =
                typed_test_child(&parent, "typed-child", &typed_test_report().to_string());
            store.save_session(&child).await.unwrap();
            let mut completion = typed_test_completion(&child);
            match mutation {
                "birth" => {
                    completion.source.as_mut().unwrap().child_created_at +=
                        chrono::Duration::seconds(1)
                }
                "assignment" => {
                    child.messages[0].content.push_str("changed assignment");
                }
                "input" => child.add_message(Message::user("new user input")),
                "result" => child
                    .messages
                    .last_mut()
                    .unwrap()
                    .content
                    .push_str("changed"),
                _ => {
                    parent.messages[0].content.push_str("changed");
                    store.save_session(&parent).await.unwrap();
                }
            }
            store.save_session(&child).await.unwrap();
            coordinator.on_child_completed(completion).await;
            assert_eq!(
                inbox.inspect(&parent.id).await.unwrap().pending,
                0,
                "{mutation}"
            );
            assert_eq!(launches.load(Ordering::SeqCst), 0, "{mutation}");
            assert!(
                read_runtime_state(&store.load_session(&parent.id).await.unwrap().unwrap())
                    .waiting_for_children
                    .is_some(),
                "{mutation}"
            );
        }
    }

    #[tokio::test]
    async fn typed_cold_watchdog_and_terminal_siblings_admit_each_source_once() {
        for case in [
            "valid",
            "prose",
            "malformed",
            "oversized",
            "multipart",
            "commentary",
            "blank",
            "old_final",
            "error",
            "absent",
        ] {
            for cold_watchdog in [false, true] {
                let (_temp, store, inbox, coordinator, reservations, launches) =
                    completion_inbox_fixture().await;
                let ids = vec!["typed-a".into(), "typed-b".into()];
                save_waiting_test_parent(&store, "typed-parent", ids.clone(), ChildWaitPolicy::All)
                    .await;
                let mut parent = store.load_session("typed-parent").await.unwrap().unwrap();
                parent.add_message(Message::user("required source"));
                store.save_session(&parent).await.unwrap();
                let mut completions = Vec::new();
                for id in &ids {
                    let child = if case == "valid" {
                        typed_test_child(&parent, id, &typed_test_report().to_string())
                    } else {
                        unusable_typed_test_child(&parent, id, case).0
                    };
                    store.save_session(&child).await.unwrap();
                    completions.push(typed_test_completion(&child));
                }
                if cold_watchdog {
                    // No callbacks or cached Session snapshots precede this disk-only sweep.
                    let mut wait = read_runtime_state(&parent).waiting_for_children.unwrap();
                    wait.registered_at = Utc::now() - chrono::Duration::seconds(61);
                    coordinator.sweep_child_wait(&parent.id, wait.clone()).await;
                    coordinator.sweep_child_wait(&parent.id, wait).await;
                } else {
                    // The terminal index is ahead of the second sibling's callback.
                    coordinator.on_child_completed(completions[0].clone()).await;
                }
                for completion in completions {
                    coordinator.on_child_completed(completion).await;
                }
                let claims = inbox.claim(&parent.id, 8).await.unwrap();
                assert_eq!(claims.len(), 2);
                assert_eq!(reservations.load(Ordering::SeqCst), 1);
                assert_eq!(launches.load(Ordering::SeqCst), 1);
                for claim in claims {
                    assert!(fits_child_result_budget(&claim.envelope));
                    assert!(fits_child_result_budget(
                        &claim.envelope.to_provider_message().unwrap()
                    ));
                    let actual = claim.envelope.to_provider_message().unwrap();
                    assert!(!actual.content.contains("UNUSABLE_RAW"));
                    let SessionMessageBody::ChildOutcome(outcome) = claim.envelope.body else {
                        panic!("child outcome");
                    };
                    let projection: serde_json::Value = serde_json::from_str(
                        outcome
                            .provider_message
                            .as_ref()
                            .unwrap()
                            .content
                            .text
                            .split_once("Child typed result:\n")
                            .unwrap()
                            .1
                            .lines()
                            .next()
                            .unwrap(),
                    )
                    .unwrap();
                    if case != "valid" {
                        let expected =
                            unusable_typed_test_child(&parent, &outcome.child_session_id, case).1;
                        assert_eq!(
                            projection,
                            unavailable_child_result("typed_result", expected)
                        );
                        assert!(outcome.result.is_none());
                        assert!(outcome.error.is_none());
                        assert!(fits_child_result_budget(&outcome.provider_message));
                        continue;
                    }
                    assert_eq!(projection["child_report"], typed_test_report());
                    assert_eq!(
                        projection["host_observation"]["terminal_source"]["child_session_id"],
                        outcome.child_session_id
                    );
                    assert_eq!(
                        projection["host_observation"]["last_run_status"],
                        "completed"
                    );
                    assert!(fits_child_result_budget(&outcome.provider_message));
                }
            }
        }
    }

    async fn completion_inbox_fixture() -> (
        tempfile::TempDir,
        Arc<bamboo_storage::SessionStoreV2>,
        Arc<dyn SessionInboxPort>,
        Arc<ChildCompletionCoordinator>,
        Arc<AtomicUsize>,
        Arc<AtomicUsize>,
    ) {
        completion_inbox_fixture_with_limits(bamboo_domain::SessionInboxLimits::default()).await
    }

    async fn completion_inbox_fixture_with_limits(
        limits: bamboo_domain::SessionInboxLimits,
    ) -> (
        tempfile::TempDir,
        Arc<bamboo_storage::SessionStoreV2>,
        Arc<dyn SessionInboxPort>,
        Arc<ChildCompletionCoordinator>,
        Arc<AtomicUsize>,
        Arc<AtomicUsize>,
    ) {
        let temp = tempfile::tempdir().unwrap();
        let store = Arc::new(
            bamboo_storage::SessionStoreV2::new(temp.path().to_path_buf())
                .await
                .unwrap(),
        );
        let storage: Arc<dyn Storage> = store.clone();
        let locked = Arc::new(LockedSessionStore::new(storage.clone()));
        let inbox: Arc<dyn SessionInboxPort> =
            Arc::new(bamboo_storage::FileSessionInbox::new(store.clone(), limits));
        let router = crate::SessionActivationRouter::new();
        let messenger = Arc::new(crate::SessionMessenger::new(
            storage.clone(),
            inbox.clone(),
            router.clone(),
        ));
        let reservations = Arc::new(AtomicUsize::new(0));
        let launches = Arc::new(AtomicUsize::new(0));
        router
            .set_spawner(Arc::new(CountingActivationSpawner {
                reservations: reservations.clone(),
                launches: launches.clone(),
            }))
            .await;
        let provider: Arc<dyn bamboo_llm::LLMProvider> = Arc::new(CompletedTestProvider);
        let config = Arc::new(RwLock::new(Config::default()));
        let metrics = bamboo_metrics::MetricsCollector::spawn(
            Arc::new(bamboo_metrics::SqliteMetricsStorage::new(
                temp.path().join("metrics.db"),
            )),
            7,
        );
        let tools: Arc<dyn ToolExecutor> = Arc::new(EmptyTools);
        let agent = Arc::new(
            Agent::builder()
                .storage(storage.clone())
                .persistence(locked.clone())
                .session_inbox(inbox.clone())
                .activation_router(router)
                .session_messenger(messenger)
                .attachment_reader(store.clone())
                .skill_manager(Arc::new(bamboo_skills::SkillManager::new()))
                .metrics_collector(metrics)
                .config(config.clone())
                .provider(provider.clone())
                .default_tools(tools)
                .build()
                .unwrap(),
        );
        let mut providers = HashMap::new();
        providers.insert("test".to_string(), provider);
        let registry = Arc::new(ProviderRegistry::new(providers, "test".to_string()));
        let provider_router = Arc::new(ProviderModelRouter::new(registry.clone()));
        let coordinator = Arc::new(ChildCompletionCoordinator::new(
            storage,
            locked,
            Arc::default(),
            Arc::new(RwLock::new(HashMap::new())),
            Arc::new(RwLock::new(HashMap::new())),
            agent,
            config,
            registry,
            provider_router,
            temp.path().to_path_buf(),
            None,
        ));
        (temp, store, inbox, coordinator, reservations, launches)
    }

    async fn save_waiting_test_parent(
        store: &bamboo_storage::SessionStoreV2,
        parent_id: &str,
        ids: Vec<String>,
        policy: ChildWaitPolicy,
    ) {
        let mut parent = Session::new(parent_id, "model");
        let mut runtime = AgentRuntimeState::new("waiting-run");
        runtime.status = AgentStatusState::Suspended;
        runtime.waiting_for_children = Some(WaitingForChildrenState::for_children(
            ids,
            policy,
            Utc::now(),
        ));
        runtime.suspension = Some(SuspensionState {
            reason: "waiting_for_children".into(),
            suspended_at: Utc::now(),
            resumable: true,
            hook_point: Some("ChildCompletion".into()),
        });
        write_runtime_state(&mut parent, &runtime);
        parent.metadata.insert(
            "runtime.suspend_reason".into(),
            "waiting_for_children".into(),
        );
        store.save_session(&parent).await.unwrap();
    }

    async fn save_completion_test_child(
        store: &bamboo_storage::SessionStoreV2,
        parent_id: &str,
        child_id: &str,
        status: &str,
    ) -> ChildCompletion {
        let mut child = store
            .load_session(child_id)
            .await
            .unwrap()
            .unwrap_or_else(|| Session::new_child(child_id, parent_id, "model", "Child"));
        child.add_message(Message::assistant(format!("result-{child_id}"), None));
        child.set_last_run_status(status);
        save_sourced_test_child(store, &child).await;
        ChildCompletion {
            parent_session_id: parent_id.into(),
            child_session_id: child_id.into(),
            status: status.into(),
            error: None,
            completed_at: child.updated_at,
            source: test_child_source(store, child_id).await,
        }
    }

    #[tokio::test]
    async fn unrelated_invalid_pending_body_does_not_block_incremental_child_activation() {
        for queue in ["new", "cur"] {
            let (_temp, store, inbox, coordinator, reservations, launches) =
                completion_inbox_fixture().await;
            let parent_id = "invalid-pending-parent";
            save_waiting_test_parent(
                &store,
                parent_id,
                vec!["invalid-pending-fast".into(), "invalid-pending-slow".into()],
                ChildWaitPolicy::All,
            )
            .await;
            let before = read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap());
            let fast =
                save_completion_test_child(&store, parent_id, "invalid-pending-fast", "completed")
                    .await;
            save_completion_test_child(&store, parent_id, "invalid-pending-slow", "running").await;

            let mut unrelated = SessionMessageEnvelope::user_input(parent_id, "unrelated bash");
            unrelated.source = SessionMessageSource::Runtime {
                subsystem: "bash".into(),
            };
            unrelated.kind = SessionMessageKind::RuntimeInstruction;
            unrelated.body =
                SessionMessageBody::RuntimeInstruction(bamboo_domain::SessionRuntimeInstruction {
                    instruction: "bash_completion".into(),
                    content: Some(bamboo_domain::SessionMessageContent::text("unrelated bash")),
                    data: None,
                    provider_message: None,
                });
            inbox.deliver(&unrelated).await.unwrap();
            let dir = store
                .bamboo_home_dir()
                .join(store.resolve_rel_path(parent_id).await.unwrap())
                .join("inbox");
            let staged = tokio::fs::read_dir(dir.join("new"))
                .await
                .unwrap()
                .next_entry()
                .await
                .unwrap()
                .unwrap()
                .path();
            let mut transport: serde_json::Value =
                serde_json::from_slice(&tokio::fs::read(&staged).await.unwrap()).unwrap();
            transport["body"] = serde_json::json!({"not": "an envelope"});
            let invalid_bytes = serde_json::to_vec(&transport).unwrap();
            tokio::fs::write(&staged, &invalid_bytes).await.unwrap();
            let invalid_path = if queue == "cur" {
                let destination = dir.join("cur").join(staged.file_name().unwrap());
                tokio::fs::create_dir_all(dir.join("cur")).await.unwrap();
                tokio::fs::rename(&staged, &destination).await.unwrap();
                destination
            } else {
                staged
            };

            coordinator.on_child_completed(fast.clone()).await;
            let outcomes = inbox
                .pending_child_outcomes(parent_id, "invalid-pending-fast")
                .await
                .unwrap();
            assert_eq!(outcomes.len(), 1);
            assert!(fits_child_result_budget(&outcomes[0]));
            let after = read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap());
            assert_eq!(after, before, "All wait and suspension remain armed");
            assert_eq!(reservations.load(Ordering::SeqCst), 1);
            assert_eq!(launches.load(Ordering::SeqCst), 1);
            let backlog = inbox.inspect(parent_id).await.unwrap();
            assert!(backlog.interrupt_pending());
            assert_eq!(backlog.coordinator_generation, 0);
            let ready = inbox
                .inspect_wake_readiness(parent_id, Utc::now())
                .await
                .unwrap()
                .ready
                .unwrap();
            assert_eq!(ready.id, outcomes[0].id);
            assert_eq!(
                ready.activation_policy,
                SessionActivationPolicy::InterruptSpecificWait
            );
            assert_eq!(tokio::fs::read(&invalid_path).await.unwrap(), invalid_bytes);
            assert!(!inbox.was_admitted(parent_id, &unrelated.id).await.unwrap());

            let claims = inbox.claim(parent_id, 8).await.unwrap();
            assert_eq!(claims.len(), 1);
            assert_eq!(claims[0].envelope.id, outcomes[0].id);
            inbox.ack(parent_id, &claims[0]).await.unwrap();
            coordinator.on_child_completed(fast).await;
            assert_eq!(launches.load(Ordering::SeqCst), 1);
            assert!(!inbox.was_admitted(parent_id, &unrelated.id).await.unwrap());
        }
    }

    #[tokio::test]
    async fn partial_all_outcome_preserves_wait_and_active_parent_control_state() {
        let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
        let parent_id = "incremental-parent";
        save_waiting_test_parent(
            &store,
            parent_id,
            vec!["ready-child".into(), "slow-child".into()],
            ChildWaitPolicy::All,
        )
        .await;
        let mut parent = store.load_session(parent_id).await.unwrap().unwrap();
        let mut runtime = read_runtime_state(&parent);
        runtime.status = AgentStatusState::Running;
        runtime.suspension = None;
        parent.metadata.remove("runtime.suspend_reason");
        write_runtime_state(&mut parent, &runtime);
        store.save_session(&parent).await.unwrap();
        let ready = save_completion_test_child(&store, parent_id, "ready-child", "completed").await;
        save_completion_test_child(&store, parent_id, "slow-child", "running").await;

        coordinator.on_child_completed(ready.clone()).await;
        let after = store.load_session(parent_id).await.unwrap().unwrap();
        assert_eq!(read_runtime_state(&after), runtime);
        assert_eq!(after.updated_at, parent.updated_at);
        assert_eq!(launches.load(Ordering::SeqCst), 1);
        let backlog = inbox.inspect(parent_id).await.unwrap();
        assert!(backlog.interrupt_pending());
        assert_eq!(backlog.coordinator_generation, 0, "no blanket prefix grant");
        let claims = inbox.claim(parent_id, 8).await.unwrap();
        assert_eq!(claims.len(), 1);
        inbox.ack(parent_id, &claims[0]).await.unwrap();

        coordinator.on_child_completed(ready).await;
        assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 0);
        let slow = save_completion_test_child(&store, parent_id, "slow-child", "completed").await;
        coordinator.on_child_completed(slow).await;
        assert!(
            read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                .waiting_for_children
                .is_none()
        );
        let claims = inbox.claim(parent_id, 8).await.unwrap();
        assert_eq!(
            claims.len(),
            1,
            "the acknowledged first result is not repeated at wait release"
        );
        let SessionMessageBody::ChildOutcome(outcome) = &claims[0].envelope.body else {
            panic!("child outcome");
        };
        assert_eq!(outcome.child_session_id, "slow-child");
    }

    #[tokio::test]
    async fn incremental_completions_during_owner_exit_share_one_successor() {
        let (_temp, store, inbox, coordinator, reservations, launches) =
            completion_inbox_fixture().await;
        let parent_id = "handoff-parent";
        save_waiting_test_parent(
            &store,
            parent_id,
            vec![
                "handoff-a".into(),
                "handoff-b".into(),
                "handoff-slow".into(),
            ],
            ChildWaitPolicy::All,
        )
        .await;
        let a = save_completion_test_child(&store, parent_id, "handoff-a", "completed").await;
        let b = save_completion_test_child(&store, parent_id, "handoff-b", "completed").await;
        save_completion_test_child(&store, parent_id, "handoff-slow", "running").await;
        let router = coordinator.agent.activation_router().unwrap();
        router.set_inbox(inbox.clone());
        let mut owner = router
            .register_run(parent_id, "old-parent-run")
            .await
            .unwrap();

        coordinator.on_child_completed(a.clone()).await;
        assert!(inbox
            .claim_for_turn(parent_id, 8, Some("old-parent-run"))
            .await
            .unwrap()
            .is_empty());
        owner.begin_finalization().await;
        coordinator.on_child_completed(b.clone()).await;
        assert_eq!(reservations.load(Ordering::SeqCst), 0);
        assert_eq!(launches.load(Ordering::SeqCst), 0);
        owner.finish(0).await.unwrap();
        assert_eq!(reservations.load(Ordering::SeqCst), 1);
        assert_eq!(launches.load(Ordering::SeqCst), 1);
        let successor = router.current_run_id(parent_id).await.unwrap();
        let claims = inbox
            .claim_for_turn(parent_id, 8, Some(&successor))
            .await
            .unwrap();
        assert_eq!(claims.len(), 2);
        for claim in &claims {
            inbox.ack(parent_id, claim).await.unwrap();
        }
        tokio::join!(
            coordinator.on_child_completed(a),
            coordinator.on_child_completed(b)
        );
        assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 0);
        assert_eq!(reservations.load(Ordering::SeqCst), 1);
        assert!(
            read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                .waiting_for_children
                .is_some()
        );
    }

    #[tokio::test]
    async fn cold_watchdog_replays_each_ready_child_before_all_wait_is_satisfied() {
        let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
        let parent_id = "cold-incremental-parent";
        save_waiting_test_parent(
            &store,
            parent_id,
            vec!["cold-a".into(), "cold-b".into(), "cold-slow".into()],
            ChildWaitPolicy::All,
        )
        .await;
        let mut parent = store.load_session(parent_id).await.unwrap().unwrap();
        let mut runtime = read_runtime_state(&parent);
        runtime.waiting_for_children.as_mut().unwrap().registered_at -=
            chrono::Duration::seconds(61);
        write_runtime_state(&mut parent, &runtime);
        store.save_session(&parent).await.unwrap();
        save_completion_test_child(&store, parent_id, "cold-a", "completed").await;
        save_completion_test_child(&store, parent_id, "cold-b", "completed").await;
        save_completion_test_child(&store, parent_id, "cold-slow", "running").await;

        let wait = runtime.waiting_for_children.unwrap();
        coordinator.sweep_child_wait(parent_id, wait.clone()).await;
        let reopened = bamboo_storage::FileSessionInbox::new(store.clone(), Default::default());
        let backlog = reopened.inspect(parent_id).await.unwrap();
        assert_eq!(backlog.pending, 2);
        assert!(backlog.interrupt_pending());
        assert_eq!(backlog.coordinator_generation, 0);
        assert_eq!(launches.load(Ordering::SeqCst), 1);
        let claims = reopened.claim(parent_id, 8).await.unwrap();
        assert_eq!(claims.len(), 2);
        for claim in &claims {
            reopened.ack(parent_id, claim).await.unwrap();
        }
        coordinator.sweep_child_wait(parent_id, wait).await;
        assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 0);
        assert!(
            read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                .waiting_for_children
                .is_some()
        );
    }

    #[tokio::test]
    async fn any_wait_delivers_a_late_direct_child_without_repeating_the_first_outcome() {
        let (_temp, store, inbox, coordinator, _, _) = completion_inbox_fixture().await;
        let parent_id = "any-incremental-parent";
        save_waiting_test_parent(
            &store,
            parent_id,
            vec!["any-first".into(), "any-late".into()],
            ChildWaitPolicy::Any,
        )
        .await;
        let first = save_completion_test_child(&store, parent_id, "any-first", "completed").await;
        save_completion_test_child(&store, parent_id, "any-late", "running").await;
        coordinator.on_child_completed(first.clone()).await;
        assert!(
            read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                .waiting_for_children
                .is_none()
        );
        let first_claims = inbox.claim(parent_id, 8).await.unwrap();
        assert_eq!(first_claims.len(), 1);
        inbox.ack(parent_id, &first_claims[0]).await.unwrap();

        let late = save_completion_test_child(&store, parent_id, "any-late", "completed").await;
        coordinator.on_child_completed(late.clone()).await;
        coordinator.on_child_completed(first).await;
        coordinator.on_child_completed(late).await;
        let claims = inbox.claim(parent_id, 8).await.unwrap();
        assert_eq!(claims.len(), 1);
        let SessionMessageBody::ChildOutcome(outcome) = &claims[0].envelope.body else {
            panic!("child outcome");
        };
        assert_eq!(outcome.child_session_id, "any-late");
        assert!(
            read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                .waiting_for_children
                .is_none()
        );
    }

    #[tokio::test]
    async fn terminal_batch_upgrades_lower_pending_outcome_below_an_acknowledged_sibling() {
        let (_temp, store, inbox, coordinator, _, _) = completion_inbox_fixture().await;
        let parent_id = "terminal-batch-parent";
        save_waiting_test_parent(
            &store,
            parent_id,
            vec!["batch-low".into(), "batch-high".into()],
            ChildWaitPolicy::All,
        )
        .await;
        let mut parent = store.load_session(parent_id).await.unwrap().unwrap();
        let wait = read_runtime_state(&parent).waiting_for_children.unwrap();
        let low = save_completion_test_child(&store, parent_id, "batch-low", "completed").await;
        let high = save_completion_test_child(&store, parent_id, "batch-high", "completed").await;
        let mut receipts = Vec::new();
        for completion in [&low, &high] {
            let result = format!("result-{}", completion.child_session_id);
            let presentation = runtime_resume_message(completion, 0, Some(&result));
            let envelope = child_completion_envelope(
                completion,
                wait.registered_at,
                Some(result),
                &presentation,
            );
            receipts.push(inbox.deliver(&envelope).await.unwrap());
        }
        inbox
            .mark_message_activation_eligible(
                parent_id,
                &receipts[1],
                SessionActivationPolicy::InterruptSpecificWait,
            )
            .await
            .unwrap();
        let claims = inbox.claim(parent_id, 8).await.unwrap();
        assert_eq!(claims.len(), 1);
        assert_eq!(claims[0].generation, receipts[1].generation);
        parent.add_message(claims[0].envelope.to_provider_message().unwrap());
        store.save_session(&parent).await.unwrap();
        inbox.ack(parent_id, &claims[0]).await.unwrap();

        coordinator.on_child_completed(low).await;
        let mut rearmed = store.load_session(parent_id).await.unwrap().unwrap();
        let mut runtime = read_runtime_state(&rearmed);
        assert!(runtime.waiting_for_children.is_none());
        let new_wait = WaitingForChildrenState::for_children(
            vec!["next-slow-child".into()],
            ChildWaitPolicy::All,
            Utc::now(),
        );
        runtime.waiting_for_children = Some(new_wait.clone());
        runtime.status = AgentStatusState::Suspended;
        write_runtime_state(&mut rearmed, &runtime);
        store.save_session(&rearmed).await.unwrap();
        let backlog = inbox.inspect(parent_id).await.unwrap();
        assert_eq!(backlog.pending, 1);
        assert!(
            backlog.interrupt_pending(),
            "the low generation has its own grant"
        );
        let (prepared, ready) = prepare_session_inbox_activation(
            &coordinator.persistence,
            parent_id,
            backlog.interrupt_pending(),
        )
        .await
        .unwrap()
        .unwrap();
        assert!(ready);
        assert_eq!(
            read_runtime_state(&prepared).waiting_for_children,
            Some(new_wait)
        );
        let claims = inbox.claim(parent_id, 8).await.unwrap();
        assert_eq!(claims.len(), 1);
        assert_eq!(claims[0].generation, receipts[0].generation);
    }

    #[tokio::test]
    async fn large_legal_typed_source_falls_back_within_the_delivery_budget() {
        let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
        let parent_id = "large-source-parent";
        let child_id = "large-source-child";
        save_waiting_test_parent(
            &store,
            parent_id,
            vec![child_id.into()],
            ChildWaitPolicy::All,
        )
        .await;
        let mut parent = store.load_session(parent_id).await.unwrap().unwrap();
        parent.add_message(Message::user("required original source"));
        store.save_session(&parent).await.unwrap();
        let mut child = typed_test_child(&parent, child_id, &typed_test_report().to_string());
        let source = ChildCompletionSource::from_committed_session(&child).unwrap();
        let run_length =
            3900 - serde_json::to_vec(&source).unwrap().len() + source.activation_run_id.len();
        ChildCompletionSource::prepare(&mut child, &"r".repeat(run_length), &Default::default());
        let completion = typed_test_completion(&child);
        let source_bytes = serde_json::to_vec(completion.source.as_ref().unwrap())
            .unwrap()
            .len();
        assert!((3800..=4096).contains(&source_bytes));
        store.save_session(&child).await.unwrap();

        coordinator.on_child_completed(completion).await;
        let claims = inbox.claim(parent_id, 8).await.unwrap();
        assert_eq!(claims.len(), 1);
        assert_eq!(launches.load(Ordering::SeqCst), 1);
        assert!(fits_child_result_budget(&claims[0].envelope));
        let message = claims[0].envelope.to_provider_message().unwrap();
        assert!(fits_child_result_budget(&message));
        assert!(message.content.contains("result_budget_exceeded"));
    }

    #[tokio::test]
    async fn upgrade_promotes_one_existing_legacy_partial_delivery_without_a_second_id() {
        let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
        let parent_id = "legacy-partial-parent";
        save_waiting_test_parent(
            &store,
            parent_id,
            vec!["legacy-ready".into(), "legacy-slow".into()],
            ChildWaitPolicy::All,
        )
        .await;
        let completion =
            save_completion_test_child(&store, parent_id, "legacy-ready", "completed").await;
        save_completion_test_child(&store, parent_id, "legacy-slow", "running").await;
        let parent = store.load_session(parent_id).await.unwrap().unwrap();
        let wait = read_runtime_state(&parent).waiting_for_children.unwrap();
        let result = "result-legacy-ready";
        let presentation = runtime_resume_message(&completion, 1, Some(result));
        let mut legacy = child_completion_envelope(
            &completion,
            wait.registered_at,
            Some(result.into()),
            &presentation,
        );
        legacy.id = child_completion_id(&completion, wait.registered_at, Some(result), true);
        let receipt = inbox.deliver(&legacy).await.unwrap();
        assert!(!inbox.inspect(parent_id).await.unwrap().activation_pending());

        coordinator.on_child_completed(completion.clone()).await;
        coordinator.on_child_completed(completion).await;
        let claims = inbox.claim(parent_id, 8).await.unwrap();
        assert_eq!(claims.len(), 1);
        assert_eq!(claims[0].envelope.id, legacy.id);
        assert_eq!(claims[0].generation, receipt.generation);
        assert_eq!(launches.load(Ordering::SeqCst), 1);
        assert_eq!(
            read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                .waiting_for_children
                .as_ref(),
            Some(&wait)
        );
    }

    #[tokio::test]
    async fn upgrade_keeps_admitted_legacy_raw_guardian_and_typed_sources_deduplicated_without_a_wait(
    ) {
        for presentation_kind in ["raw", "guardian", "typed", "typed_unavailable_digest"] {
            let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
            let parent_id = "legacy-admitted-parent";
            let child_id = "legacy-admitted-child";
            let mut parent = Session::new(parent_id, "model");
            parent.add_message(Message::user("required original source"));
            store.save_session(&parent).await.unwrap();
            let completion = if presentation_kind.starts_with("typed") {
                let child = typed_test_child(&parent, child_id, &typed_test_report().to_string());
                store.save_session(&child).await.unwrap();
                typed_test_completion(&child)
            } else {
                save_completion_test_child(&store, parent_id, child_id, "completed").await
            };
            let child = store.load_session(child_id).await.unwrap().unwrap();
            let result = completion.source.as_ref().unwrap().result(&child, &parent);
            let mut projection = sealed_child_report(&completion, &child, &parent);
            if presentation_kind == "typed_unavailable_digest" {
                projection = Some(unavailable_child_result(
                    "typed_result",
                    "result_budget_exceeded",
                ));
            }
            let mut presentation = match presentation_kind {
                "guardian" => guardian_resume_message(&completion, &GuardianVerdict::approved()),
                "typed" | "typed_unavailable_digest" => {
                    typed_child_resume_message(&completion, 0, projection.as_ref().unwrap())
                }
                _ => runtime_resume_message(&completion, 0, result.as_deref()),
            };
            if presentation_kind == "typed" {
                presentation
                    .metadata
                    .as_mut()
                    .unwrap()
                    .as_object_mut()
                    .unwrap()
                    .remove("child_completion_source_sha256");
            }
            let old_wait = Utc::now() - chrono::Duration::minutes(2);
            let mut legacy =
                child_completion_envelope(&completion, old_wait, result.clone(), &presentation);
            legacy.id = child_completion_id(&completion, old_wait, result.as_deref(), true);
            if let Some(projection) = projection.as_mut() {
                assert!(bound_typed_child_delivery(
                    &mut legacy,
                    &mut presentation,
                    projection,
                    &completion,
                    0
                ));
            }
            let receipt = inbox.deliver(&legacy).await.unwrap();
            inbox
                .mark_message_activation_eligible(
                    parent_id,
                    &receipt,
                    SessionActivationPolicy::InterruptSpecificWait,
                )
                .await
                .unwrap();
            let claims = inbox.claim(parent_id, 1).await.unwrap();
            parent.add_message(claims[0].envelope.to_provider_message().unwrap());
            store.save_session(&parent).await.unwrap();
            inbox.ack(parent_id, &claims[0]).await.unwrap();

            coordinator.on_child_completed(completion).await;
            assert_eq!(
                inbox.inspect(parent_id).await.unwrap().generation,
                receipt.generation,
                "{presentation_kind}"
            );
            assert_eq!(
                inbox.inspect(parent_id).await.unwrap().pending,
                0,
                "{presentation_kind}"
            );
            assert_eq!(launches.load(Ordering::SeqCst), 0, "{presentation_kind}");
        }
    }

    #[tokio::test]
    async fn upgrade_reuses_a_pending_legacy_source_after_its_wait_was_cleared() {
        for (arm, acked) in [
            ("none", false),
            ("partial", false),
            ("terminal", false),
            ("terminal", true),
        ] {
            let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
            let parent_id = "legacy-pending-parent";
            let child_id = "legacy-pending-child";
            let mut parent = Session::new(parent_id, "model");
            store.save_session(&parent).await.unwrap();
            let completion =
                save_completion_test_child(&store, parent_id, child_id, "completed").await;
            let result = "result-legacy-pending-child";
            let presentation = runtime_resume_message(&completion, 0, Some(result));
            let old_wait = Utc::now() - chrono::Duration::minutes(2);
            let mut legacy = child_completion_envelope(
                &completion,
                old_wait,
                Some(result.into()),
                &presentation,
            );
            legacy.id = child_completion_id(&completion, old_wait, Some(result), true);
            let receipt = inbox.deliver(&legacy).await.unwrap();
            inbox
                .mark_message_activation_eligible(
                    parent_id,
                    &receipt,
                    SessionActivationPolicy::RespectSpecificWait,
                )
                .await
                .unwrap();
            if acked {
                let claims = inbox.claim(parent_id, 1).await.unwrap();
                parent.add_message(claims[0].envelope.to_provider_message().unwrap());
                store.save_session(&parent).await.unwrap();
                inbox.ack(parent_id, &claims[0]).await.unwrap();
            }
            let new_wait = if arm != "none" {
                let mut ids = vec![child_id.into()];
                if arm == "partial" {
                    ids.push("legacy-pending-slow".into());
                    save_completion_test_child(&store, parent_id, "legacy-pending-slow", "running")
                        .await;
                }
                let wait =
                    WaitingForChildrenState::for_children(ids, ChildWaitPolicy::All, Utc::now());
                let mut runtime = read_runtime_state(&parent);
                runtime.waiting_for_children = Some(wait.clone());
                runtime.status = AgentStatusState::Suspended;
                write_runtime_state(&mut parent, &runtime);
                store.save_session(&parent).await.unwrap();
                Some(wait)
            } else {
                None
            };

            coordinator.on_child_completed(completion.clone()).await;
            coordinator.on_child_completed(completion).await;
            let backlog = inbox.inspect(parent_id).await.unwrap();
            assert_eq!(
                backlog.generation, receipt.generation,
                "{arm}, acked={acked}"
            );
            assert_eq!(backlog.pending, usize::from(!acked), "{arm}, acked={acked}");
            assert_eq!(backlog.interrupt_pending(), !acked, "{arm}, acked={acked}");
            // Clearing a fresh terminal arm still requests activation. This
            // counting adapter reserves unconditionally; the production adapter
            // rechecks the Inbox and returns NoWork for an ACK-only receipt.
            assert_eq!(
                launches.load(Ordering::SeqCst),
                usize::from(!acked || arm == "terminal"),
                "{arm}, acked={acked}"
            );
            let wait = read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                .waiting_for_children;
            assert_eq!(wait, if arm == "partial" { new_wait } else { None });
            let claims = inbox.claim(parent_id, 8).await.unwrap();
            assert_eq!(claims.len(), usize::from(!acked));
            if !acked {
                assert_eq!(claims[0].envelope.id, legacy.id);
                assert_eq!(claims[0].generation, receipt.generation);
            }
        }
    }

    #[tokio::test]
    async fn admitted_legacy_guardian_source_still_commits_a_pending_verdict() {
        let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
        let parent_id = "legacy-pending-guardian-parent";
        let child_id = "legacy-pending-guardian-child";
        let mut parent = Session::new(parent_id, "model");
        let mut guardian = crate::runtime::guardian_state::ensure_guardian_state(&parent);
        guardian.record_spawn(child_id);
        write_guardian_state(&mut parent, guardian);
        store.save_session(&parent).await.unwrap();
        let verdict = GuardianVerdict::approved();
        let mut child = Session::new_child(child_id, parent_id, "model", "Guardian");
        child.set_subagent_type("guardian");
        child.add_message(Message::assistant(
            serde_json::to_string(&verdict).unwrap(),
            None,
        ));
        child.set_last_run_status("completed");
        save_sourced_test_child(&store, &child).await;
        let completion = ChildCompletion {
            parent_session_id: parent_id.into(),
            child_session_id: child_id.into(),
            status: "completed".into(),
            error: None,
            completed_at: child.updated_at,
            source: test_child_source(&store, child_id).await,
        };
        let result = serde_json::to_string(&verdict).unwrap();
        let old_wait = Utc::now() - chrono::Duration::minutes(2);
        let presentation = guardian_resume_message(&completion, &verdict);
        let mut legacy =
            child_completion_envelope(&completion, old_wait, Some(result.clone()), &presentation);
        legacy.id = child_completion_id(&completion, old_wait, Some(&result), true);
        let receipt = inbox.deliver(&legacy).await.unwrap();
        inbox
            .mark_message_activation_eligible(
                parent_id,
                &receipt,
                SessionActivationPolicy::InterruptSpecificWait,
            )
            .await
            .unwrap();
        let claims = inbox.claim(parent_id, 1).await.unwrap();
        parent.add_message(claims[0].envelope.to_provider_message().unwrap());
        store.save_session(&parent).await.unwrap();
        inbox.ack(parent_id, &claims[0]).await.unwrap();

        coordinator.on_child_completed(completion.clone()).await;
        let saved = store.load_session(parent_id).await.unwrap().unwrap();
        let guardian = read_guardian_state(&saved).unwrap();
        assert!(guardian.phase.is_reviewed());
        assert!(guardian.last_approved());
        assert_eq!(guardian.review_count, 1);
        let saved_guardian = saved.metadata["guardian.state"].clone();
        coordinator.on_child_completed(completion).await;
        assert_eq!(
            store
                .load_session(parent_id)
                .await
                .unwrap()
                .unwrap()
                .metadata["guardian.state"],
            saved_guardian
        );
        let backlog = inbox.inspect(parent_id).await.unwrap();
        assert_eq!(backlog.generation, receipt.generation);
        assert_eq!(backlog.pending, 0);
        assert_eq!(launches.load(Ordering::SeqCst), 0);
    }

    #[tokio::test]
    async fn terminal_index_ahead_of_callbacks_preserves_every_waited_child_outcome() {
        for partial_first in [false, true] {
            let (_temp, store, inbox, coordinator, reservations, launches) =
                completion_inbox_fixture().await;
            let parent_id = "parallel-parent";
            let ids: Vec<_> = (0..4)
                .map(|slot| format!("parallel-child-{slot}"))
                .collect();
            save_waiting_test_parent(&store, parent_id, ids.clone(), ChildWaitPolicy::All).await;
            for (slot, id) in ids.iter().enumerate() {
                let mut child = Session::new_child(id, parent_id, "model", "Child");
                child.add_message(Message::assistant(format!("result-{id}"), None));
                child.set_last_run_status(if partial_first && slot > 0 {
                    "running"
                } else {
                    "completed"
                });
                save_sourced_test_child(&store, &child).await;
            }
            let mut sources = HashMap::new();
            for id in &ids {
                sources.insert(id.clone(), test_child_source(&store, id).await);
            }
            let completion = |id: &str| ChildCompletion {
                parent_session_id: parent_id.into(),
                child_session_id: id.into(),
                status: "completed".into(),
                error: None,
                completed_at: Utc::now(),
                source: sources.get(id).cloned().flatten(),
            };
            if partial_first {
                ChildCompletionHandler::on_child_completed(
                    coordinator.as_ref(),
                    completion(&ids[0]),
                )
                .await;
                assert!(
                    read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                        .waiting_for_children
                        .is_some()
                );
                assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 1);
                assert_eq!(launches.load(Ordering::SeqCst), 1);
                for id in &ids[1..] {
                    let mut child = store.load_session(id).await.unwrap().unwrap();
                    child.set_last_run_status("completed");
                    save_sourced_test_child(&store, &child).await;
                }
            }
            // The index knows all four outcomes, but only one completion
            // callback has reached the coordinator at this wake boundary.
            for id in &ids {
                sources.insert(id.clone(), test_child_source(&store, id).await);
            }
            let completion = |id: &str| ChildCompletion {
                parent_session_id: parent_id.into(),
                child_session_id: id.into(),
                status: "completed".into(),
                error: None,
                completed_at: Utc::now(),
                source: sources.get(id).cloned().flatten(),
            };
            ChildCompletionHandler::on_child_completed(coordinator.as_ref(), completion(&ids[1]))
                .await;
            assert!(
                read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                    .waiting_for_children
                    .is_none()
            );
            assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 4);
            assert_eq!(reservations.load(Ordering::SeqCst), 1);
            assert_eq!(launches.load(Ordering::SeqCst), 1);
            for id in &ids {
                ChildCompletionHandler::on_child_completed(coordinator.as_ref(), completion(id))
                    .await;
            }
            let claims = inbox.claim(parent_id, 8).await.unwrap();
            assert_eq!(claims.len(), 4, "late callbacks cannot duplicate outcomes");
            assert_eq!(reservations.load(Ordering::SeqCst), 1);
            for id in &ids {
                let matching: Vec<_> = claims
                    .iter()
                    .filter_map(|claim| match &claim.envelope.body {
                        SessionMessageBody::ChildOutcome(outcome)
                            if &outcome.child_session_id == id =>
                        {
                            Some(outcome)
                        }
                        _ => None,
                    })
                    .collect();
                assert_eq!(matching.len(), 1);
                assert_eq!(
                    matching[0].result.as_deref(),
                    Some(format!("result-{id}").as_str())
                );
            }
        }
    }

    #[tokio::test]
    async fn saved_a_callback_barrier_never_attributes_completed_b_to_a() {
        for guardian in [false, true] {
            let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
            let parent_id = "source-barrier-parent";
            let child_id = "source-barrier-child";
            save_waiting_test_parent(
                &store,
                parent_id,
                vec![child_id.into()],
                ChildWaitPolicy::All,
            )
            .await;
            if guardian {
                let mut parent = store.load_session(parent_id).await.unwrap().unwrap();
                let mut state = crate::runtime::guardian_state::ensure_guardian_state(&parent);
                state.record_spawn(child_id);
                write_guardian_state(&mut parent, state);
                store.save_session(&parent).await.unwrap();
            }
            let mut child = Session::new_child(child_id, parent_id, "model", "Child");
            if guardian {
                child
                    .metadata
                    .insert("subagent_type".into(), "guardian".into());
            }
            child.add_message(Message::user("assignment A"));
            store.save_session(&child).await.unwrap();
            let prior = child.messages.iter().map(|m| m.id.clone()).collect();
            child.add_message(Message::assistant("answer A", None));
            child.set_last_run_status("completed");
            ChildCompletionSource::prepare(&mut child, "actual-run-A", &prior);
            coordinator
                .agent
                .persistence()
                .save_runtime_session(&mut child)
                .await
                .unwrap();
            let completion_a = ChildCompletion {
                parent_session_id: parent_id.into(),
                child_session_id: child_id.into(),
                status: "completed".into(),
                error: None,
                completed_at: Utc::now(),
                source: ChildCompletionSource::after_final_save(&child, true),
            };
            assert_eq!(
                completion_a.source.as_ref().unwrap().activation_run_id,
                "actual-run-A"
            );
            // Pause the real callback boundary after A's canonical save. B can
            // now finish on the same Child before A enters the coordinator.
            let (release, barrier) = tokio::sync::oneshot::channel();
            let target = coordinator.clone();
            let callback = tokio::spawn(async move {
                barrier.await.unwrap();
                target.on_child_completed(completion_a).await;
            });
            let prior = child.messages.iter().map(|m| m.id.clone()).collect();
            child.add_message(Message::user("assignment B"));
            let answer_b = if guardian {
                r#"{"approve":true,"summary":"B approved","findings":[]}"#
            } else {
                "answer B"
            };
            child.add_message(Message::assistant(answer_b, None));
            child.set_last_run_status("completed");
            ChildCompletionSource::prepare(&mut child, "actual-run-B", &prior);
            coordinator
                .agent
                .persistence()
                .save_runtime_session(&mut child)
                .await
                .unwrap();
            release.send(()).unwrap();
            callback.await.unwrap();
            assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 0);
            let parent = store.load_session(parent_id).await.unwrap().unwrap();
            assert!(read_runtime_state(&parent).waiting_for_children.is_some());
            assert_eq!(launches.load(Ordering::SeqCst), 0);
            if guardian {
                assert!(read_guardian_state(&parent).unwrap().phase.is_pending());
            }

            let completion_b = ChildCompletion {
                parent_session_id: parent_id.into(),
                child_session_id: child_id.into(),
                status: "completed".into(),
                error: None,
                completed_at: Utc::now(),
                source: ChildCompletionSource::after_final_save(&child, true),
            };
            coordinator.on_child_completed(completion_b.clone()).await;
            coordinator.on_child_completed(completion_b).await;
            let claims = inbox.claim(parent_id, 8).await.unwrap();
            assert_eq!(
                claims.len(),
                1,
                "B is admitted once with B's own provenance"
            );
            let SessionMessageBody::ChildOutcome(outcome) = &claims[0].envelope.body else {
                panic!("expected ChildOutcome");
            };
            assert_eq!(outcome.result.as_deref(), Some(answer_b));
            assert_eq!(launches.load(Ordering::SeqCst), 1);
        }
    }

    #[tokio::test]
    async fn missing_source_cannot_import_latest_guardian_approval() {
        let (_temp, store, inbox, coordinator, _, _) = completion_inbox_fixture().await;
        let parent_id = "unavailable-guardian-parent";
        let child_id = "unavailable-guardian-child";
        save_waiting_test_parent(
            &store,
            parent_id,
            vec![child_id.into()],
            ChildWaitPolicy::All,
        )
        .await;
        let mut parent = store.load_session(parent_id).await.unwrap().unwrap();
        let mut state = crate::runtime::guardian_state::ensure_guardian_state(&parent);
        state.record_spawn(child_id);
        write_guardian_state(&mut parent, state);
        store.save_session(&parent).await.unwrap();
        let mut child = Session::new_child(child_id, parent_id, "model", "Child");
        child
            .metadata
            .insert("subagent_type".into(), "guardian".into());
        child.add_message(Message::assistant(
            r#"{"approve":true,"summary":"unbound latest","findings":[]}"#,
            None,
        ));
        child.set_last_run_status("completed");
        store.save_session(&child).await.unwrap();
        coordinator
            .on_child_completed(ChildCompletion {
                parent_session_id: parent_id.into(),
                child_session_id: child_id.into(),
                status: "completed".into(),
                error: None,
                completed_at: Utc::now(),
                source: None,
            })
            .await;
        let parent = store.load_session(parent_id).await.unwrap().unwrap();
        let state = read_guardian_state(&parent).unwrap();
        assert!(state.phase.is_reviewed());
        assert!(
            !state.last_approved(),
            "source-less latest approval is never authoritative"
        );
        let claims = inbox.claim(parent_id, 8).await.unwrap();
        let SessionMessageBody::ChildOutcome(outcome) = &claims[0].envelope.body else {
            panic!("expected ChildOutcome");
        };
        assert!(outcome.result.is_none());
    }

    #[tokio::test]
    async fn lost_callback_replay_uses_the_committed_seal() {
        for valid in [true, false] {
            let (_temp, store, inbox, coordinator, _, _) = completion_inbox_fixture().await;
            save_waiting_test_parent(
                &store,
                "replay-parent",
                vec!["replay-child".into()],
                ChildWaitPolicy::All,
            )
            .await;
            let mut child = Session::new_child("replay-child", "replay-parent", "model", "Child");
            child.add_message(Message::assistant("sealed replay result", None));
            child.set_last_run_status("completed");
            save_sourced_test_child(&store, &child).await;
            if !valid {
                let mut child = store.load_session("replay-child").await.unwrap().unwrap();
                child.messages[0].content = "unsealed successor result".into();
                store.save_session(&child).await.unwrap();
            }
            let parent = store.load_session("replay-parent").await.unwrap().unwrap();
            let mut wait = read_runtime_state(&parent).waiting_for_children.unwrap();
            wait.registered_at = Utc::now() - chrono::Duration::seconds(61);
            coordinator.sweep_child_wait("replay-parent", wait).await;
            let claims = inbox.claim("replay-parent", 1).await.unwrap();
            if !valid {
                assert!(claims.is_empty());
                let parent = store.load_session("replay-parent").await.unwrap().unwrap();
                assert!(read_runtime_state(&parent).waiting_for_children.is_some());
                continue;
            }
            let SessionMessageBody::ChildOutcome(outcome) = &claims[0].envelope.body else {
                panic!("expected ChildOutcome");
            };
            assert_eq!(outcome.result.as_deref(), Some("sealed replay result"));
        }
    }

    #[tokio::test]
    async fn partial_admission_then_source_mutation_ignores_stale_retry() {
        for parent_changed in [false, true] {
            let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
            let parent_id = "partial-source-parent";
            let child_id = "partial-source-child";
            save_waiting_test_parent(
                &store,
                parent_id,
                vec![child_id.into(), "still-running-child".into()],
                ChildWaitPolicy::All,
            )
            .await;
            let mut child = Session::new_child(child_id, parent_id, "model", "Child");
            if parent_changed {
                let mut parent = store.load_session(parent_id).await.unwrap().unwrap();
                let mut source = Message::user("required parent constraint");
                source.id = "parent-source".into();
                parent.add_message(source);
                store.save_session(&parent).await.unwrap();
                let packet = bamboo_domain::ChildContextPacket {
                    version: 1,
                    objective: "task".into(),
                    constraints: vec![],
                    acceptance: vec!["source remains attributable".into()],
                    non_goals: vec![],
                    necessary_user_instructions: vec![],
                    recorded_decisions: vec![],
                    source_user_message_ids: vec!["parent-source".into()],
                    background_message_ids: vec![],
                };
                let resolved = packet.resolve(&parent, "task").unwrap();
                let binding = bamboo_domain::ChildContextBinding::new(
                    &parent,
                    child_id,
                    resolved.required_brief.clone(),
                    resolved,
                )
                .unwrap();
                binding.install(&mut child).unwrap();
                child.add_message(binding.assignment_message());
            }
            child.add_message(Message::assistant("original source", None));
            child.set_last_run_status("completed");
            save_sourced_test_child(&store, &child).await;
            let completion = ChildCompletion {
                parent_session_id: parent_id.into(),
                child_session_id: child_id.into(),
                status: "completed".into(),
                error: None,
                completed_at: Utc::now(),
                source: test_child_source(&store, child_id).await,
            };
            coordinator.on_child_completed(completion.clone()).await;
            assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 1);
            if parent_changed {
                let mut parent = store.load_session(parent_id).await.unwrap().unwrap();
                parent.messages[0].content = "modified parent source".into();
                store.save_session(&parent).await.unwrap();
            } else {
                child = store.load_session(child_id).await.unwrap().unwrap();
                child.messages[0].content = "modified after admission".into();
                store.save_session(&child).await.unwrap();
            }
            coordinator.on_child_completed(completion).await;
            assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 1);
            assert_eq!(launches.load(Ordering::SeqCst), 1);
            assert!(
                read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                    .waiting_for_children
                    .is_some()
            );
            let mut sibling =
                Session::new_child("still-running-child", parent_id, "model", "Child");
            sibling.add_message(Message::assistant("sibling result", None));
            sibling.set_last_run_status("completed");
            save_sourced_test_child(&store, &sibling).await;
            coordinator
                .on_child_completed(ChildCompletion {
                    parent_session_id: parent_id.into(),
                    child_session_id: sibling.id.clone(),
                    status: "completed".into(),
                    error: None,
                    completed_at: Utc::now(),
                    source: test_child_source(&store, &sibling.id).await,
                })
                .await;
            assert_eq!(
                inbox.inspect(parent_id).await.unwrap().pending,
                2,
                "sibling completion cannot restage mutated A as a neutral outcome"
            );
            assert_eq!(launches.load(Ordering::SeqCst), 1);
            assert!(
                read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                    .waiting_for_children
                    .is_some()
            );
            assert_eq!(
                inbox.inspect(parent_id).await.unwrap().claimed,
                0,
                "stale replay does not create another claim or reasoning owner"
            );
        }
    }

    #[tokio::test]
    async fn short_circuit_wait_stages_terminal_siblings_without_waiting_or_foreign_content() {
        for policy in [ChildWaitPolicy::Any, ChildWaitPolicy::FirstError] {
            let (_temp, store, inbox, coordinator, _, launches) = completion_inbox_fixture().await;
            let parent_id = "short-circuit-parent";
            let ids = vec![
                "failed-child".into(),
                "done-child".into(),
                "running-child".into(),
                "foreign-child".into(),
            ];
            save_waiting_test_parent(&store, parent_id, ids, policy).await;
            for (id, owner, status, result) in [
                ("failed-child", parent_id, "error", "failed-owned-result"),
                ("done-child", parent_id, "completed", "done-owned-result"),
                ("running-child", parent_id, "running", "not-terminal"),
                (
                    "foreign-child",
                    "another-parent",
                    "completed",
                    "foreign-secret",
                ),
            ] {
                let mut child = Session::new_child(id, owner, "model", "Child");
                child.add_message(Message::assistant(result, None));
                child.set_last_run_status(status);
                save_sourced_test_child(&store, &child).await;
            }
            ChildCompletionHandler::on_child_completed(
                coordinator.as_ref(),
                ChildCompletion {
                    parent_session_id: parent_id.into(),
                    child_session_id: "failed-child".into(),
                    status: "error".into(),
                    error: None,
                    completed_at: Utc::now(),
                    source: test_child_source(&store, "failed-child").await,
                },
            )
            .await;
            assert!(
                read_runtime_state(&store.load_session(parent_id).await.unwrap().unwrap())
                    .waiting_for_children
                    .is_none()
            );
            let claims = inbox.claim(parent_id, 8).await.unwrap();
            assert_eq!(claims.len(), 2);
            let content = serde_json::to_string(
                &claims
                    .iter()
                    .map(|claim| &claim.envelope)
                    .collect::<Vec<_>>(),
            )
            .unwrap();
            assert!(content.contains("done-owned-result"));
            assert!(!content.contains("foreign-secret"));
            assert!(!content.contains("not-terminal"));
            assert_eq!(
                store
                    .load_session("running-child")
                    .await
                    .unwrap()
                    .unwrap()
                    .last_run_status()
                    .as_deref(),
                Some("running")
            );
            assert_eq!(launches.load(Ordering::SeqCst), 1);
        }
    }

    #[tokio::test]
    async fn sibling_admission_failure_keeps_wait_armed_without_activation() {
        let limits = bamboo_domain::SessionInboxLimits {
            max_backlog: 1,
            ..Default::default()
        };
        let (_temp, store, inbox, coordinator, reservations, launches) =
            completion_inbox_fixture_with_limits(limits).await;
        let parent_id = "partial-stage-parent";
        let ids = vec!["first-child".into(), "second-child".into()];
        save_waiting_test_parent(&store, parent_id, ids, ChildWaitPolicy::All).await;
        for id in ["first-child", "second-child"] {
            let mut child = Session::new_child(id, parent_id, "model", "Child");
            child.add_message(Message::assistant(format!("result-{id}"), None));
            child.set_last_run_status("completed");
            save_sourced_test_child(&store, &child).await;
        }
        ChildCompletionHandler::on_child_completed(
            coordinator.as_ref(),
            ChildCompletion {
                parent_session_id: parent_id.into(),
                child_session_id: "first-child".into(),
                status: "completed".into(),
                error: None,
                completed_at: Utc::now(),
                source: test_child_source(&store, "first-child").await,
            },
        )
        .await;
        let durable = store.load_session(parent_id).await.unwrap().unwrap();
        assert!(read_runtime_state(&durable).waiting_for_children.is_some());
        assert_eq!(
            read_runtime_state(&durable).status,
            AgentStatusState::Suspended
        );
        assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 1);
        let backlog = inbox.inspect(parent_id).await.unwrap();
        assert!(backlog.activation_pending());
        assert!(
            !backlog.interrupt_pending(),
            "failed wait transition is still respected"
        );
        assert_eq!(reservations.load(Ordering::SeqCst), 0);
        assert_eq!(launches.load(Ordering::SeqCst), 0);
    }

    #[tokio::test]
    async fn supervisor_resume_rejects_old_incarnation_without_replacing_cache() {
        let (_temp, store, _inbox, coordinator, _reservations, _launches) =
            completion_inbox_fixture().await;
        let original = store
            .get_or_create_default_supervisor("model")
            .await
            .unwrap();
        let mut stale = store
            .load_session(&original.session_id)
            .await
            .unwrap()
            .unwrap();
        store.delete_session(&original.session_id).await.unwrap();
        let recreated = store
            .get_or_create_default_supervisor("replacement")
            .await
            .unwrap();
        assert_ne!(original.incarnation_id, recreated.incarnation_id);
        let mut current = store
            .load_session(&recreated.session_id)
            .await
            .unwrap()
            .unwrap();
        coordinator.save_and_cache(&mut current).await;
        let cached = coordinator
            .sessions
            .get(&current.id)
            .unwrap()
            .value()
            .clone();
        coordinator.save_and_cache(&mut stale).await;
        assert!(Arc::ptr_eq(
            &cached,
            coordinator.sessions.get(&current.id).unwrap().value()
        ));
        assert_eq!(
            store
                .load_session(&current.id)
                .await
                .unwrap()
                .unwrap()
                .authority_identity,
            current.authority_identity
        );
    }

    // ── child-wait watchdog pure helpers (issue #546) ────────────────────

    #[test]
    fn dead_child_candidate_status_matrix() {
        // Never ran / lost before the running marker: dead candidate.
        assert!(is_dead_child_candidate_status(None));
        // Actively-reported non-terminal statuses: dead candidates when nothing
        // is driving them.
        assert!(is_dead_child_candidate_status(Some("running")));
        assert!(is_dead_child_candidate_status(Some("pending")));
        // Legitimately quiescent: waiting on a human / own children / bash.
        assert!(!is_dead_child_candidate_status(Some("suspended")));
        // Terminal statuses can never be "dead" — they are already done.
        for status in ["completed", "error", "timeout", "cancelled", "skipped"] {
            assert!(!is_dead_child_candidate_status(Some(status)), "{status}");
        }
    }

    #[test]
    fn completion_child_ownership_gates_content_fold() {
        // Owned: the child's own parent linkage matches the reporting parent.
        assert!(completion_child_is_owned("parent-1", Some("parent-1")));
        // Foreign: a real session that belongs to a DIFFERENT parent — its
        // content must never be folded into parent-1's transcript.
        assert!(!completion_child_is_owned("parent-1", Some("parent-2")));
        // Root/unparented session, or a nonexistent id (linkage None).
        assert!(!completion_child_is_owned("parent-1", None));
    }

    #[test]
    fn watchdog_resume_messages_are_hidden_runtime_messages() {
        for message in [
            empty_child_wait_message(),
            child_wait_lease_expired_message(&["c-1".to_string(), "c-2".to_string()]),
        ] {
            assert!(matches!(message.role, Role::User));
            let meta = message.metadata.expect("hidden runtime metadata");
            assert_eq!(meta[RUNTIME_RESUME_MESSAGE_HIDDEN_KEY], true);
            assert_eq!(
                meta[RUNTIME_RESUME_MESSAGE_KIND_KEY],
                "child_wait_watchdog_resume"
            );
        }
        let lease = child_wait_lease_expired_message(&["c-1".to_string()]);
        // The lease message must never claim the children finished.
        assert!(lease.content.contains("NOT cancelled"));
        assert!(lease.content.contains("c-1"));
    }

    #[test]
    fn runtime_child_guidance_uses_compact_inspection_calls() {
        let completion = make_completion("completed");
        let child_inspection = serde_json::json!({
            "intent": "inspect", "target": "child-1", "message": "messages",
        })
        .to_string();
        let tree_inspection =
            serde_json::json!({"intent": "inspect", "message": "tree"}).to_string();
        for (message, expected_call) in [
            (
                runtime_resume_message(&completion, 0, Some("evidence")),
                &child_inspection,
            ),
            (
                guardian_resume_message(&completion, &GuardianVerdict::approved()),
                &child_inspection,
            ),
            (empty_child_wait_message(), &tree_inspection),
            (
                child_wait_lease_expired_message(&["child-1".to_string()]),
                &tree_inspection,
            ),
        ] {
            assert!(message.content.contains(expected_call));
            for legacy in [
                "SubAgent.get",
                "SubAgent.list",
                "SubAgent.wait",
                "child_session_id=",
            ] {
                assert!(!message.content.contains(legacy), "stale hint: {legacy}");
            }
        }
    }

    // ── on_child_completed terminality guard (issue #546) ────────────────

    #[test]
    fn non_terminal_statuses_never_satisfy_wait_policies() {
        // The guard keys on `is_terminal_child_status`; "suspended" (and any
        // unknown non-terminal string) must not count toward any policy.
        assert!(!is_terminal_child_status("suspended"));
        assert!(!is_terminal_child_status("running"));
        assert!(!is_terminal_child_status("pending"));
    }

    fn make_completion(status: &str) -> ChildCompletion {
        ChildCompletion {
            parent_session_id: "parent-1".to_string(),
            child_session_id: "child-1".to_string(),
            status: status.to_string(),
            error: None,
            completed_at: Utc::now(),
            source: None,
        }
    }

    #[test]
    fn oversized_child_outcome_is_bounded_and_keeps_full_content_identity() {
        let mut completion = make_completion("completed");
        let wait_registered_at = Utc::now();
        let huge = format!("prefix-A-{}", "x".repeat(300 * 1024));
        let presentation = runtime_resume_message(&completion, 0, Some(&huge));
        let first = child_completion_envelope(
            &completion,
            wait_registered_at,
            Some(huge.clone()),
            &presentation,
        );
        assert!(
            serde_json::to_vec(&first).unwrap().len()
                < bamboo_domain::SessionInboxLimits::default().max_payload_bytes
        );
        let SessionMessageBody::ChildOutcome(outcome) = &first.body else {
            panic!("typed child outcome");
        };
        let stored = outcome.result.as_deref().unwrap();
        assert!(stored.contains("sha256="));
        assert!(stored.contains("\"intent\":\"inspect\""));
        assert!(stored.contains("\"target\":\"child-1\""));
        assert!(stored.contains("\"message\":\"result\""));
        assert!(!stored.contains("SubAgent.get"));
        assert!(stored.len() < CHILD_COMPLETION_INLINE_FIELD_BYTES);

        // Retry-only completion timestamps and provider presentation do not
        // change the logical id; full oversized content does.
        completion.completed_at += chrono::Duration::seconds(1);
        let exact_retry = child_completion_envelope(
            &completion,
            wait_registered_at,
            Some(huge.clone()),
            &runtime_resume_message(&completion, 9, Some(&huge)),
        );
        assert_eq!(exact_retry.id, first.id);
        let changed = format!("prefix-B-{}", "x".repeat(300 * 1024));
        let corrected = child_completion_envelope(
            &completion,
            wait_registered_at,
            Some(changed.clone()),
            &runtime_resume_message(&completion, 0, Some(&changed)),
        );
        assert_ne!(corrected.id, first.id);
    }

    #[test]
    fn oversized_child_error_points_to_bounded_error_view() {
        let mut completion = make_completion("error");
        completion.error = Some("失败🙂".repeat(30_000));
        let presentation = runtime_resume_message(&completion, 0, None);
        let envelope = child_completion_envelope(&completion, Utc::now(), None, &presentation);
        assert!(
            serde_json::to_vec(&envelope).unwrap().len()
                < bamboo_domain::SessionInboxLimits::default().max_payload_bytes
        );
        let SessionMessageBody::ChildOutcome(outcome) = &envelope.body else {
            panic!("typed child outcome");
        };
        let stored = outcome.error.as_deref().expect("bounded error");
        assert!(stored.contains(r#""message":"diagnostics""#));
        assert!(stored.contains("authenticated inspector"));
        assert!(!stored.contains("bounded slices"));
        assert!(!stored.contains("cursors"));
        assert!(stored.len() < CHILD_COMPLETION_INLINE_FIELD_BYTES);
        let metadata_error = outcome
            .provider_message
            .as_ref()
            .and_then(|provider| provider.metadata.get("child_error"))
            .and_then(serde_json::Value::as_str)
            .expect("bounded metadata error");
        assert_eq!(metadata_error, stored);
        assert!(!metadata_error.contains(&"失败🙂".repeat(30_000)));
    }

    #[test]
    fn escaped_and_mixed_child_fields_fit_the_serialized_inbox_budget() {
        let mut completion = make_completion("error");
        completion.error = Some("\u{0001}".repeat(CHILD_COMPLETION_INLINE_FIELD_BYTES));
        let result = "\n".repeat(CHILD_COMPLETION_INLINE_FIELD_BYTES);
        let wait_registered_at = Utc::now();
        let presentation = runtime_resume_message(&completion, 0, Some(&result));
        let envelope = child_completion_envelope(
            &completion,
            wait_registered_at,
            Some(result.clone()),
            &presentation,
        );
        assert!(
            serde_json::to_vec(&envelope).unwrap().len()
                < bamboo_domain::SessionInboxLimits::default().max_payload_bytes
        );
        let SessionMessageBody::ChildOutcome(outcome) = &envelope.body else {
            panic!("typed child outcome");
        };
        let error = outcome.error.as_deref().unwrap();
        let stored_result = outcome.result.as_deref().unwrap();
        assert!(error.contains(r#""message":"diagnostics""#));
        assert!(stored_result.contains(r#""message":"result""#));
        assert!(!error.contains("Bounded tail"));
        assert!(!stored_result.contains("Bounded tail"));
        assert_eq!(
            outcome
                .provider_message
                .as_ref()
                .and_then(|provider| provider.metadata.get("child_error"))
                .and_then(serde_json::Value::as_str),
            Some(error)
        );
        let retry = child_completion_envelope(
            &completion,
            wait_registered_at,
            Some(result),
            &runtime_resume_message(&completion, 8, None),
        );
        assert_eq!(envelope.id, retry.id);
    }

    #[test]
    fn compact_guardian_outcome_keeps_rejected_verdict_and_result_retrieval() {
        let completion = make_completion("completed");
        let result = format!(
            "{}{{\"approve\":false,\"findings\":[\"needs fix\"]}}",
            "\u{0001}".repeat(CHILD_COMPLETION_INLINE_FIELD_BYTES - 80)
        );
        assert!(result.len() <= CHILD_COMPLETION_INLINE_FIELD_BYTES);
        let verdict = parse_guardian_verdict(&result).expect("trailing guardian verdict");
        let presentation = guardian_resume_message(&completion, &verdict);
        assert!(
            serde_json::to_vec(&presentation).unwrap().len() <= CHILD_COMPLETION_INLINE_FIELD_BYTES,
            "the original guardian presentation must fit the inline cap"
        );
        let wait_registered_at = Utc::now();
        let expanded = build_child_completion_envelope(
            &completion,
            wait_registered_at,
            Some(&result),
            &presentation,
            false,
        );
        assert!(
            serde_json::to_vec(&expanded).unwrap().len()
                > bamboo_domain::SessionInboxLimits::default().max_payload_bytes,
            "JSON escaping must trigger the compact fallback"
        );

        let envelope =
            child_completion_envelope(&completion, wait_registered_at, Some(result), &presentation);
        assert!(
            serde_json::to_vec(&envelope).unwrap().len()
                <= bamboo_domain::SessionInboxLimits::default().max_payload_bytes
        );
        let SessionMessageBody::ChildOutcome(outcome) = &envelope.body else {
            panic!("typed child outcome");
        };
        let provider = outcome.provider_message.as_ref().unwrap();
        assert!(provider.content.text.contains("Guardian review REJECTED"));
        assert!(provider.content.text.contains(r#""message":"result""#));
        assert_eq!(
            provider.metadata.get("guardian_approved"),
            Some(&serde_json::json!(false))
        );
        assert!(envelope
            .to_provider_message()
            .unwrap()
            .content
            .contains("Guardian review REJECTED"));
    }

    #[tokio::test]
    async fn oversized_multibyte_child_error_is_durable_and_replayable() {
        let home = tempfile::tempdir().unwrap();
        let store = Arc::new(
            bamboo_storage::SessionStoreV2::new(home.path().to_path_buf())
                .await
                .unwrap(),
        );
        let parent_id = "large-error-parent";
        store
            .save_session(&Session::new(parent_id, "model"))
            .await
            .unwrap();
        let inbox = bamboo_storage::FileSessionInbox::new(
            store.clone(),
            bamboo_domain::SessionInboxLimits::default(),
        );
        let mut completion = make_completion("error");
        completion.parent_session_id = parent_id.to_string();
        completion.error = Some("失败🙂".repeat(30_000));
        let presentation = runtime_resume_message(&completion, 0, None);
        let envelope = child_completion_envelope(&completion, Utc::now(), None, &presentation);
        inbox.deliver(&envelope).await.unwrap();
        completion.child_session_id = "escaped-error-child".to_string();
        completion.error = Some("\u{0001}".repeat(CHILD_COMPLETION_INLINE_FIELD_BYTES));
        let result = "\n".repeat(CHILD_COMPLETION_INLINE_FIELD_BYTES);
        let mixed = child_completion_envelope(
            &completion,
            Utc::now(),
            Some(result.clone()),
            &runtime_resume_message(&completion, 0, Some(&result)),
        );
        let receipt = inbox.deliver(&mixed).await.unwrap();
        inbox
            .mark_activation_eligible(
                parent_id,
                receipt.generation,
                bamboo_domain::SessionActivationPolicy::RespectSpecificWait,
            )
            .await
            .unwrap();
        drop(inbox);
        drop(store);

        let reopened_store = Arc::new(
            bamboo_storage::SessionStoreV2::new(home.path().to_path_buf())
                .await
                .unwrap(),
        );
        let reopened = bamboo_storage::FileSessionInbox::new(
            reopened_store,
            bamboo_domain::SessionInboxLimits::default(),
        );
        let claims = reopened.claim(parent_id, 2).await.unwrap();
        assert_eq!(claims.len(), 2);
        assert_eq!(claims[0].envelope.id, envelope.id);
        assert_eq!(claims[1].envelope.id, mixed.id);
        for claim in claims {
            assert!(
                serde_json::to_vec(&claim.envelope).unwrap().len()
                    < bamboo_domain::SessionInboxLimits::default().max_payload_bytes
            );
        }
    }

    #[tokio::test]
    async fn failed_child_wait_sidecar_save_keeps_cache_wait_and_transcript() {
        let (_temp, store, inbox, mut coordinator, reservations, launches) =
            completion_inbox_fixture().await;
        let parent_id = "failed-sidecar-parent";
        let child_id = "failed-sidecar-child";
        let now = Utc::now();
        let mut parent = Session::new(parent_id, "model");
        parent.add_message(Message::user("durable root turn"));
        let mut runtime = AgentRuntimeState::new("waiting-run");
        runtime.status = AgentStatusState::Suspended;
        runtime.waiting_for_children = Some(WaitingForChildrenState::for_children(
            vec![child_id.to_string()],
            ChildWaitPolicy::All,
            now,
        ));
        write_runtime_state(&mut parent, &runtime);
        parent.metadata.insert(
            "runtime.suspend_reason".to_string(),
            "waiting_for_children".to_string(),
        );
        store.save_session(&parent).await.unwrap();
        let mut cached = parent.clone();
        cached.add_message(Message::assistant("newer cached root turn", None));
        coordinator.sessions.insert(
            parent_id.to_string(),
            Arc::new(crate::SessionSnapshot::new(cached)),
        );

        let mut child = Session::new_child(child_id, parent_id, "model", "Child");
        child.set_last_run_status("completed");
        save_sourced_test_child(&store, &child).await;
        Arc::get_mut(&mut coordinator).unwrap().persistence = Arc::new(LockedSessionStore::new(
            Arc::new(FailingRuntimeSaveStorage {
                inner: store.clone(),
            }),
        ));

        ChildCompletionHandler::on_child_completed(
            coordinator.as_ref(),
            ChildCompletion {
                parent_session_id: parent_id.to_string(),
                child_session_id: child_id.to_string(),
                status: "completed".to_string(),
                error: None,
                completed_at: Utc::now(),
                source: None,
            },
        )
        .await;

        let durable = store.load_session(parent_id).await.unwrap().unwrap();
        assert!(read_runtime_state(&durable).waiting_for_children.is_some());
        assert_eq!(durable.messages.len(), 1);
        assert_eq!(durable.messages[0].content, "durable root turn");
        let cached = coordinator.sessions.get(parent_id).unwrap();
        let cached = cached.value().read();
        assert!(read_runtime_state(&cached).waiting_for_children.is_some());
        assert!(cached
            .messages
            .iter()
            .any(|message| message.content == "newer cached root turn"));
        let backlog = inbox.inspect(parent_id).await.unwrap();
        assert_eq!(backlog.pending + backlog.claimed, 1);
        assert_eq!(reservations.load(Ordering::SeqCst), 0);
        assert_eq!(launches.load(Ordering::SeqCst), 0);
    }

    #[tokio::test]
    async fn oversized_child_completion_clears_wait_and_activates_exactly_once() {
        let (_temp, store, inbox, coordinator, reservations, launches) =
            completion_inbox_fixture().await;
        let parent_id = "oversized-parent";
        let child_id = "oversized-child";
        let now = Utc::now();
        let mut parent = Session::new(parent_id, "model");
        let mut parent_runtime = AgentRuntimeState::new("waiting-run");
        parent_runtime.status = AgentStatusState::Suspended;
        parent_runtime.waiting_for_children = Some(WaitingForChildrenState::for_children(
            vec![child_id.to_string()],
            ChildWaitPolicy::All,
            now,
        ));
        parent_runtime.suspension = Some(SuspensionState {
            reason: "waiting_for_children".to_string(),
            suspended_at: now,
            resumable: true,
            hook_point: Some("ChildCompletion".to_string()),
        });
        write_runtime_state(&mut parent, &parent_runtime);
        parent.metadata.insert(
            "runtime.suspend_reason".to_string(),
            "waiting_for_children".to_string(),
        );
        store.save_session(&parent).await.unwrap();
        // The running Root can have a newer in-memory transcript while this
        // coordinator reads the durable wait. Publishing the control-plane
        // transition must not replace that transcript with the stale load.
        let mut cached_parent = parent.clone();
        cached_parent.add_message(Message::assistant("concurrent root turn", None));
        coordinator.sessions.insert(
            parent_id.to_string(),
            Arc::new(crate::SessionSnapshot::new(cached_parent)),
        );

        let mut child = Session::new_child(child_id, parent_id, "model", "Child");
        child.add_message(Message::assistant("z".repeat(300 * 1024), None));
        child.set_last_run_status("completed");
        save_sourced_test_child(&store, &child).await;
        let completion = ChildCompletion {
            parent_session_id: parent_id.to_string(),
            child_session_id: child_id.to_string(),
            status: "completed".to_string(),
            error: None,
            completed_at: Utc::now(),
            source: test_child_source(&store, child_id).await,
        };

        ChildCompletionHandler::on_child_completed(coordinator.as_ref(), completion.clone()).await;
        let durable_parent = store.load_session(parent_id).await.unwrap().unwrap();
        let durable_runtime = read_runtime_state(&durable_parent);
        assert!(durable_runtime.waiting_for_children.is_none());
        assert_eq!(durable_runtime.status, AgentStatusState::Idle);
        assert!(!durable_parent
            .metadata
            .contains_key("runtime.suspend_reason"));
        let cached_parent = coordinator.sessions.get(parent_id).unwrap();
        let cached_parent = cached_parent.value().read();
        assert!(cached_parent
            .messages
            .iter()
            .any(|message| message.content == "concurrent root turn"));
        assert!(read_runtime_state(&cached_parent)
            .waiting_for_children
            .is_none());
        let backlog = inbox.inspect(parent_id).await.unwrap();
        assert_eq!(backlog.pending + backlog.claimed, 1);
        assert!(backlog.activation_pending());
        assert_eq!(reservations.load(Ordering::SeqCst), 1);
        assert_eq!(launches.load(Ordering::SeqCst), 1);

        let claim = inbox.claim(parent_id, 1).await.unwrap().remove(0);
        assert!(
            serde_json::to_vec(&claim.envelope).unwrap().len()
                < bamboo_domain::SessionInboxLimits::default().max_payload_bytes
        );
        // Duplicate terminal notification after the wait was already cleared
        // cannot enqueue or activate a second outcome.
        ChildCompletionHandler::on_child_completed(coordinator.as_ref(), completion).await;
        assert_eq!(reservations.load(Ordering::SeqCst), 1);
        assert_eq!(launches.load(Ordering::SeqCst), 1);
        assert_eq!(inbox.inspect(parent_id).await.unwrap().claimed, 1);
    }

    #[tokio::test]
    async fn completion_before_parent_wait_does_not_checkpoint_stale_root() {
        let (_temp, store, inbox, coordinator, reservations, launches) =
            completion_inbox_fixture().await;
        let parent_id = "parent-without-wait";
        let child_id = "already-finished-child";
        let mut parent = Session::new(parent_id, "model");
        parent.add_message(Message::assistant("root is still reasoning", None));
        store.save_session(&parent).await.unwrap();
        let before = store.load_session(parent_id).await.unwrap().unwrap();

        let mut child = Session::new_child(child_id, parent_id, "model", "Child");
        child.set_last_run_status("completed");
        save_sourced_test_child(&store, &child).await;
        ChildCompletionHandler::on_child_completed(
            coordinator.as_ref(),
            ChildCompletion {
                parent_session_id: parent_id.to_string(),
                child_session_id: child_id.to_string(),
                status: "completed".to_string(),
                error: None,
                completed_at: Utc::now(),
                source: None,
            },
        )
        .await;

        let after = store.load_session(parent_id).await.unwrap().unwrap();
        assert_eq!(after.updated_at, before.updated_at);
        assert_eq!(
            serde_json::to_value(&after.messages).unwrap(),
            serde_json::to_value(&before.messages).unwrap()
        );
        assert!(read_runtime_state(&after).waiting_for_children.is_none());
        assert_eq!(inbox.inspect(parent_id).await.unwrap().pending, 0);
        assert_eq!(reservations.load(Ordering::SeqCst), 0);
        assert_eq!(launches.load(Ordering::SeqCst), 0);
    }

    #[tokio::test]
    async fn latest_locked_activation_mutation_preserves_wait_armed_after_stale_load() {
        let temp = tempfile::tempdir().unwrap();
        let store = Arc::new(
            bamboo_storage::SessionStoreV2::new(temp.path().to_path_buf())
                .await
                .unwrap(),
        );
        let storage: Arc<dyn Storage> = store.clone();
        let locked = LockedSessionStore::new(storage);
        let mut session = Session::new("activation-stale-wait", "model");
        session.agent_runtime_state = Some(AgentRuntimeState::new("old-run"));
        store.save_session(&session).await.unwrap();

        // The activation path has already loaded this stale, wait-free snapshot.
        let stale = store.load_session(&session.id).await.unwrap().unwrap();
        assert!(read_runtime_state(&stale).waiting_for_children.is_none());

        locked
            .update_runtime_config(&session.id, |latest| {
                let mut state = read_runtime_state(latest);
                state.status = AgentStatusState::Suspended;
                state.waiting_for_children = Some(WaitingForChildrenState::for_children(
                    vec!["child-new".to_string()],
                    ChildWaitPolicy::All,
                    Utc::now(),
                ));
                state.suspension = Some(SuspensionState {
                    reason: "waiting_for_children".to_string(),
                    suspended_at: Utc::now(),
                    resumable: true,
                    hook_point: None,
                });
                write_runtime_state(latest, &state);
                latest.metadata.insert(
                    "runtime.suspend_reason".to_string(),
                    "waiting_for_children".to_string(),
                );
            })
            .await
            .unwrap();

        let (prepared, ready) = prepare_session_inbox_activation(&locked, &session.id, false)
            .await
            .unwrap()
            .unwrap();
        assert!(!ready, "latest durable specific wait must block activation");
        let state = read_runtime_state(&prepared);
        assert!(state.waiting_for_children.is_some());
        assert_eq!(state.status, AgentStatusState::Suspended);
        assert_eq!(
            prepared
                .metadata
                .get("runtime.suspend_reason")
                .map(String::as_str),
            Some("waiting_for_children")
        );
    }

    #[tokio::test]
    async fn partial_child_and_bash_backlogs_remain_inert_after_inbox_reopen() {
        let temp = tempfile::tempdir().unwrap();
        let store = Arc::new(
            bamboo_storage::SessionStoreV2::new(temp.path().to_path_buf())
                .await
                .unwrap(),
        );
        store
            .save_session(&Session::new("restart-child-parent", "model"))
            .await
            .unwrap();
        store
            .save_session(&Session::new("restart-bash-parent", "model"))
            .await
            .unwrap();
        let inbox = bamboo_storage::FileSessionInbox::new(
            store.clone(),
            bamboo_domain::SessionInboxLimits::default(),
        );

        let completion = ChildCompletion {
            parent_session_id: "restart-child-parent".to_string(),
            child_session_id: "child-a".to_string(),
            status: "completed".to_string(),
            error: None,
            completed_at: Utc::now(),
            source: None,
        };
        let child_resume = runtime_resume_message(&completion, 1, Some("first child"));
        inbox
            .deliver(&child_completion_envelope(
                &completion,
                Utc::now(),
                Some("first child".to_string()),
                &child_resume,
            ))
            .await
            .unwrap();
        let bash = BashCompletionInfo {
            session_id: "restart-bash-parent".to_string(),
            bash_id: "bash-a".to_string(),
            command: "true".to_string(),
            exit_code: Some(0),
            status: "completed".to_string(),
            output_tail: String::new(),
        };
        inbox
            .deliver(&bash_completion_envelope(&bash))
            .await
            .unwrap();

        let reopened = bamboo_storage::FileSessionInbox::new(
            store,
            bamboo_domain::SessionInboxLimits::default(),
        );
        for session_id in ["restart-child-parent", "restart-bash-parent"] {
            let backlog = reopened.inspect(session_id).await.unwrap();
            assert_eq!(backlog.pending, 1);
            assert_eq!(backlog.activation_generation, 0);
            assert!(
                !backlog.activation_pending(),
                "startup must not run a partial wait backlog for {session_id}"
            );
        }
    }

    // ── ② derive completed children from the index ──────────────────────

    struct StubChildIndex {
        children: Vec<(String, Option<String>)>,
    }

    #[async_trait]
    impl Storage for StubChildIndex {
        async fn save_session(&self, _session: &Session) -> std::io::Result<()> {
            Ok(())
        }
        async fn load_session(&self, _id: &str) -> std::io::Result<Option<Session>> {
            Ok(None)
        }
        async fn delete_session(&self, _id: &str) -> std::io::Result<bool> {
            Ok(false)
        }
        async fn list_child_run_statuses(
            &self,
            _parent_session_id: &str,
        ) -> std::io::Result<Vec<(String, Option<String>)>> {
            Ok(self.children.clone())
        }
    }

    #[tokio::test]
    async fn derive_completed_only_includes_terminal_children() {
        let storage: Arc<dyn Storage> = Arc::new(StubChildIndex {
            children: vec![
                ("a".into(), Some("completed".into())),
                ("b".into(), Some("running".into())),
                ("c".into(), Some("error".into())),
                ("d".into(), None),
            ],
        });
        let completed = derive_completed_child_ids(&storage, "parent-1", "b").await;
        // Terminal from index: a, c. Plus the just-completed child b folded in.
        assert_eq!(
            completed,
            vec!["a".to_string(), "b".to_string(), "c".to_string()]
        );
    }

    #[tokio::test]
    async fn derive_completed_folds_in_just_completed_when_index_lags() {
        // Index hasn't caught up — reports the child as still running.
        let storage: Arc<dyn Storage> = Arc::new(StubChildIndex {
            children: vec![("only".into(), Some("running".into()))],
        });
        let completed = derive_completed_child_ids(&storage, "parent-1", "only").await;
        assert_eq!(completed, vec!["only".to_string()]);
    }

    #[test]
    fn wait_policy_all_uses_derived_completed_set() {
        let waited = vec!["a".to_string(), "b".to_string()];
        assert!(!wait_policy_satisfied(
            ChildWaitPolicy::All,
            &waited,
            &["a".to_string()],
            "a",
            "completed"
        ));
        assert!(wait_policy_satisfied(
            ChildWaitPolicy::All,
            &waited,
            &["a".to_string(), "b".to_string()],
            "b",
            "completed"
        ));
    }

    #[test]
    fn wait_policy_first_error_requires_tracked_membership() {
        let waited = vec!["a".to_string(), "b".to_string()];
        // An error from a TRACKED child resumes immediately.
        assert!(wait_policy_satisfied(
            ChildWaitPolicy::FirstError,
            &waited,
            &["a".to_string()],
            "a",
            "error"
        ));
        // An error-like completion from an UNTRACKED child (e.g. a zombie
        // task from an earlier run waking up late) must not resume the wait.
        assert!(!wait_policy_satisfied(
            ChildWaitPolicy::FirstError,
            &waited,
            &["a".to_string()],
            "stray-child",
            "timeout"
        ));
        // The all-complete fallback still applies regardless of the reporter.
        assert!(wait_policy_satisfied(
            ChildWaitPolicy::FirstError,
            &waited,
            &["a".to_string(), "b".to_string()],
            "stray-child",
            "completed"
        ));
    }

    #[test]
    fn child_final_assistant_text_returns_last_assistant() {
        let mut session = Session::new("child-1", "gpt-4");
        session.messages.push(Message::user("hi"));
        session
            .messages
            .push(Message::assistant("first answer", None));
        session.messages.push(Message::user("again"));
        session
            .messages
            .push(Message::assistant("final answer", None));

        assert_eq!(
            child_final_assistant_text(&session).as_deref(),
            Some("final answer")
        );
    }

    #[test]
    fn child_final_assistant_text_returns_none_when_blank() {
        let mut session = Session::new("child-1", "gpt-4");
        session.messages.push(Message::assistant("   ", None));
        assert!(child_final_assistant_text(&session).is_none());
    }

    #[test]
    fn child_final_assistant_text_returns_none_when_no_assistant() {
        let mut session = Session::new("child-1", "gpt-4");
        session.messages.push(Message::user("hi"));
        assert!(child_final_assistant_text(&session).is_none());
    }

    #[test]
    fn runtime_resume_message_folds_full_response_without_truncation() {
        // A very long child final response is folded in verbatim (no 4000-char
        // cap, no truncation marker).
        let completion = make_completion("completed");
        let long: String = "a".repeat(10_000);
        let message = runtime_resume_message(&completion, 0, Some(&long));
        assert!(message.content.contains(&long));
        assert!(!message.content.contains("truncated"));
    }

    #[test]
    fn runtime_resume_message_includes_child_response_when_provided() {
        let completion = make_completion("completed");
        let message = runtime_resume_message(&completion, 0, Some("the answer is 42"));

        assert!(matches!(message.role, Role::User));
        // Folded child results are now compressible so the parent context can
        // reclaim them under compaction.
        assert!(!message.never_compress);
        assert!(message.content.contains("Child final response:"));
        assert!(message.content.contains("the answer is 42"));

        let metadata = message.metadata.expect("metadata present");
        assert_eq!(
            metadata.get("hidden_from_ui").and_then(|v| v.as_bool()),
            Some(true)
        );
        assert_eq!(
            metadata.get("runtime_kind").and_then(|v| v.as_str()),
            Some("child_completion_resume")
        );
        assert_eq!(
            metadata
                .get("child_final_response_included")
                .and_then(|v| v.as_bool()),
            Some(true)
        );
    }

    #[test]
    fn runtime_resume_message_falls_back_to_error_when_no_response() {
        let mut completion = make_completion("error");
        completion.error = Some("boom".to_string());

        let message = runtime_resume_message(&completion, 1, None);
        assert!(message.content.contains("Child error:"));
        assert!(message.content.contains("boom"));
        let metadata = message.metadata.expect("metadata present");
        assert_eq!(
            metadata
                .get("child_final_response_included")
                .and_then(|v| v.as_bool()),
            Some(false)
        );
    }

    #[test]
    fn runtime_resume_message_minimal_when_no_response_and_no_error() {
        let completion = make_completion("completed");
        let message = runtime_resume_message(&completion, 2, None);
        assert!(!message.content.contains("Child final response:"));
        assert!(!message.content.contains("Child error:"));
        assert!(message.content.contains("Resume the parent task"));
    }

    #[test]
    fn read_config_snapshot_refreshes_cached_snapshot_from_live_config() {
        let runtime = tokio::runtime::Runtime::new().expect("runtime");

        runtime.block_on(async {
            let config = Arc::new(RwLock::new(Config::default()));
            config.write().await.provider = "copilot".to_string();
            let cached_config = StdRwLock::new(Config::default());

            let snapshot = read_config_snapshot(&config, &cached_config);

            assert_eq!(snapshot.provider, "copilot");
            assert_eq!(
                cached_config.read().expect("cached snapshot lock").provider,
                "copilot"
            );
        });
    }

    #[test]
    fn read_config_snapshot_uses_cached_snapshot_when_live_lock_is_busy() {
        let runtime = tokio::runtime::Runtime::new().expect("runtime");

        runtime.block_on(async {
            let mut cached_snapshot = Config::default();
            cached_snapshot.provider = "cached-provider".to_string();

            let config = Arc::new(RwLock::new(Config::default()));
            let cached_config = StdRwLock::new(cached_snapshot);
            let _write_guard = config.write().await;

            let snapshot = read_config_snapshot(&config, &cached_config);

            assert_eq!(snapshot.provider, "cached-provider");
        });
    }

    // ── Bash self-resume (issue #84 Phase 2b): deadline message + clobber-retry ──

    #[test]
    fn bash_completion_resume_message_normal_announces_completion() {
        let ids = vec!["bg-1".to_string(), "bg-2".to_string()];
        let message = bash_completion_resume_message(&ids, false);
        // Normal path: the shells genuinely finished.
        assert!(
            message.content.contains("have completed"),
            "normal resume message must announce completion: {}",
            message.content
        );
        // Hidden + compressible so the resume gate sees it but the UI hides it.
        let metadata = message.metadata.expect("metadata present");
        assert_eq!(
            metadata
                .get(RUNTIME_RESUME_MESSAGE_HIDDEN_KEY)
                .and_then(|v| v.as_bool()),
            Some(true),
            "resume message must be hidden from the UI"
        );
        assert_eq!(
            metadata
                .get(RUNTIME_RESUME_MESSAGE_KIND_KEY)
                .and_then(|v| v.as_str()),
            Some(BASH_COMPLETION_RESUME_KIND),
            "resume message must carry the bash-completion kind discriminant"
        );
    }

    #[test]
    fn bash_completion_resume_message_deadline_does_not_claim_completion() {
        // The 6h+10m deadline force-breaks with shells STILL running. The message
        // must NOT say "have completed" — that would let the model assume success
        // on a false premise. It must direct the model to verify with BashOutput.
        let ids = vec!["bg-long".to_string()];
        let message = bash_completion_resume_message(&ids, true);
        assert!(
            !message.content.contains("have completed"),
            "deadline resume message must NOT claim the shells completed: {}",
            message.content
        );
        assert!(
            message.content.contains("may still be running"),
            "deadline resume message must warn shells may still be running: {}",
            message.content
        );
        assert!(
            message.content.contains("BashOutput"),
            "deadline resume message must direct verification via BashOutput: {}",
            message.content
        );
        // Same hidden/kind shape so the resume gate is satisfied identically.
        let metadata = message.metadata.expect("metadata present");
        assert_eq!(
            metadata
                .get(RUNTIME_RESUME_MESSAGE_KIND_KEY)
                .and_then(|v| v.as_str()),
            Some(BASH_COMPLETION_RESUME_KIND)
        );
    }

    #[test]
    fn bash_resume_should_retry_matrix() {
        // The finalize-clobber retry predicate (issue #84 Phase 2b). Retry only
        // when the resume did NOT spawn (Completed / AlreadyRunning) AND the
        // persisted bash wait is still set on reload — the clobber signature.

        // Started: the resume fired — never retry, regardless of persisted state.
        assert!(!bash_resume_should_retry(
            &ResumeOutcome::Started { run_id: "r".into() },
            true
        ));
        assert!(!bash_resume_should_retry(
            &ResumeOutcome::Started { run_id: "r".into() },
            false
        ));

        // NotFound: session gone — never retry.
        assert!(!bash_resume_should_retry(&ResumeOutcome::NotFound, true));
        assert!(!bash_resume_should_retry(&ResumeOutcome::NotFound, false));

        // Completed + persisted wait still set ⇒ finalize-clobber ⇒ retry.
        assert!(bash_resume_should_retry(&ResumeOutcome::Completed, true));
        // Completed + persisted wait cleared ⇒ handled (our message stuck, or a
        // concurrent resume finished) ⇒ stop.
        assert!(!bash_resume_should_retry(&ResumeOutcome::Completed, false));

        // AlreadyRunning + persisted wait still set ⇒ clobbered while a runner is
        // (stale-)active ⇒ retry to re-establish the resume message.
        assert!(bash_resume_should_retry(
            &ResumeOutcome::AlreadyRunning { run_id: "r".into() },
            true
        ));
        // AlreadyRunning + wait cleared ⇒ a runner owns the session ⇒ stop.
        assert!(!bash_resume_should_retry(
            &ResumeOutcome::AlreadyRunning { run_id: "r".into() },
            false
        ));
    }

    // ── bash completion injection body (Phase 2b follow-up) ──────────────

    #[test]
    fn injection_body_includes_status_exit_command_and_tail() {
        let info = BashCompletionInfo {
            session_id: "s".into(),
            bash_id: "abc123".into(),
            command: "make build".into(),
            exit_code: Some(0),
            status: "completed".into(),
            output_tail: "BUILD OK".into(),
        };
        let body = bash_completion_injection_body(&info);
        assert!(body.contains("abc123"), "body: {body}");
        assert!(body.contains("make build"), "body: {body}");
        assert!(body.contains("completed"), "body: {body}");
        assert!(body.contains("exit code 0"), "body: {body}");
        assert!(body.contains("BUILD OK"), "body: {body}");
        // The model is pointed at BashOutput for the full log.
        assert!(body.contains("BashOutput"), "body: {body}");
        assert!(body.contains("bash_id=\"abc123\""), "body: {body}");
    }

    #[test]
    fn injection_body_handles_no_output_and_signal_kill() {
        let info = BashCompletionInfo {
            session_id: "s".into(),
            bash_id: "xyz".into(),
            command: "sleep 99".into(),
            exit_code: None,
            status: "killed".into(),
            output_tail: String::new(),
        };
        let body = bash_completion_injection_body(&info);
        assert!(body.contains("killed"), "body: {body}");
        assert!(body.contains("none (signal/killed)"), "body: {body}");
        assert!(body.contains("no captured output"), "body: {body}");
        // No output tail section when there is nothing to show.
        assert!(!body.contains("Output tail:"), "body: {body}");
    }

    #[test]
    fn background_completion_builds_post_tool_use_payload_and_feedback() {
        let mut info = BashCompletionInfo {
            session_id: "s".into(),
            bash_id: "bg-7".into(),
            command: "cargo test".into(),
            exit_code: Some(0),
            status: "completed".into(),
            output_tail: "test result: ok".into(),
        };

        let payload = background_bash_post_tool_payload(&info);
        match payload {
            HookPayload::ToolResult {
                tool_name,
                tool_call_id,
                outcome,
            } => {
                assert_eq!(tool_name, "Bash");
                assert_eq!(tool_call_id, "bg-7");
                assert!(outcome.success);
                let response: serde_json::Value =
                    serde_json::from_str(outcome.result.as_deref().unwrap()).unwrap();
                assert_eq!(response["command"], "cargo test");
                assert_eq!(response["exit_code"], 0);
                assert_eq!(response["status"], "completed");
                assert_eq!(response["output_tail"], "test result: ok");
            }
            other => panic!("expected PostToolUse payload, got {other:?}"),
        }

        append_background_bash_hook_feedback(
            &mut info,
            vec!["Run the formatter before continuing".to_string()],
        );
        assert!(info.output_tail.contains("<post_tool_use_feedback>"));
        assert!(info
            .output_tail
            .contains("Run the formatter before continuing"));
    }

    #[tokio::test]
    async fn bash_completion_payload_identity_matches_file_inbox_idempotency() {
        let baseline = BashCompletionInfo {
            session_id: "session".into(),
            bash_id: "bg-7".into(),
            command: "cargo test".into(),
            exit_code: Some(0),
            status: "completed".into(),
            output_tail: "first tail".into(),
        };
        let mut retried = baseline.clone();
        retried.output_tail = "first tail\nlater bytes\n<hook feedback>".into();
        retried.status = "completed-after-hook".into();
        let baseline_envelope = bash_completion_envelope(&baseline);
        let exact_retry = bash_completion_envelope(&baseline);
        let corrected_envelope = bash_completion_envelope(&retried);
        assert_eq!(baseline_envelope.id, exact_retry.id);
        assert_ne!(
            baseline_envelope.id, corrected_envelope.id,
            "changed payload semantics must receive a distinct id"
        );

        let mut other_shell = baseline.clone();
        other_shell.bash_id = "bg-8".into();
        assert_ne!(
            bash_completion_envelope(&baseline).id,
            bash_completion_envelope(&other_shell).id
        );

        let temp = tempfile::tempdir().unwrap();
        let store = Arc::new(
            bamboo_storage::SessionStoreV2::new(temp.path().to_path_buf())
                .await
                .unwrap(),
        );
        store
            .save_session(&Session::new("session", "model"))
            .await
            .unwrap();
        let inbox = bamboo_storage::FileSessionInbox::new(
            store,
            bamboo_domain::SessionInboxLimits::default(),
        );
        let first = inbox.deliver(&baseline_envelope).await.unwrap();
        let duplicate = inbox.deliver(&exact_retry).await.unwrap();
        let corrected = inbox.deliver(&corrected_envelope).await.unwrap();
        assert_eq!(duplicate, first, "exact payload retry is idempotent");
        assert_ne!(corrected.id, first.id);
        assert_eq!(corrected.generation, first.generation + 1);
        assert_eq!(inbox.inspect("session").await.unwrap().pending, 2);
    }

    #[cfg(unix)]
    #[tokio::test]
    async fn background_completion_fires_configured_post_tool_use_command() {
        use bamboo_config::{
            LifecycleHookGroup, LifecycleHookHandler, LifecycleHooksConfig,
            DEFAULT_LIFECYCLE_HOOK_TIMEOUT_MS,
        };

        let dir = tempfile::tempdir().unwrap();
        let output = dir.path().join("background-post-tool.json");
        let command = format!(
            "cat > '{}'; printf '%s' '{{\"additional_context\":\"inspect the completed build log\"}}'",
            output.display()
        );
        let config = LifecycleHooksConfig {
            enabled: true,
            post_tool_use: vec![LifecycleHookGroup {
                enabled: true,
                matcher: Some("^Bash$".to_string()),
                hooks: vec![LifecycleHookHandler::command(
                    command,
                    DEFAULT_LIFECYCLE_HOOK_TIMEOUT_MS,
                )],
            }],
            ..Default::default()
        };
        let mut session = Session::new("session-bg-hook", "test-model");
        session.workspace = Some(dir.path().to_string_lossy().into_owned());
        let mut info = BashCompletionInfo {
            session_id: session.id.clone(),
            bash_id: "bg-9".into(),
            command: "cargo test".into(),
            exit_code: Some(0),
            status: "completed".into(),
            output_tail: "test result: ok".into(),
        };

        assert!(run_background_bash_post_tool_hooks(&config, None, &session, &mut info).await);
        let envelope: serde_json::Value =
            serde_json::from_str(&std::fs::read_to_string(output).unwrap()).unwrap();
        assert_eq!(envelope["hook_event_name"], "PostToolUse");
        assert_eq!(envelope["tool_name"], "Bash");
        assert_eq!(envelope["payload"]["tool_call_id"], "bg-9");
        let response = envelope["tool_response"]["result"]
            .as_str()
            .map(serde_json::from_str::<serde_json::Value>)
            .transpose()
            .unwrap()
            .unwrap();
        assert_eq!(response["command"], "cargo test");
        assert_eq!(response["status"], "completed");
        assert!(info.output_tail.contains("inspect the completed build log"));
    }

    async fn temp_store() -> (tempfile::TempDir, Arc<dyn Storage>, LockedSessionStore) {
        let temp = tempfile::tempdir().unwrap();
        let storage: Arc<dyn Storage> = Arc::new(
            bamboo_storage::v2::SessionStoreV2::new(temp.path().to_path_buf())
                .await
                .expect("storage init"),
        );
        let persistence = LockedSessionStore::new(storage.clone());
        (temp, storage, persistence)
    }

    #[tokio::test]
    async fn enqueue_writes_pending_injection_and_preserves_messages() {
        let (_temp, storage, persistence) = temp_store().await;

        let mut session = Session::new("sess-enq", "test-model");
        session.add_message(Message::user("do the build"));
        storage.save_session(&session).await.unwrap();

        let info = BashCompletionInfo {
            session_id: "sess-enq".into(),
            bash_id: "sh-1".into(),
            command: "make".into(),
            exit_code: Some(0),
            status: "completed".into(),
            output_tail: "done".into(),
        };
        let saved = enqueue_bash_completion_injection(&persistence, &info)
            .await
            .expect("enqueue io ok")
            .expect("session exists");

        let pending = saved
            .pending_injected_messages()
            .expect("pending injection present");
        assert_eq!(pending.len(), 1);
        let content = pending[0].get("content").and_then(|v| v.as_str()).unwrap();
        assert!(content.contains("sh-1"), "content: {content}");
        assert!(content.contains("make"), "content: {content}");
        assert!(content.contains("done"), "content: {content}");
        // The pre-existing conversation is untouched (no clobber).
        assert_eq!(saved.messages.len(), 1);
    }

    #[tokio::test]
    async fn enqueue_returns_none_for_missing_session() {
        let (_temp, _storage, persistence) = temp_store().await;
        let info = BashCompletionInfo {
            session_id: "does-not-exist".into(),
            bash_id: "x".into(),
            command: "true".into(),
            exit_code: Some(0),
            status: "completed".into(),
            output_tail: String::new(),
        };
        let result = enqueue_bash_completion_injection(&persistence, &info)
            .await
            .expect("io ok");
        assert!(result.is_none(), "no session → nothing enqueued");
    }

    // ── push-driven resume: the state transition + decision the push applies ──

    /// A session suspended on `waiting_for_bash`, given the rich completion
    /// message, is transitioned to a resumable state: the wait is cleared, the
    /// runtime is Idle, the suspend-reason marker is gone, and the resume message
    /// is appended. This is exactly what the PUSH does to wake the loop
    /// event-driven (vs the old backstop poll).
    #[test]
    fn apply_bash_resume_transition_clears_wait_and_appends_message() {
        use bamboo_domain::session::runtime_state::WaitingForBashState;

        let mut session = Session::new("sess-resume", "test-model");
        session.add_message(Message::user("kick off the build"));
        let mut rt = read_runtime_state(&session);
        rt.status = AgentStatusState::Running;
        rt.waiting_for_bash = Some(WaitingForBashState::for_bash(
            vec!["sh-1".into()],
            Utc::now(),
        ));
        write_runtime_state(&mut session, &rt);
        session.metadata.insert(
            "runtime.suspend_reason".to_string(),
            "waiting_for_bash".to_string(),
        );

        let resume = bash_completion_resume_message(&["sh-1".to_string()], false);
        let did = apply_bash_resume_transition(&mut session, &resume);

        assert!(did, "a suspended session must transition");
        let after = read_runtime_state(&session);
        assert!(
            after.waiting_for_bash.is_none(),
            "bash wait must be cleared"
        );
        assert_eq!(after.status, AgentStatusState::Idle, "runtime must be Idle");
        assert!(
            !session.metadata.contains_key("runtime.suspend_reason"),
            "suspend-reason marker must be removed"
        );
        assert_eq!(session.messages.len(), 2, "resume message must be appended");
        assert!(matches!(
            session.messages.last().map(|m| &m.role),
            Some(Role::User)
        ));
    }

    /// The double-resume guard: a session NOT waiting on bash is a no-op — no
    /// message appended, nothing mutated. This is what makes the backstop poll
    /// harmlessly yield once the push has already resumed (and vice versa).
    #[test]
    fn apply_bash_resume_transition_noops_when_not_waiting() {
        let mut session = Session::new("sess-live", "test-model");
        session.add_message(Message::user("hi"));

        let resume = bash_completion_resume_message(&["sh-1".to_string()], false);
        let did = apply_bash_resume_transition(&mut session, &resume);

        assert!(!did, "a non-waiting session must not transition");
        assert_eq!(session.messages.len(), 1, "no resume message appended");
    }

    /// The resume invariant: push-resume fires ONLY when the loop is suspended on
    /// bash AND every waited shell has finished. A still-running sibling shell
    /// keeps it on the enqueue path.
    #[test]
    fn bash_completion_should_resume_only_when_suspended_and_all_done() {
        assert!(bash_completion_should_resume(true, true));
        assert!(!bash_completion_should_resume(true, false)); // other shells still running
        assert!(!bash_completion_should_resume(false, true)); // live loop, not suspended
        assert!(!bash_completion_should_resume(false, false));
    }

    #[test]
    fn two_shell_delivery_stages_backlog_before_one_final_activation() {
        let mut waiting = true;
        let mut durable_backlog = 0;
        let mut reservations = 0;

        // First shell: its completion is durable, but a sibling still runs.
        durable_backlog += 1;
        let first = bash_completion_delivery_plan(waiting, false);
        assert_eq!(first, BashCompletionDeliveryPlan::DurableOnly);
        assert!(waiting);
        assert_eq!(durable_backlog, 1);
        assert_eq!(reservations, 0);

        // Last shell: its own completion joins the same ordered backlog, then
        // the durable wait is cleared and exactly one activation is requested.
        durable_backlog += 1;
        let last = bash_completion_delivery_plan(waiting, true);
        assert_eq!(last, BashCompletionDeliveryPlan::ClearWaitThenActivate);
        waiting = false;
        reservations += 1;
        assert!(!waiting);
        assert_eq!(durable_backlog, 2);
        assert_eq!(reservations, 1);
    }

    /// The push's resume message carries the shell's identity + status + output
    /// tail (so the model needs no `BashOutput` round-trip) and is tagged as a
    /// bash-completion resume so it satisfies the `has_pending_user_message` gate.
    #[test]
    fn bash_resume_message_from_info_carries_bashid_tail_and_kind() {
        let info = BashCompletionInfo {
            session_id: "s".into(),
            bash_id: "sh-42".into(),
            command: "cargo test".into(),
            exit_code: Some(0),
            status: "completed".into(),
            output_tail: "test result: ok".into(),
        };
        let msg = bash_resume_message_from_info(&info);

        assert!(matches!(msg.role, Role::User));
        assert!(msg.content.contains("sh-42"), "content: {}", msg.content);
        assert!(
            msg.content.contains("cargo test"),
            "content: {}",
            msg.content
        );
        assert!(
            msg.content.contains("test result: ok"),
            "content: {}",
            msg.content
        );
        assert!(
            msg.content.contains("BashOutput"),
            "content: {}",
            msg.content
        );
        let meta = serde_json::to_string(&msg.metadata).unwrap();
        assert!(
            meta.contains(BASH_COMPLETION_RESUME_KIND),
            "resume message must be tagged as a bash-completion resume: {meta}"
        );
    }
}