anda_engine 0.14.2

//! Hook system for observing and customizing runtime behavior.
//!
//! Hooks let applications wrap agent and tool execution without changing the
//! agent or tool implementations themselves. Engine-level hooks are registered
//! through [`EngineBuilder::with_hooks`](crate::engine::EngineBuilder::with_hooks),
//! while typed tool and agent hooks are stored on [`BaseCtx`] state and consumed
//! by specific extensions.

use anda_core::{
    AgentOutput, BoxError, CacheExpiry, CacheFeatures, CancellationToken, CompletionRequest, Json,
    Resource, StateFeatures, ToolOutput,
};
use async_trait::async_trait;
use core::{fmt, str::FromStr};
use parking_lot::RwLock;
use std::{any::Any, collections::HashMap, sync::Arc, time::Duration};
use structured_logger::unix_ms;

use crate::context::{AgentCtx, BaseCtx};

/// Identifier split into a namespace prefix and local ID.
#[derive(Clone, Debug, PartialEq, Eq, Hash)]
pub struct PrefixedId {
    /// Namespace or hook prefix.
    pub prefix: String,
    /// Local identifier within the prefix.
    pub id: String,
}

impl fmt::Display for PrefixedId {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(f, "{}:{}", self.prefix, self.id)
    }
}

impl FromStr for PrefixedId {
    type Err = BoxError;

    fn from_str(s: &str) -> Result<Self, Self::Err> {
        let parts: Vec<&str> = s.splitn(2, ':').collect();
        if parts.len() != 2 {
            return Err(format!("Invalid PrefixedId format: {}", s).into());
        }
        if parts[0].trim().is_empty() || parts[1].trim().is_empty() {
            return Err(format!("Prefix and ID cannot be empty: {}", s).into());
        }
        if parts[0].trim() != parts[0] || parts[1].trim() != parts[1] {
            return Err(format!(
                "Prefix and ID cannot have leading or trailing whitespace: {}",
                s
            )
            .into());
        }
        Ok(Self {
            prefix: parts[0].to_string(),
            id: parts[1].to_string(),
        })
    }
}

/// A cloneable capability handed to background-task subscribers so they can request
/// cancellation of a single background task without tearing down the whole request.
///
/// The handle wraps a per-task [`CancellationToken`] (a child of the task context's
/// token). Calling [`stop`](BackgroundHandle::stop) cancels only that task: the
/// producer's background loop observes the cancellation and terminates the task, while
/// sibling tasks and the enclosing request keep running. Request-level cancellation still
/// cascades down, because the per-task token is a child of the request token.
///
/// The handle also carries a creation timestamp (for observing run time) and an optional,
/// type-erased application payload attached with [`with_data`](BackgroundHandle::with_data).
/// The payload is shared by [`Arc`] across every clone (the subscriber's copy and the one
/// kept in [`BackgroundTaskControls`]), so a subscriber can keep per-task bookkeeping on the
/// handle itself instead of maintaining a parallel map. Use interior mutability (e.g.
/// [`Mutex`](std::sync::Mutex)) inside the payload when it must change over the task's lifetime.
#[derive(Clone)]
pub struct BackgroundHandle {
    task_id: String,
    token: CancellationToken,
    created_at: u64,
    data: Option<Arc<dyn Any + Send + Sync>>,
}

impl BackgroundHandle {
    /// Wraps a per-task cancellation token under its task ID. The creation timestamp is
    /// stamped now and no payload is attached.
    pub fn new(task_id: impl Into<String>, token: CancellationToken) -> Self {
        Self {
            task_id: task_id.into(),
            token,
            created_at: unix_ms(),
            data: None,
        }
    }

    /// Attaches an application-defined payload, wrapping it in an [`Arc`] so every clone of
    /// the handle observes the same value.
    pub fn with_data<T: Any + Send + Sync>(mut self, data: T) -> Self {
        self.data = Some(Arc::new(data));
        self
    }

    /// Attaches an already-shared payload, so the producer can keep its own [`Arc`] clone in
    /// addition to the copy carried by the handle.
    pub fn with_shared_data<T: Any + Send + Sync>(mut self, data: Arc<T>) -> Self {
        let data: Arc<dyn Any + Send + Sync> = data;
        self.data = Some(data);
        self
    }

    /// Downcasts the attached payload to `T`, returning a shared handle to it. Returns `None`
    /// when no payload is attached or it is of another type.
    pub fn data<T: Any + Send + Sync>(&self) -> Option<Arc<T>> {
        self.data.clone()?.downcast::<T>().ok()
    }

    /// The background task's identifier.
    pub fn task_id(&self) -> &str {
        &self.task_id
    }

    /// Wall-clock time in milliseconds (Unix epoch) when this handle, and thus the task, was
    /// created.
    pub fn created_at(&self) -> u64 {
        self.created_at
    }

    /// Milliseconds elapsed since the task was created, for observing its run time.
    pub fn elapsed_ms(&self) -> u64 {
        unix_ms().saturating_sub(self.created_at)
    }

    /// Requests the producer stop this background task. Idempotent and safe to call
    /// after the task has already ended.
    pub fn stop(&self) {
        self.token.cancel();
    }

    /// Returns whether a stop has already been requested for this task.
    pub fn is_stopped(&self) -> bool {
        self.token.is_cancelled()
    }
}

/// Registry of the [`BackgroundHandle`]s for a subscriber's live background tasks, keyed by
/// task ID.
///
/// Producers register a handle when a task starts and drop it (via
/// [`finish`](BackgroundTaskControls::finish)) when the task ends; subscribers call
/// [`stop_background_task`](BackgroundTaskControls::stop_background_task) to actively
/// terminate a still-running task, and [`get`](BackgroundTaskControls::get) /
/// [`handles`](BackgroundTaskControls::handles) to read the handles (and their payloads) for
/// progress reporting. Cloning shares the same underlying registry.
///
/// Stopping a task only signals its cancellation token; the handle stays registered until
/// [`finish`](BackgroundTaskControls::finish) removes it. This lets a late progress or end
/// callback — which only carries a task ID — still find the handle and observe that the task
/// was stopped, so it can suppress duplicate output instead of forwarding it.
#[derive(Clone, Default)]
pub struct BackgroundTaskControls {
    tasks: Arc<RwLock<HashMap<String, BackgroundHandle>>>,
}

impl BackgroundTaskControls {
    /// Creates an empty registry.
    pub fn new() -> Self {
        Self::default()
    }

    /// Records a handle so the task can later be observed and stopped by ID.
    pub fn register(&self, handle: BackgroundHandle) {
        self.tasks
            .write()
            .insert(handle.task_id().to_string(), handle);
    }

    /// Returns a clone of the handle for `task_id`, if still registered. Gives callers access
    /// to the handle's application payload for progress and end bookkeeping.
    pub fn get(&self, task_id: &str) -> Option<BackgroundHandle> {
        self.tasks.read().get(task_id).cloned()
    }

    /// Downcasts the payload of the handle for `task_id` to `T` without cloning the whole
    /// handle. Returns `None` when the task is not registered or its payload is of another type.
    pub fn get_data<T: Any + Send + Sync>(&self, task_id: &str) -> Option<Arc<T>> {
        self.tasks.read().get(task_id)?.data::<T>()
    }

    /// Returns clones of every currently registered handle, for status reporting.
    pub fn handles(&self) -> Vec<BackgroundHandle> {
        self.tasks.read().values().cloned().collect()
    }

    /// Whether any task is still registered.
    pub fn is_empty(&self) -> bool {
        self.tasks.read().is_empty()
    }

    /// Forgets a task's handle once it has ended, without stopping it, returning the removed
    /// handle so a caller can read its payload one last time. This is the only path that removes
    /// a handle, so a stopped-but-not-yet-ended task stays observable to late progress and end
    /// callbacks.
    pub fn finish(&self, task_id: &str) -> Option<BackgroundHandle> {
        self.tasks.write().remove(task_id)
    }

    /// Forgets every registered handle at once, returning the removed handles so a caller can do
    /// final bookkeeping on their payloads. Like [`finish`](BackgroundTaskControls::finish) but
    /// for the whole registry; it does not stop the tasks, which a torn-down request cancels on
    /// its own through the context-token cascade.
    pub fn finish_all(&self) -> Vec<BackgroundHandle> {
        self.tasks
            .write()
            .drain()
            .map(|(_, handle)| handle)
            .collect()
    }

    /// Actively stops a single background task by ID.
    ///
    /// The handle stays registered so a later progress or end callback can still observe that
    /// the task was stopped; removal happens through [`finish`](BackgroundTaskControls::finish).
    /// Returns `true` if a task was found and signalled, `false` if no such task was registered
    /// (it already ended or never started).
    pub fn stop_background_task(&self, task_id: &str) -> bool {
        let handle = self.tasks.read().get(task_id).cloned();
        match handle {
            Some(handle) => {
                handle.stop();
                true
            }
            None => false,
        }
    }
}

/// Engine-level hook for agent runs and direct tool calls.
///
/// Returning an error from a start hook aborts execution. End hooks may inspect
/// or replace the output before it is returned to the caller.
#[async_trait]
pub trait Hook: Send + Sync {
    /// Called before an agent is executed.
    async fn on_agent_start(&self, _ctx: &AgentCtx, _agent: &str) -> Result<(), BoxError> {
        Ok(())
    }

    /// Called after an agent is executed.
    async fn on_agent_end(
        &self,
        _ctx: &AgentCtx,
        _agent: &str,
        output: AgentOutput,
    ) -> Result<AgentOutput, BoxError> {
        Ok(output)
    }

    /// Called before a tool is called.
    async fn on_tool_start(&self, _ctx: &BaseCtx, _tool: &str) -> Result<(), BoxError> {
        Ok(())
    }

    /// Called after a tool is called.
    async fn on_tool_end(
        &self,
        _ctx: &BaseCtx,
        _tool: &str,
        output: ToolOutput<Json>,
    ) -> Result<ToolOutput<Json>, BoxError> {
        Ok(output)
    }
}

/// Typed hook for a specific tool's input and output types.
///
/// Tool hooks provide fine-grained interception for extension tools. They can
/// transform arguments before execution, transform results after execution, and
/// observe background task lifecycle events.
#[async_trait]
pub trait ToolHook<I, O>: Send + Sync
where
    I: Send + Sync + 'static,
    O: Send + Sync + 'static,
{
    /// This method is called before a tool is called, allowing you to modify the input arguments.
    async fn before_tool_call(&self, _ctx: &BaseCtx, args: I) -> Result<I, BoxError> {
        Ok(args)
    }

    /// This method is called after a tool is called, allowing you to modify the output.
    async fn after_tool_call(
        &self,
        _ctx: &BaseCtx,
        output: ToolOutput<O>,
    ) -> Result<ToolOutput<O>, BoxError> {
        Ok(output)
    }

    /// Called when a tool starts a background task.
    ///
    /// `handle` lets the subscriber later stop this specific task via
    /// [`BackgroundHandle::stop`]; its [`task_id`](BackgroundHandle::task_id)
    /// correlates the progress and end callbacks.
    async fn on_background_start(&self, _ctx: &BaseCtx, _handle: BackgroundHandle, _args: &I) {}

    /// Called when a tool reports progress from a background task.
    async fn on_background_progress(
        &self,
        _ctx: &BaseCtx,
        _task_id: String,
        _output: ToolOutput<O>,
    ) {
    }

    /// Called with the final output from a background tool task.
    async fn on_background_end(&self, _ctx: &BaseCtx, _task_id: String, _output: ToolOutput<O>) {}
}

/// Type-erased background hook used when a tool's typed arguments are not available.
#[async_trait]
pub trait ToolBackgroundHook: Send + Sync {
    /// Called when a tool starts a background task.
    ///
    /// `handle` lets the subscriber later stop this specific task via
    /// [`BackgroundHandle::stop`]; its [`task_id`](BackgroundHandle::task_id)
    /// correlates the progress and end callbacks.
    async fn on_background_start(&self, _ctx: &BaseCtx, _handle: BackgroundHandle, _args: Json) {}

    /// Called when a tool reports progress from a background task.
    async fn on_background_progress(
        &self,
        _ctx: &BaseCtx,
        _task_id: String,
        _output: ToolOutput<Json>,
    ) {
    }

    /// Called with the final output from a background tool task.
    async fn on_background_end(&self, _ctx: &BaseCtx, _task_id: String, _output: ToolOutput<Json>) {
    }
}

/// Cloneable type-erased wrapper for a typed [`ToolHook`].
#[derive(Clone)]
pub struct DynToolHook<I, O> {
    inner: Arc<dyn ToolHook<I, O>>,
}

impl<I, O> DynToolHook<I, O>
where
    I: Send + Sync + 'static,
    O: Send + Sync + 'static,
{
    /// Wraps a concrete hook implementation.
    pub fn new(inner: Arc<dyn ToolHook<I, O>>) -> Self {
        Self { inner }
    }
}

#[async_trait]
impl<I, O> ToolHook<I, O> for DynToolHook<I, O>
where
    I: Send + Sync + 'static,
    O: Send + Sync + 'static,
{
    async fn before_tool_call(&self, ctx: &BaseCtx, args: I) -> Result<I, BoxError> {
        self.inner.before_tool_call(ctx, args).await
    }

    async fn after_tool_call(
        &self,
        ctx: &BaseCtx,
        output: ToolOutput<O>,
    ) -> Result<ToolOutput<O>, BoxError> {
        self.inner.after_tool_call(ctx, output).await
    }

    async fn on_background_start(&self, ctx: &BaseCtx, handle: BackgroundHandle, args: &I) {
        self.inner.on_background_start(ctx, handle, args).await;
    }

    async fn on_background_progress(&self, ctx: &BaseCtx, task_id: String, _output: ToolOutput<O>) {
        self.inner
            .on_background_progress(ctx, task_id, _output)
            .await;
    }

    async fn on_background_end(&self, ctx: &BaseCtx, task_id: String, output: ToolOutput<O>) {
        self.inner.on_background_end(ctx, task_id, output).await;
    }
}

/// Type-erased wrapper for [`ToolBackgroundHook`].
#[derive(Clone)]
pub struct DynToolJsonHook {
    inner: Arc<dyn ToolBackgroundHook>,
}

impl DynToolJsonHook {
    /// Wraps a concrete hook implementation.
    pub fn new(inner: Arc<dyn ToolBackgroundHook>) -> Self {
        Self { inner }
    }
}

#[async_trait]
impl ToolBackgroundHook for DynToolJsonHook {
    async fn on_background_start(&self, ctx: &BaseCtx, handle: BackgroundHandle, args: Json) {
        self.inner.on_background_start(ctx, handle, args).await;
    }

    async fn on_background_progress(
        &self,
        ctx: &BaseCtx,
        task_id: String,
        _output: ToolOutput<Json>,
    ) {
        self.inner
            .on_background_progress(ctx, task_id, _output)
            .await;
    }

    async fn on_background_end(&self, ctx: &BaseCtx, task_id: String, output: ToolOutput<Json>) {
        self.inner.on_background_end(ctx, task_id, output).await;
    }
}

/// Typed hook for nested agent calls and agent-runner extensions.
#[async_trait]
pub trait AgentHook: Send + Sync {
    /// Called before an agent is executed, allowing you to modify the prompt and resources.
    async fn before_agent_run(
        &self,
        _ctx: &AgentCtx,
        prompt: String,
        resources: Vec<Resource>,
    ) -> Result<(String, Vec<Resource>), BoxError> {
        Ok((prompt, resources))
    }

    /// Called after an agent is executed, allowing you to modify the output.
    ///
    /// For background execution, implementations can use
    /// [`AgentHook::on_background_end`] to observe the final result.
    async fn after_agent_run(
        &self,
        _ctx: &AgentCtx,
        output: AgentOutput,
    ) -> Result<AgentOutput, BoxError> {
        Ok(output)
    }

    /// Called when an agent starts in the background.
    ///
    /// `handle` lets the subscriber later stop this specific session via
    /// [`BackgroundHandle::stop`]; its [`task_id`](BackgroundHandle::task_id) is
    /// the session ID and correlates the progress and final output hooks.
    async fn on_background_start(
        &self,
        _ctx: &AgentCtx,
        _handle: BackgroundHandle,
        _req: &CompletionRequest,
    ) {
    }

    /// Called when an agent reports progress from a background task.
    async fn on_background_progress(
        &self,
        _ctx: &AgentCtx,
        _session_id: String,
        _progress: AgentOutput,
    ) {
    }

    /// Called with the final output from a background agent task.
    async fn on_background_end(&self, _ctx: &AgentCtx, _session_id: String, _output: AgentOutput) {}
}

/// Cloneable type-erased wrapper for an [`AgentHook`].
#[derive(Clone)]
pub struct DynAgentHook {
    inner: Arc<dyn AgentHook>,
}

impl DynAgentHook {
    /// Wraps a concrete agent hook implementation.
    pub fn new(inner: Arc<dyn AgentHook>) -> Self {
        Self { inner }
    }
}

#[async_trait]
impl AgentHook for DynAgentHook {
    async fn before_agent_run(
        &self,
        ctx: &AgentCtx,
        prompt: String,
        resources: Vec<Resource>,
    ) -> Result<(String, Vec<Resource>), BoxError> {
        self.inner.before_agent_run(ctx, prompt, resources).await
    }

    async fn after_agent_run(
        &self,
        ctx: &AgentCtx,
        output: AgentOutput,
    ) -> Result<AgentOutput, BoxError> {
        self.inner.after_agent_run(ctx, output).await
    }

    async fn on_background_start(
        &self,
        ctx: &AgentCtx,
        handle: BackgroundHandle,
        req: &CompletionRequest,
    ) {
        self.inner.on_background_start(ctx, handle, req).await;
    }

    async fn on_background_progress(
        &self,
        ctx: &AgentCtx,
        session_id: String,
        progress: AgentOutput,
    ) {
        self.inner
            .on_background_progress(ctx, session_id, progress)
            .await;
    }

    async fn on_background_end(&self, ctx: &AgentCtx, session_id: String, output: AgentOutput) {
        self.inner.on_background_end(ctx, session_id, output).await;
    }
}

/// Ordered collection of engine-level hooks.
///
/// Hooks run in insertion order. End hooks receive the output from the previous
/// hook, so each hook can transform the value before the next one observes it.
pub struct Hooks {
    hooks: Vec<Box<dyn Hook>>,
}

impl Default for Hooks {
    fn default() -> Self {
        Self::new()
    }
}

impl Hooks {
    /// Creates an empty hook collection.
    pub fn new() -> Self {
        Self { hooks: Vec::new() }
    }

    /// Adds a hook to the end of the collection.
    pub fn add(&mut self, hook: Box<dyn Hook>) {
        self.hooks.push(hook);
    }
}

#[async_trait]
impl Hook for Hooks {
    async fn on_agent_start(&self, ctx: &AgentCtx, agent: &str) -> Result<(), BoxError> {
        for hook in &self.hooks {
            hook.on_agent_start(ctx, agent).await?;
        }
        Ok(())
    }

    async fn on_agent_end(
        &self,
        ctx: &AgentCtx,
        agent: &str,
        mut output: AgentOutput,
    ) -> Result<AgentOutput, BoxError> {
        for hook in &self.hooks {
            output = hook.on_agent_end(ctx, agent, output).await?;
        }
        Ok(output)
    }

    async fn on_tool_start(&self, ctx: &BaseCtx, tool: &str) -> Result<(), BoxError> {
        for hook in &self.hooks {
            hook.on_tool_start(ctx, tool).await?;
        }
        Ok(())
    }

    async fn on_tool_end(
        &self,
        ctx: &BaseCtx,
        tool: &str,
        mut output: ToolOutput<Json>,
    ) -> Result<ToolOutput<Json>, BoxError> {
        for hook in &self.hooks {
            output = hook.on_tool_end(ctx, tool, output).await?;
        }
        Ok(output)
    }
}

/// Hook that limits each caller to one active agent prompt at a time.
///
/// The hook stores a per-caller lease in the context cache when an agent starts
/// and removes it when the agent ends. The TTL prevents stale leases from
/// blocking a caller forever after an interrupted process.
pub struct SingleThreadHook {
    ttl: Duration,
}

impl SingleThreadHook {
    /// Creates a single-thread hook with the lease time-to-live.
    pub fn new(ttl: Duration) -> Self {
        Self { ttl }
    }
}

#[async_trait]
impl Hook for SingleThreadHook {
    async fn on_agent_start(&self, ctx: &AgentCtx, _agent: &str) -> Result<(), BoxError> {
        let caller = ctx.caller();
        let now_ms = unix_ms();
        let ok = ctx
            .cache_set_if_not_exists(
                caller.to_string().as_str(),
                (now_ms, Some(CacheExpiry::TTL(self.ttl))),
            )
            .await;
        if !ok {
            return Err("Only one prompt can run at a time.".into());
        }
        Ok(())
    }

    async fn on_agent_end(
        &self,
        ctx: &AgentCtx,
        _agent: &str,
        output: AgentOutput,
    ) -> Result<AgentOutput, BoxError> {
        let caller = ctx.caller();
        ctx.cache_delete(caller.to_string().as_str()).await;
        Ok(output)
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use anda_core::{AgentSet, CancellationToken, ToolProviderSet, ToolSet};
    use parking_lot::Mutex;

    use crate::{
        context::{RemoteEngines, Web3SDK},
        model::Models,
        store::{InMemory, Store},
        subagent::SubAgentSetManager,
    };

    fn base_ctx() -> BaseCtx {
        BaseCtx::new(
            candid::Principal::self_authenticating([9; 32]),
            "engine".to_string(),
            "agent".to_string(),
            CancellationToken::new(),
            std::collections::BTreeSet::from([anda_core::Path::default()]),
            Arc::new(Web3SDK::not_implemented()),
            Store::new(Arc::new(InMemory::new())),
            Arc::new(RemoteEngines::new()),
        )
    }

    fn agent_ctx() -> AgentCtx {
        AgentCtx::new(
            base_ctx(),
            Arc::new(Models::default()),
            Arc::new(ToolSet::new()),
            Arc::new(ToolProviderSet::new()),
            Arc::new(AgentSet::new()),
            Arc::new(SubAgentSetManager::new()),
        )
    }

    #[test]
    fn background_handle_cancels_only_its_token() {
        let parent = CancellationToken::new();
        let task_token = parent.child_token();
        let handle = BackgroundHandle::new("shell:1", task_token.clone());

        assert_eq!(handle.task_id(), "shell:1");
        assert!(!handle.is_stopped());
        assert!(!task_token.is_cancelled());

        handle.stop();
        assert!(handle.is_stopped());
        assert!(task_token.is_cancelled());
        // Stopping one task must not cancel the enclosing request token.
        assert!(!parent.is_cancelled());
        // Idempotent.
        handle.stop();
    }

    #[test]
    fn background_handle_carries_payload_and_created_at() {
        let handle =
            BackgroundHandle::new("shell:1", CancellationToken::new()).with_data(Mutex::new(7u32));

        // The creation timestamp is stamped and elapsed time is observable.
        assert!(handle.created_at() > 0);
        let _ = handle.elapsed_ms();

        // The payload is shared across clones: a mutation through one is seen through the other.
        let clone = handle.clone();
        *handle.data::<Mutex<u32>>().unwrap().lock() = 42;
        assert_eq!(*clone.data::<Mutex<u32>>().unwrap().lock(), 42);

        // Downcasting to the wrong type, or reading a handle without payload, returns None.
        assert!(handle.data::<Mutex<String>>().is_none());
        let bare = BackgroundHandle::new("shell:2", CancellationToken::new());
        assert!(bare.data::<Mutex<u32>>().is_none());
    }

    #[test]
    fn background_task_controls_stop_finish_and_finish_all() {
        let controls = BackgroundTaskControls::new();
        let token_a = CancellationToken::new();
        let token_b = CancellationToken::new();
        controls.register(BackgroundHandle::new("a", token_a.clone()));
        controls.register(BackgroundHandle::new("b", token_b.clone()));

        // Stopping a live task signals its token and reports it was found, but keeps the handle
        // registered so a late progress/end callback can still observe the stop.
        assert!(controls.stop_background_task("a"));
        assert!(token_a.is_cancelled());
        assert!(controls.get("a").is_some());
        // Stop is idempotent and still reports the task as found until it is finished.
        assert!(controls.stop_background_task("a"));
        // Unknown task ids report not found.
        assert!(!controls.stop_background_task("missing"));

        // finish() forgets a handle and returns it, without cancelling.
        assert_eq!(controls.finish("a").unwrap().task_id(), "a");
        assert!(controls.get("a").is_none());
        assert!(!controls.stop_background_task("a"));

        controls.finish("b");
        assert!(!token_b.is_cancelled());
        assert!(!controls.stop_background_task("b"));

        // finish_all() drains every remaining handle without stopping the tasks.
        let token_c = CancellationToken::new();
        controls.register(BackgroundHandle::new("c", token_c.clone()));
        assert!(!controls.is_empty());
        let drained = controls.finish_all();
        assert_eq!(drained.len(), 1);
        assert_eq!(drained[0].task_id(), "c");
        assert!(!token_c.is_cancelled());
        assert!(controls.is_empty());
    }

    #[test]
    fn prefixed_id_display_and_parse_validate_shape() {
        let id: PrefixedId = "agent:run-1".parse().unwrap();
        assert_eq!(id.prefix, "agent");
        assert_eq!(id.id, "run-1");
        assert_eq!(id.to_string(), "agent:run-1");

        assert!("missing_separator".parse::<PrefixedId>().is_err());
        assert!(":empty_prefix".parse::<PrefixedId>().is_err());
        assert!("empty_id:".parse::<PrefixedId>().is_err());
        assert!(" agent:run".parse::<PrefixedId>().is_err());
        assert!("agent:run ".parse::<PrefixedId>().is_err());
    }

    struct AppendHook(&'static str);

    #[async_trait]
    impl Hook for AppendHook {
        async fn on_agent_start(&self, _ctx: &AgentCtx, agent: &str) -> Result<(), BoxError> {
            assert!(!agent.is_empty());
            Ok(())
        }

        async fn on_agent_end(
            &self,
            _ctx: &AgentCtx,
            _agent: &str,
            mut output: AgentOutput,
        ) -> Result<AgentOutput, BoxError> {
            output.content.push_str(self.0);
            Ok(output)
        }

        async fn on_tool_start(&self, _ctx: &BaseCtx, tool: &str) -> Result<(), BoxError> {
            assert!(!tool.is_empty());
            Ok(())
        }

        async fn on_tool_end(
            &self,
            _ctx: &BaseCtx,
            _tool: &str,
            mut output: ToolOutput<Json>,
        ) -> Result<ToolOutput<Json>, BoxError> {
            let current = output.output.as_str().unwrap_or_default().to_string();
            output.output = Json::String(format!("{current}{}", self.0));
            Ok(output)
        }
    }

    #[tokio::test(flavor = "current_thread")]
    async fn hooks_apply_start_checks_and_end_transforms_in_order() {
        let agent_ctx = agent_ctx();
        let base_ctx = base_ctx();
        let mut hooks = Hooks::new();
        hooks.add(Box::new(AppendHook("-one")));
        hooks.add(Box::new(AppendHook("-two")));

        hooks.on_agent_start(&agent_ctx, "worker").await.unwrap();
        let output = hooks
            .on_agent_end(
                &agent_ctx,
                "worker",
                AgentOutput {
                    content: "done".to_string(),
                    ..Default::default()
                },
            )
            .await
            .unwrap();
        assert_eq!(output.content, "done-one-two");

        hooks.on_tool_start(&base_ctx, "tool").await.unwrap();
        let output = hooks
            .on_tool_end(
                &base_ctx,
                "tool",
                ToolOutput::new(Json::String("ok".to_string())),
            )
            .await
            .unwrap();
        assert_eq!(output.output, Json::String("ok-one-two".to_string()));
    }

    struct PrefixToolHook {
        events: Arc<Mutex<Vec<String>>>,
    }

    #[async_trait]
    impl ToolHook<String, String> for PrefixToolHook {
        async fn before_tool_call(&self, _ctx: &BaseCtx, args: String) -> Result<String, BoxError> {
            Ok(format!("before:{args}"))
        }

        async fn after_tool_call(
            &self,
            _ctx: &BaseCtx,
            mut output: ToolOutput<String>,
        ) -> Result<ToolOutput<String>, BoxError> {
            output.output = format!("after:{}", output.output);
            Ok(output)
        }

        async fn on_background_start(
            &self,
            _ctx: &BaseCtx,
            handle: BackgroundHandle,
            _args: &String,
        ) {
            self.events
                .lock()
                .push(format!("start:{}", handle.task_id()));
        }

        async fn on_background_progress(
            &self,
            _ctx: &BaseCtx,
            task_id: String,
            _output: ToolOutput<String>,
        ) {
            self.events.lock().push(format!("progress:{task_id}"));
        }

        async fn on_background_end(
            &self,
            _ctx: &BaseCtx,
            task_id: String,
            _output: ToolOutput<String>,
        ) {
            self.events.lock().push(format!("end:{task_id}"));
        }
    }

    struct JsonBackgroundHook {
        events: Arc<Mutex<Vec<String>>>,
    }

    #[async_trait]
    impl ToolBackgroundHook for JsonBackgroundHook {
        async fn on_background_start(&self, _ctx: &BaseCtx, handle: BackgroundHandle, _args: Json) {
            self.events
                .lock()
                .push(format!("json-start:{}", handle.task_id()));
        }

        async fn on_background_progress(
            &self,
            _ctx: &BaseCtx,
            task_id: String,
            _output: ToolOutput<Json>,
        ) {
            self.events.lock().push(format!("json-progress:{task_id}"));
        }

        async fn on_background_end(
            &self,
            _ctx: &BaseCtx,
            task_id: String,
            _output: ToolOutput<Json>,
        ) {
            self.events.lock().push(format!("json-end:{task_id}"));
        }
    }

    #[tokio::test(flavor = "current_thread")]
    async fn dynamic_tool_hooks_forward_typed_and_json_background_calls() {
        let ctx = base_ctx();
        let events = Arc::new(Mutex::new(Vec::new()));
        let hook = DynToolHook::new(Arc::new(PrefixToolHook {
            events: events.clone(),
        }));

        let args = hook
            .before_tool_call(&ctx, "input".to_string())
            .await
            .unwrap();
        assert_eq!(args, "before:input");
        let output = hook
            .after_tool_call(&ctx, ToolOutput::new("value".to_string()))
            .await
            .unwrap();
        assert_eq!(output.output, "after:value");
        hook.on_background_start(
            &ctx,
            BackgroundHandle::new("task", CancellationToken::new()),
            &"args".to_string(),
        )
        .await;
        hook.on_background_progress(&ctx, "task".to_string(), ToolOutput::new("p".to_string()))
            .await;
        hook.on_background_end(&ctx, "task".to_string(), ToolOutput::new("e".to_string()))
            .await;
        assert_eq!(
            events.lock().clone(),
            vec!["start:task", "progress:task", "end:task"]
        );

        let json_events = Arc::new(Mutex::new(Vec::new()));
        let json_hook = DynToolJsonHook::new(Arc::new(JsonBackgroundHook {
            events: json_events.clone(),
        }));
        json_hook
            .on_background_start(
                &ctx,
                BackgroundHandle::new("json-task", CancellationToken::new()),
                Json::Null,
            )
            .await;
        json_hook
            .on_background_progress(&ctx, "json-task".to_string(), ToolOutput::new(Json::Null))
            .await;
        json_hook
            .on_background_end(&ctx, "json-task".to_string(), ToolOutput::new(Json::Null))
            .await;
        assert_eq!(
            json_events.lock().clone(),
            vec![
                "json-start:json-task",
                "json-progress:json-task",
                "json-end:json-task"
            ]
        );
    }

    struct PrefixAgentHook {
        events: Arc<Mutex<Vec<String>>>,
    }

    #[async_trait]
    impl AgentHook for PrefixAgentHook {
        async fn before_agent_run(
            &self,
            _ctx: &AgentCtx,
            prompt: String,
            mut resources: Vec<Resource>,
        ) -> Result<(String, Vec<Resource>), BoxError> {
            resources.push(Resource {
                name: "added".to_string(),
                tags: vec!["text".to_string()],
                ..Default::default()
            });
            Ok((format!("before:{prompt}"), resources))
        }

        async fn after_agent_run(
            &self,
            _ctx: &AgentCtx,
            mut output: AgentOutput,
        ) -> Result<AgentOutput, BoxError> {
            output.content = format!("after:{}", output.content);
            Ok(output)
        }

        async fn on_background_start(
            &self,
            _ctx: &AgentCtx,
            handle: BackgroundHandle,
            _req: &CompletionRequest,
        ) {
            self.events
                .lock()
                .push(format!("agent-start:{}", handle.task_id()));
        }

        async fn on_background_progress(
            &self,
            _ctx: &AgentCtx,
            session_id: String,
            _progress: AgentOutput,
        ) {
            self.events
                .lock()
                .push(format!("agent-progress:{session_id}"));
        }

        async fn on_background_end(
            &self,
            _ctx: &AgentCtx,
            session_id: String,
            _output: AgentOutput,
        ) {
            self.events.lock().push(format!("agent-end:{session_id}"));
        }
    }

    #[tokio::test(flavor = "current_thread")]
    async fn dynamic_agent_hook_forwards_prompt_output_and_background_calls() {
        let ctx = agent_ctx();
        let events = Arc::new(Mutex::new(Vec::new()));
        let hook = DynAgentHook::new(Arc::new(PrefixAgentHook {
            events: events.clone(),
        }));
        let (prompt, resources) = hook
            .before_agent_run(&ctx, "input".to_string(), Vec::new())
            .await
            .unwrap();
        assert_eq!(prompt, "before:input");
        assert_eq!(resources.len(), 1);

        let output = hook
            .after_agent_run(
                &ctx,
                AgentOutput {
                    content: "value".to_string(),
                    ..Default::default()
                },
            )
            .await
            .unwrap();
        assert_eq!(output.content, "after:value");
        hook.on_background_start(
            &ctx,
            BackgroundHandle::new("session", CancellationToken::new()),
            &CompletionRequest::default(),
        )
        .await;
        hook.on_background_progress(&ctx, "session".to_string(), AgentOutput::default())
            .await;
        hook.on_background_end(&ctx, "session".to_string(), AgentOutput::default())
            .await;
        assert_eq!(
            events.lock().clone(),
            vec![
                "agent-start:session",
                "agent-progress:session",
                "agent-end:session"
            ]
        );
    }

    #[tokio::test(flavor = "current_thread")]
    async fn single_thread_hook_rejects_second_prompt_until_end_releases_lease() {
        let ctx = agent_ctx().with_caller(candid::Principal::self_authenticating([4; 32]));
        let hook = SingleThreadHook::new(Duration::from_secs(30));

        hook.on_agent_start(&ctx, "agent").await.unwrap();
        assert!(
            hook.on_agent_start(&ctx, "agent")
                .await
                .unwrap_err()
                .to_string()
                .contains("Only one prompt")
        );
        let output = hook
            .on_agent_end(
                &ctx,
                "agent",
                AgentOutput {
                    content: "done".to_string(),
                    ..Default::default()
                },
            )
            .await
            .unwrap();
        assert_eq!(output.content, "done");
        hook.on_agent_start(&ctx, "agent").await.unwrap();
    }
}