eidetic-engine 0.18.0

//! Command effect manifest (EE-TST-009).
//!
//! Defines a first-class taxonomy of what each public command may do,
//! enabling agents to mechanically choose safe commands and proving
//! that read-only commands stay read-only.
//!
//! Effect classes:
//! - `ReadOnly` — no durable mutation, safe to call mid-task
//! - `DerivedArtifactWrite` — writes rebuildable indexes/caches
//! - `DurableMemoryWrite` — writes memories, audit records
//! - `WorkspaceFileWrite` — writes workspace files beyond DB
//! - `ConfigWrite` — modifies configuration
//! - `ExternalIo` — network or subprocess I/O
//!
//! The manifest maps each command path to its default effect, dry-run
//! effect, allowed write surfaces, and idempotency posture.

use std::collections::HashMap;

/// Side-effect class names shared with the command-boundary matrix.
#[derive(Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd, Hash)]
pub enum SideEffectClass {
    /// No DB, index, cache, or filesystem mutation.
    ReadOnly,
    /// Current command is read-only or degraded; future writes need a new audited class.
    ReadOnlyNow,
    /// Report computed from explicit inputs without durable mutation.
    ReportOnly,
    /// Static reads are allowed; missing or judgment-heavy work degrades or hands off.
    ReadOnlyOrUnavailable,
    /// Append new records, or return an existing record by idempotency key.
    AppendOnly,
    /// Durable mutation in one audited transaction.
    AuditedMutation,
    /// Rebuild only derived, rebuildable assets keyed by source generation.
    DerivedAssetRebuild,
    /// Create or verify a side-path artifact without overwriting source data.
    SidePathArtifact,
    /// Long-running job mutation through a supervised job ledger.
    SupervisedJobs,
    /// Family contains both read-only and mutating subcommands.
    Mixed,
    /// No mutation until the real implementation exists.
    DegradedUnavailable,
    /// Read-only extraction today; future candidate writes require an explicit append path.
    ReportOnlyOrAppend,
    /// Read-only reports today; relation writes require an explicit audited transaction.
    ReportOnlyOrAuditedMutation,
}

impl SideEffectClass {
    /// Stable vocabulary token used in docs, JSON logs, and tests.
    #[must_use]
    pub const fn as_str(self) -> &'static str {
        match self {
            Self::ReadOnly => "class=read_only",
            Self::ReadOnlyNow => "class=read_only_now",
            Self::ReportOnly => "class=report_only",
            Self::ReadOnlyOrUnavailable => "class=read_only_or_unavailable",
            Self::AppendOnly => "class=append_only",
            Self::AuditedMutation => "class=audited_mutation",
            Self::DerivedAssetRebuild => "class=derived_asset_rebuild",
            Self::SidePathArtifact => "class=side_path_artifact",
            Self::SupervisedJobs => "class=supervised_jobs",
            Self::Mixed => "class=mixed",
            Self::DegradedUnavailable => "class=degraded_unavailable",
            Self::ReportOnlyOrAppend => "class=report_only_or_append",
            Self::ReportOnlyOrAuditedMutation => "class=report_only_or_audited_mutation",
        }
    }

    /// `true` if this class forbids durable mutation.
    #[must_use]
    pub const fn declares_no_durable_mutation(self) -> bool {
        matches!(
            self,
            Self::ReadOnly
                | Self::ReadOnlyNow
                | Self::ReportOnly
                | Self::ReadOnlyOrUnavailable
                | Self::DegradedUnavailable
        )
    }

    /// `true` if this class must carry no-overwrite side-path behavior.
    #[must_use]
    pub const fn requires_no_overwrite_contract(self) -> bool {
        matches!(self, Self::SidePathArtifact)
    }

    /// `true` if this class must carry transaction/audit metadata.
    #[must_use]
    pub const fn requires_audited_transaction_contract(self) -> bool {
        matches!(self, Self::AppendOnly | Self::AuditedMutation)
    }
}

/// Effect class describing what a command may mutate.
#[derive(Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd, Hash)]
pub enum EffectClass {
    /// No durable mutation. Safe to call at any time.
    ReadOnly,
    /// Writes derived artifacts (indexes, caches) that can be rebuilt.
    DerivedArtifactWrite,
    /// Writes durable memory records, audit log, or user-visible state.
    DurableMemoryWrite,
    /// Writes files in the workspace beyond the database.
    WorkspaceFileWrite,
    /// Modifies configuration (ee.toml, workspace config).
    ConfigWrite,
    /// Performs external I/O (network, subprocess).
    ExternalIo,
}

impl EffectClass {
    /// Stable string for JSON serialization and logs.
    #[must_use]
    pub const fn as_str(self) -> &'static str {
        match self {
            Self::ReadOnly => "read_only",
            Self::DerivedArtifactWrite => "derived_artifact_write",
            Self::DurableMemoryWrite => "durable_memory_write",
            Self::WorkspaceFileWrite => "workspace_file_write",
            Self::ConfigWrite => "config_write",
            Self::ExternalIo => "external_io",
        }
    }

    /// `true` if this effect class mutates durable user-visible state.
    #[must_use]
    pub const fn is_mutating(self) -> bool {
        !matches!(self, Self::ReadOnly)
    }

    /// `true` if mutations are rebuildable (indexes, caches).
    #[must_use]
    pub const fn is_derived(self) -> bool {
        matches!(self, Self::DerivedArtifactWrite)
    }
}

/// Cross-cutting mutation contract for a command manifest entry.
#[derive(Clone, Debug, Eq, PartialEq)]
pub struct CommandMutationContract {
    /// Side-effect class from the command-boundary matrix vocabulary.
    pub side_effect_class: SideEffectClass,
    /// Named transaction scope, if the command mutates durable state.
    pub transaction_scope: Option<&'static str>,
    /// Idempotency key or retry posture.
    pub idempotency_key: Option<&'static str>,
    /// Audit surface written by the command, if any.
    pub audit_surface: Option<&'static str>,
    /// Effect on source database generation.
    pub db_generation_effect: &'static str,
    /// Effect on derived index or cache generation.
    pub index_generation_effect: &'static str,
    /// Dry-run or preview behavior, if exposed by this class.
    ///
    /// THIS IS NOT A `--dry-run` FLAG CLAIM AND MUST NOT BE READ AS ONE
    /// (bd-3j6l3). It describes previewability BY ANY MECHANISM, and the
    /// mechanism is frequently not `--dry-run`: `harness_hook_settings_write`
    /// documents `--print`, `certificate_key_file_write` documents `--show`,
    /// `ask` documents `--read-only`, and `health scorecard
    /// --record-snapshot` documents omitting the flag. For the narrow
    /// question "does the parser accept `--dry-run`", read
    /// `CommandEffect::dry_run_effect`.
    ///
    /// The two fields are therefore deliberately NOT required to agree. Only
    /// one implication is enforced: a command declaring `dry_run_effect:
    /// Some(_)` must also describe what that preview does. Prose without a
    /// flag is allowed and is how a non-`--dry-run` preview path is recorded.
    pub dry_run_behavior: Option<&'static str>,
    /// Recovery, rollback, or degraded behavior.
    pub recovery_behavior: &'static str,
    /// Side-path no-overwrite/no-delete behavior, when applicable.
    pub no_overwrite_behavior: Option<&'static str>,
    /// Degraded/error code returned when the command intentionally abstains.
    pub degraded_code: Option<&'static str>,
}

impl CommandMutationContract {
    #[must_use]
    pub const fn read_only() -> Self {
        Self {
            side_effect_class: SideEffectClass::ReadOnly,
            transaction_scope: None,
            idempotency_key: Some("full command argv and explicit inputs"),
            audit_surface: None,
            db_generation_effect: "none",
            index_generation_effect: "none",
            dry_run_behavior: None,
            recovery_behavior: "no durable changes to recover",
            no_overwrite_behavior: None,
            degraded_code: None,
        }
    }

    #[must_use]
    pub const fn derived_asset_rebuild(
        idempotency_key: &'static str,
        recovery_behavior: &'static str,
    ) -> Self {
        Self {
            side_effect_class: SideEffectClass::DerivedAssetRebuild,
            transaction_scope: Some("derived asset rebuild keyed by source generation"),
            idempotency_key: Some(idempotency_key),
            audit_surface: None,
            db_generation_effect: "source DB generation unchanged",
            index_generation_effect: "derived generation may advance to source generation",
            dry_run_behavior: Some("preview only; no derived files are written"),
            recovery_behavior,
            no_overwrite_behavior: None,
            degraded_code: None,
        }
    }

    #[must_use]
    pub const fn audited_mutation(idempotency_key: &'static str) -> Self {
        Self {
            side_effect_class: SideEffectClass::AuditedMutation,
            transaction_scope: Some("single DB transaction across write surfaces"),
            idempotency_key: Some(idempotency_key),
            audit_surface: Some("audit_log"),
            db_generation_effect: "advances on commit; unchanged on rollback",
            index_generation_effect: "queues or refreshes derived index after commit when applicable",
            dry_run_behavior: Some("no DB rows, audit rows, or derived index jobs are written"),
            recovery_behavior: "transaction rollback leaves no partial durable records",
            no_overwrite_behavior: None,
            degraded_code: None,
        }
    }

    #[must_use]
    pub const fn append_only(idempotency_key: &'static str) -> Self {
        Self {
            side_effect_class: SideEffectClass::AppendOnly,
            transaction_scope: Some("single append transaction across write surfaces"),
            idempotency_key: Some(idempotency_key),
            audit_surface: Some("audit_log"),
            db_generation_effect: "advances only when a new record commits; unchanged when idempotency key matches",
            index_generation_effect: "queues or refreshes derived index after new records commit",
            dry_run_behavior: Some("no DB rows, audit rows, or derived index jobs are written"),
            recovery_behavior: "transaction rollback leaves no partial append records",
            no_overwrite_behavior: None,
            degraded_code: None,
        }
    }

    #[must_use]
    pub const fn side_path_artifact(
        idempotency_key: &'static str,
        no_overwrite_behavior: &'static str,
    ) -> Self {
        Self {
            side_effect_class: SideEffectClass::SidePathArtifact,
            transaction_scope: Some("side-path artifact creation outside source DB mutation"),
            idempotency_key: Some(idempotency_key),
            audit_surface: Some("artifact manifest or audit_log when DB backing exists"),
            db_generation_effect: "source DB generation unchanged unless manifest audit is committed",
            index_generation_effect: "none",
            dry_run_behavior: Some("preview artifact path and manifest only; no files are written"),
            recovery_behavior: "partial side-path output is reported as failed, never deleted by ee, and not treated as a valid artifact",
            no_overwrite_behavior: Some(no_overwrite_behavior),
            degraded_code: None,
        }
    }

    #[must_use]
    pub const fn degraded_unavailable(degraded_code: &'static str) -> Self {
        Self {
            side_effect_class: SideEffectClass::DegradedUnavailable,
            transaction_scope: None,
            idempotency_key: Some("full command argv and explicit inputs"),
            audit_surface: None,
            db_generation_effect: "none",
            index_generation_effect: "none",
            dry_run_behavior: None,
            recovery_behavior: "returns an explicit degraded response without mutation",
            no_overwrite_behavior: None,
            degraded_code: Some(degraded_code),
        }
    }

    #[must_use]
    pub const fn supervised_jobs(
        idempotency_key: &'static str,
        recovery_behavior: &'static str,
    ) -> Self {
        Self {
            side_effect_class: SideEffectClass::SupervisedJobs,
            transaction_scope: Some("supervised steward job ledger"),
            idempotency_key: Some(idempotency_key),
            audit_surface: Some("audit_log"),
            db_generation_effect: "advances only when the configured job applies durable changes",
            index_generation_effect: "unchanged unless the steward job explicitly processes index work",
            dry_run_behavior: Some(
                "runs handler planning and reports candidate changes without committing job mutations",
            ),
            recovery_behavior,
            no_overwrite_behavior: None,
            degraded_code: None,
        }
    }

    #[must_use]
    pub fn declares_no_source_mutation(&self) -> bool {
        matches!(
            self.db_generation_effect,
            "none" | "source DB generation unchanged"
        )
    }
}

/// Idempotency behavior of a command.
#[derive(Clone, Copy, Debug, Eq, PartialEq, Hash)]
pub enum IdempotencyClass {
    /// Running twice produces the same observable outcome.
    Idempotent,
    /// Running twice may produce different outcomes (e.g., new memory IDs).
    NonIdempotent,
    /// Command supports `--dry-run` to preview without mutation.
    DryRunAvailable,
}

impl IdempotencyClass {
    #[must_use]
    pub const fn as_str(self) -> &'static str {
        match self {
            Self::Idempotent => "idempotent",
            Self::NonIdempotent => "non_idempotent",
            Self::DryRunAvailable => "dry_run_available",
        }
    }
}

/// Runtime class for cancellation and budget behavior.
#[derive(Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd, Hash)]
pub enum RuntimeClass {
    /// Completes without meaningful async checkpoints.
    Immediate,
    /// Bounded local probes or DB reads; cancellation checked around boundaries.
    Bounded,
    /// Potentially long-running work with explicit budget/deadline checkpoints.
    LongRunning,
    /// Multi-stage work with commit/publish boundaries and cleanup policy.
    MultiStage,
    /// Work coordinated through a supervised child/job ledger.
    Supervised,
}

impl RuntimeClass {
    /// Stable vocabulary token used in boundary logs and tests.
    #[must_use]
    pub const fn as_str(self) -> &'static str {
        match self {
            Self::Immediate => "immediate",
            Self::Bounded => "bounded",
            Self::LongRunning => "long_running",
            Self::MultiStage => "multi_stage",
            Self::Supervised => "supervised",
        }
    }

    /// `true` when commands in this class need an explicit runtime budget.
    #[must_use]
    pub const fn requires_budget(self) -> bool {
        matches!(
            self,
            Self::LongRunning | Self::MultiStage | Self::Supervised
        )
    }
}

/// Cross-cutting runtime contract for a command manifest entry.
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
pub struct CommandRuntimeContract {
    /// Classifies the runtime shape of the command.
    pub runtime_class: RuntimeClass,
    /// Default runtime budget in milliseconds, if the command is budgeted.
    pub default_budget_ms: Option<u64>,
    /// Stable cancellation checkpoints named in boundary logs.
    pub cancellation_points: &'static [&'static str],
    /// Policy for cleanup or audit after partial progress.
    pub partial_progress_policy: &'static str,
    /// Deterministic mapping from runtime outcome to CLI result/log outcome.
    pub outcome_mapping: &'static str,
}

impl CommandRuntimeContract {
    #[must_use]
    pub const fn immediate() -> Self {
        Self {
            runtime_class: RuntimeClass::Immediate,
            default_budget_ms: None,
            cancellation_points: &["before_start"],
            partial_progress_policy: "no durable partial progress is possible",
            outcome_mapping: "success or explicit degraded/error response",
        }
    }

    #[must_use]
    pub const fn bounded_read() -> Self {
        Self {
            runtime_class: RuntimeClass::Bounded,
            default_budget_ms: Some(30_000),
            cancellation_points: &["before_start", "between_bounded_probes"],
            partial_progress_policy: "read-only; no durable partial progress is possible",
            outcome_mapping: "success, degraded, or read-side error with no mutation",
        }
    }

    #[must_use]
    pub const fn long_running_derived() -> Self {
        Self {
            runtime_class: RuntimeClass::LongRunning,
            default_budget_ms: Some(300_000),
            cancellation_points: &["before_start", "source_scan", "before_publish"],
            partial_progress_policy: "derived artifacts publish atomically; failed generations are ignored until a complete publish",
            outcome_mapping: "success, cancelled, budget_exhausted, or index_error",
        }
    }

    #[must_use]
    pub const fn transactional() -> Self {
        Self {
            runtime_class: RuntimeClass::MultiStage,
            default_budget_ms: Some(60_000),
            cancellation_points: &["before_start", "before_transaction", "before_commit"],
            partial_progress_policy: "single transaction rollback leaves no unaudited durable records",
            outcome_mapping: "success, cancelled, budget_exhausted, storage_error, or degraded",
        }
    }

    #[must_use]
    pub const fn side_path_artifact() -> Self {
        Self {
            runtime_class: RuntimeClass::MultiStage,
            default_budget_ms: Some(120_000),
            cancellation_points: &["before_start", "before_artifact_write", "before_manifest"],
            partial_progress_policy: "partial side-path output is reported, never deleted by ee, and is not a valid artifact until manifested",
            outcome_mapping: "success, cancelled, budget_exhausted, storage_error, or degraded",
        }
    }

    #[must_use]
    pub const fn supervised_unavailable() -> Self {
        Self {
            runtime_class: RuntimeClass::Supervised,
            default_budget_ms: Some(300_000),
            cancellation_points: &["before_start", "before_child_spawn", "child_outcome"],
            partial_progress_policy: "supervised jobs must record child failure or cancellation before reporting completion",
            outcome_mapping: "degraded, cancelled, budget_exhausted, or supervised_child_failed",
        }
    }

    #[must_use]
    pub const fn supervised_jobs() -> Self {
        Self {
            runtime_class: RuntimeClass::Supervised,
            default_budget_ms: Some(300_000),
            cancellation_points: &[
                "before_start",
                "before_job_schedule",
                "before_handler",
                "handler_outcome",
            ],
            partial_progress_policy: "supervised jobs report skipped, failed, cancelled, or applied handler work in the runner result",
            outcome_mapping: "success, skipped, failed, cancelled, budget_exhausted, or supervised_child_failed",
        }
    }

    #[must_use]
    pub const fn requires_budget(&self) -> bool {
        self.runtime_class.requires_budget()
    }

    pub fn effective_budget_ms(
        &self,
        requested_budget_ms: Option<u64>,
    ) -> Result<Option<u64>, &'static str> {
        match requested_budget_ms {
            Some(0) => Err("runtime budget must be greater than zero"),
            Some(budget) => Ok(Some(budget)),
            None => Ok(self.default_budget_ms),
        }
    }
}

/// Allowed write surfaces for a command.
#[derive(Clone, Debug, Default, Eq, PartialEq)]
pub struct WriteSurfaces {
    /// Database tables the command may write.
    pub db_tables: Vec<&'static str>,
    /// Derived artifact paths (relative to workspace).
    pub derived_paths: Vec<&'static str>,
    /// Workspace file patterns the command may write.
    pub workspace_files: Vec<&'static str>,
}

impl WriteSurfaces {
    #[must_use]
    pub const fn none() -> Self {
        Self {
            db_tables: Vec::new(),
            derived_paths: Vec::new(),
            workspace_files: Vec::new(),
        }
    }

    #[must_use]
    pub fn is_empty(&self) -> bool {
        self.db_tables.is_empty()
            && self.derived_paths.is_empty()
            && self.workspace_files.is_empty()
    }
}

/// Effect manifest entry for a single command.
#[derive(Clone, Debug, Eq, PartialEq)]
pub struct CommandEffect {
    /// Command path (e.g., "status", "memory list", "index rebuild").
    pub command_path: &'static str,
    /// Default effect class when run normally.
    pub default_effect: EffectClass,
    /// Effect class when run with `--dry-run`.
    ///
    /// MIGRATION NOTE FOR CONSUMERS -- THE MEANING OF `None` CHANGED (bd-3j6l3,
    /// bd-mmbzu). Before this change every mutating constructor hardcoded
    /// `Some(EffectClass::ReadOnly)`, so `None` occurred on exactly one mutating
    /// command in the whole manifest and effectively meant "unknown". An agent
    /// could not distinguish a command that had been considered from one that
    /// had never been thought about, because the constructor answered for both.
    ///
    /// `None` NOW MEANS "this command's parser does not accept `--dry-run`",
    /// and it is the default. If you previously read `Some(_)` as evidence that
    /// a destructive command could be safely previewed, that reading was wrong
    /// for 90 of 191 mutating commands and is no longer produced: support is
    /// now opt-in via `DRY_RUN_CAPABLE`, and
    /// `cli::tests::mutating_dry_run_declarations_match_the_parser` fails in
    /// both directions if the roster and Clap disagree.
    ///
    /// If you parse this field, the safe migration is to treat `Some(_)` as
    /// "`--dry-run` is accepted" and `None` as "it is not", and to stop
    /// treating `None` as missing data. To ask the weaker question "can this be
    /// previewed at all", read `mutation_contract.dry_run_behavior` instead --
    /// see its own note, because several commands preview through `--print`,
    /// `--show`, `--read-only` or by omitting a flag, and none of those are
    /// `--dry-run`.
    pub dry_run_effect: Option<EffectClass>,
    /// Idempotency behavior.
    pub idempotency: IdempotencyClass,
    /// Surfaces the command may write.
    pub write_surfaces: WriteSurfaces,
    /// Cross-cutting side-effect and mutation-safety contract.
    pub mutation_contract: CommandMutationContract,
    /// Cross-cutting runtime, cancellation, and budget contract.
    pub runtime_contract: CommandRuntimeContract,
    /// Whether the command should run through a read-side snapshot lease.
    pub requires_read_snapshot: bool,
    /// Whether command requires audit log write.
    pub requires_audit: bool,
    /// Human-readable description of the effect.
    pub description: &'static str,
}

impl CommandEffect {
    /// Create a read-only effect entry.
    #[must_use]
    pub const fn read_only(command_path: &'static str, description: &'static str) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::ReadOnly,
            dry_run_effect: None,
            idempotency: IdempotencyClass::Idempotent,
            write_surfaces: WriteSurfaces::none(),
            mutation_contract: CommandMutationContract::read_only(),
            runtime_contract: CommandRuntimeContract::bounded_read(),
            requires_read_snapshot: false,
            requires_audit: false,
            description,
        }
    }

    /// Create a DB-backed read-only effect entry that must use a read snapshot.
    #[must_use]
    pub fn read_only_db(command_path: &'static str, description: &'static str) -> Self {
        Self::read_only(command_path, description).with_read_snapshot()
    }

    /// Create a derived-artifact-write effect entry.
    #[must_use]
    pub fn derived_write(
        command_path: &'static str,
        derived_paths: Vec<&'static str>,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::DerivedArtifactWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::Idempotent,
            write_surfaces: WriteSurfaces {
                db_tables: Vec::new(),
                derived_paths,
                workspace_files: Vec::new(),
            },
            mutation_contract: CommandMutationContract::derived_asset_rebuild(
                "source DB generation",
                "derived artifacts are rebuildable from FrankenSQLite source records",
            ),
            runtime_contract: CommandRuntimeContract::long_running_derived(),
            requires_read_snapshot: false,
            requires_audit: false,
            description,
        }
    }

    /// Create a durable-memory-write effect entry.
    #[must_use]
    pub fn durable_write(
        command_path: &'static str,
        db_tables: Vec<&'static str>,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::DurableMemoryWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::NonIdempotent,
            write_surfaces: WriteSurfaces {
                db_tables,
                derived_paths: Vec::new(),
                workspace_files: Vec::new(),
            },
            mutation_contract: CommandMutationContract::audited_mutation(
                "caller-provided key or generated durable record ID",
            ),
            runtime_contract: CommandRuntimeContract::transactional(),
            requires_read_snapshot: false,
            requires_audit: true,
            description,
        }
    }

    /// Create a durable write that also declares companion workspace-file surfaces.
    #[must_use]
    pub fn durable_write_with_workspace_files(
        command_path: &'static str,
        db_tables: Vec<&'static str>,
        workspace_files: Vec<&'static str>,
        description: &'static str,
    ) -> Self {
        let mut effect = Self::durable_write(command_path, db_tables, description);
        effect.write_surfaces.workspace_files = workspace_files;
        effect
    }

    /// Create an append-only durable-write effect entry.
    #[must_use]
    pub fn append_only_write(
        command_path: &'static str,
        db_tables: Vec<&'static str>,
        idempotency_key: &'static str,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::DurableMemoryWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::Idempotent,
            write_surfaces: WriteSurfaces {
                db_tables,
                derived_paths: Vec::new(),
                workspace_files: Vec::new(),
            },
            mutation_contract: CommandMutationContract::append_only(idempotency_key),
            runtime_contract: CommandRuntimeContract::transactional(),
            requires_read_snapshot: false,
            requires_audit: true,
            description,
        }
    }

    /// Create a workspace-file-write effect entry.
    #[must_use]
    pub fn workspace_file_write(
        command_path: &'static str,
        workspace_files: Vec<&'static str>,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::WorkspaceFileWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::NonIdempotent,
            write_surfaces: WriteSurfaces {
                db_tables: Vec::new(),
                derived_paths: Vec::new(),
                workspace_files,
            },
            mutation_contract: CommandMutationContract::side_path_artifact(
                "artifact path plus manifest hash",
                "no-overwrite/no-delete: existing output paths block unless the verifier proves the same manifest",
            ),
            runtime_contract: CommandRuntimeContract::side_path_artifact(),
            requires_read_snapshot: false,
            requires_audit: true,
            description,
        }
    }

    /// Create an audited external-I/O command entry.
    #[must_use]
    pub fn external_io_write(
        command_path: &'static str,
        db_tables: Vec<&'static str>,
        workspace_files: Vec<&'static str>,
        idempotency_key: &'static str,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::ExternalIo,
            dry_run_effect: None,
            idempotency: IdempotencyClass::NonIdempotent,
            write_surfaces: WriteSurfaces {
                db_tables,
                derived_paths: Vec::new(),
                workspace_files,
            },
            mutation_contract: CommandMutationContract {
                side_effect_class: SideEffectClass::AuditedMutation,
                transaction_scope: Some("one audit ledger append per executed command step"),
                idempotency_key: Some(idempotency_key),
                audit_surface: Some("audit_log"),
                db_generation_effect: "advances audit log for each executed step; unchanged for dry-run or rejected unsafe steps",
                index_generation_effect: "none",
                dry_run_behavior: Some(
                    "parses the manifest and reports planned steps without executing commands or writing evidence",
                ),
                recovery_behavior: "failed steps keep evidence and audit rows; later steps are skipped unless the caller opts into continuing",
                no_overwrite_behavior: Some(
                    "no-overwrite/no-delete: evidence paths use fresh run IDs and ee never deletes partial demo evidence",
                ),
                degraded_code: None,
            },
            runtime_contract: CommandRuntimeContract {
                runtime_class: RuntimeClass::MultiStage,
                default_budget_ms: Some(120_000),
                cancellation_points: &[
                    "before_start",
                    "before_step_execute",
                    "after_step_evidence",
                    "before_audit_commit",
                ],
                partial_progress_policy: "each executed step writes evidence plus one audit row; remaining steps are skipped after the first failure unless explicitly continued",
                outcome_mapping: "success, policy_denied, usage_error, storage_error, or step failure with persisted evidence",
            },
            requires_read_snapshot: false,
            requires_audit: true,
            description,
        }
    }

    /// Create a config-write effect entry.
    ///
    /// The workspace registry is a separate database file, not a table of the
    /// workspace store, and only `workspace alias` writes it; that entry
    /// declares it as a derived path (bd-hb5os).
    ///
    /// bd-acxw6: `db_tables` is a parameter rather than a hard-coded
    /// `["audit_log"]`. The hard-coded form made every config-write command
    /// claim an audit row, and measurement (bd-hb5os probe v2) found that not
    /// one of the four writes one. `write_surfaces` is descriptive — it is the
    /// measured write set — while `side_effect_class` and `requires_audit`
    /// stay class-normative, exactly as bd-hb5os left them. Where the two now
    /// disagree the gap is a real product question, tracked separately, not
    /// something to paper over by relabelling the class.
    #[must_use]
    pub fn config_write(
        command_path: &'static str,
        db_tables: Vec<&'static str>,
        workspace_files: Vec<&'static str>,
        idempotency_key: &'static str,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::ConfigWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::Idempotent,
            write_surfaces: WriteSurfaces {
                db_tables,
                derived_paths: Vec::new(),
                workspace_files,
            },
            mutation_contract: CommandMutationContract::audited_mutation(idempotency_key),
            runtime_contract: CommandRuntimeContract::transactional(),
            requires_read_snapshot: false,
            requires_audit: true,
            description,
        }
    }

    /// Create a config-file-write entry that does not touch the ee database.
    #[must_use]
    pub fn config_file_write(
        command_path: &'static str,
        workspace_files: Vec<&'static str>,
        idempotency_key: &'static str,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::ConfigWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::DryRunAvailable,
            write_surfaces: WriteSurfaces {
                db_tables: Vec::new(),
                derived_paths: Vec::new(),
                workspace_files,
            },
            mutation_contract: CommandMutationContract {
                side_effect_class: SideEffectClass::SidePathArtifact,
                transaction_scope: Some("workspace config file update"),
                idempotency_key: Some(idempotency_key),
                audit_surface: None,
                db_generation_effect: "source DB generation unchanged",
                index_generation_effect: "none",
                dry_run_behavior: Some(
                    "--dry-run validates and previews config changes; no files are written",
                ),
                recovery_behavior: "partial config-file output is reported as failed, never deleted by ee, and not treated as committed configuration",
                no_overwrite_behavior: Some(
                    "no-overwrite contract preserves unrelated config keys where possible; no-delete: ee never deletes the config file",
                ),
                degraded_code: None,
            },
            runtime_contract: CommandRuntimeContract::side_path_artifact(),
            requires_read_snapshot: false,
            requires_audit: false,
            description,
        }
    }

    /// Create a harness settings-file write entry that does not touch the ee database.
    #[must_use]
    pub fn harness_hook_settings_write(
        command_path: &'static str,
        workspace_files: Vec<&'static str>,
        idempotency_key: &'static str,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::ConfigWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::DryRunAvailable,
            write_surfaces: WriteSurfaces {
                db_tables: Vec::new(),
                derived_paths: Vec::new(),
                workspace_files,
            },
            mutation_contract: CommandMutationContract {
                side_effect_class: SideEffectClass::SidePathArtifact,
                transaction_scope: Some("agent harness settings file update"),
                idempotency_key: Some(idempotency_key),
                audit_surface: None,
                db_generation_effect: "source DB generation unchanged",
                index_generation_effect: "none",
                dry_run_behavior: Some(
                    "--print previews generated harness hooks without writing settings files",
                ),
                recovery_behavior: "failed harness settings writes return storage_error, partial output is never deleted by ee, and backups are left in place for manual restore and --undo",
                no_overwrite_behavior: Some(
                    "no-overwrite/no-delete: preserves unmanaged settings entries; install writes a deterministic backup before changing managed hook entries; ee never deletes harness settings files",
                ),
                degraded_code: None,
            },
            runtime_contract: CommandRuntimeContract::side_path_artifact(),
            requires_read_snapshot: false,
            requires_audit: false,
            description,
        }
    }

    /// Create a certificate key-file write entry that does not touch the ee database.
    #[must_use]
    pub fn certificate_key_file_write(
        command_path: &'static str,
        workspace_files: Vec<&'static str>,
        idempotency_key: &'static str,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::ConfigWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::DryRunAvailable,
            write_surfaces: WriteSurfaces {
                db_tables: Vec::new(),
                derived_paths: Vec::new(),
                workspace_files,
            },
            mutation_contract: CommandMutationContract {
                side_effect_class: SideEffectClass::SidePathArtifact,
                transaction_scope: Some("key file create-or-explicit-force overwrite"),
                idempotency_key: Some(idempotency_key),
                audit_surface: None,
                db_generation_effect: "source DB generation unchanged",
                index_generation_effect: "none",
                dry_run_behavior: Some(
                    "--show reads existing key material only; no files are written",
                ),
                recovery_behavior: "partial key-file output is reported as failed, never deleted by ee, and not treated as valid key material",
                no_overwrite_behavior: Some(
                    "no-overwrite by default; --force is an explicit overwrite, and no-delete: ee never deletes key material",
                ),
                degraded_code: None,
            },
            runtime_contract: CommandRuntimeContract::side_path_artifact(),
            requires_read_snapshot: false,
            requires_audit: false,
            description,
        }
    }

    /// Create a workspace state-file write entry.
    #[must_use]
    pub fn workspace_state_write(
        command_path: &'static str,
        workspace_files: Vec<&'static str>,
        idempotency_key: &'static str,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::WorkspaceFileWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::DryRunAvailable,
            write_surfaces: WriteSurfaces {
                db_tables: Vec::new(),
                derived_paths: Vec::new(),
                workspace_files,
            },
            mutation_contract: CommandMutationContract {
                side_effect_class: SideEffectClass::AuditedMutation,
                transaction_scope: Some("workspace-local state file update"),
                idempotency_key: Some(idempotency_key),
                audit_surface: Some("workspace state file"),
                db_generation_effect: "source DB generation unchanged",
                index_generation_effect: "none",
                dry_run_behavior: Some(
                    "--dry-run previews the state transition without writing workspace files",
                ),
                recovery_behavior: "failed workspace state writes return storage_error; rerun from the last readable state file",
                no_overwrite_behavior: None,
                degraded_code: None,
            },
            runtime_contract: CommandRuntimeContract::transactional(),
            requires_read_snapshot: false,
            requires_audit: true,
            description,
        }
    }

    /// Create a durable state write backed by a non-audit-log evidence spine.
    #[must_use]
    pub fn durable_state_write(
        command_path: &'static str,
        db_tables: Vec<&'static str>,
        idempotency_key: &'static str,
        audit_surface: &'static str,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::DurableMemoryWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::DryRunAvailable,
            write_surfaces: WriteSurfaces {
                db_tables,
                derived_paths: Vec::new(),
                workspace_files: Vec::new(),
            },
            mutation_contract: CommandMutationContract {
                side_effect_class: SideEffectClass::AuditedMutation,
                transaction_scope: Some("single DB transaction across state/evidence rows"),
                idempotency_key: Some(idempotency_key),
                audit_surface: Some(audit_surface),
                db_generation_effect: "advances on commit; unchanged for dry-run or rollback",
                index_generation_effect: "none unless a downstream steward job is queued",
                dry_run_behavior: Some("validates and renders the report without writing DB rows"),
                recovery_behavior: "transaction rollback leaves no partial durable records",
                no_overwrite_behavior: None,
                degraded_code: None,
            },
            runtime_contract: CommandRuntimeContract::transactional(),
            requires_read_snapshot: false,
            requires_audit: true,
            description,
        }
    }

    /// Create the schema-migration effect entry.
    #[must_use]
    pub fn schema_migration_run() -> Self {
        Self {
            command_path: "migrate run",
            default_effect: EffectClass::DurableMemoryWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::Idempotent,
            write_surfaces: WriteSurfaces {
                db_tables: vec![
                    "ee_schema_migrations",
                    "memories",
                    "search_index_jobs",
                    "audit_log",
                ],
                derived_paths: vec![".ee/index/"],
                workspace_files: Vec::new(),
            },
            mutation_contract: CommandMutationContract {
                side_effect_class: SideEffectClass::AuditedMutation,
                transaction_scope: Some(
                    "ordered schema migrations plus post-migration backfill/index audit",
                ),
                idempotency_key: Some("database path plus compiled migration checksums"),
                audit_surface: Some("ee_schema_migrations and audit_log"),
                db_generation_effect: "advances schema migration state and any migration-owned rows on commit",
                index_generation_effect: "post-migration index rebuild may refresh derived index generation",
                dry_run_behavior: Some(
                    "--dry-run reports pending migrations and backfill/index plans without mutation",
                ),
                recovery_behavior: "migration transaction rollback leaves unapplied versions pending for retry",
                no_overwrite_behavior: None,
                degraded_code: None,
            },
            runtime_contract: CommandRuntimeContract::transactional(),
            requires_read_snapshot: false,
            requires_audit: true,
            description: "Apply pending schema migrations and post-migration repair work",
        }
    }

    /// Create the shard fan-out migration effect entry.
    #[must_use]
    pub fn shard_fanout_migration() -> Self {
        Self {
            command_path: "migrate shard-fanout",
            default_effect: EffectClass::WorkspaceFileWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::DryRunAvailable,
            write_surfaces: WriteSurfaces {
                // The shard catalog and the shard databases are files, named
                // below; only the audit log is a table (bd-hb5os).
                db_tables: vec!["audit_log"],
                derived_paths: Vec::new(),
                workspace_files: vec![
                    "<shards-dir>/catalog.db",
                    "<shards-dir>/<workspace-shard>.db",
                    "<source-db>.pre-shard-fanout",
                ],
            },
            mutation_contract: CommandMutationContract {
                side_effect_class: SideEffectClass::AuditedMutation,
                transaction_scope: Some(
                    "preserve source database, copy workspace rows, then write shard catalog",
                ),
                idempotency_key: Some("source database hash plus shard fan-out plan"),
                audit_surface: Some("shard migration audit rows"),
                db_generation_effect: "source DB generation is preserved; shard catalog and workspace shard DBs advance",
                index_generation_effect: "none",
                dry_run_behavior: Some(
                    "--dry-run reports the shard plan and blockers without writing shard files",
                ),
                recovery_behavior: "preserved source copy and shard hashes let reruns detect already-applied work",
                no_overwrite_behavior: Some(
                    "no-overwrite/no-delete: source database is preserved before copy and existing incompatible shard/catalog hashes block apply",
                ),
                degraded_code: None,
            },
            runtime_contract: CommandRuntimeContract::side_path_artifact(),
            requires_read_snapshot: false,
            requires_audit: true,
            description: "Migrate a monolithic workspace database into per-workspace shard files",
        }
    }

    /// Create a degraded/unavailable read-only effect entry.
    #[must_use]
    pub const fn degraded_unavailable(
        command_path: &'static str,
        degraded_code: &'static str,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::ReadOnly,
            dry_run_effect: None,
            idempotency: IdempotencyClass::Idempotent,
            write_surfaces: WriteSurfaces::none(),
            mutation_contract: CommandMutationContract::degraded_unavailable(degraded_code),
            runtime_contract: CommandRuntimeContract::immediate(),
            requires_read_snapshot: false,
            requires_audit: false,
            description,
        }
    }

    /// Create a supervised maintenance-job effect entry.
    #[must_use]
    pub fn supervised_job(
        command_path: &'static str,
        db_tables: Vec<&'static str>,
        description: &'static str,
    ) -> Self {
        Self {
            command_path,
            default_effect: EffectClass::DurableMemoryWrite,
            dry_run_effect: None,
            idempotency: IdempotencyClass::DryRunAvailable,
            write_surfaces: WriteSurfaces {
                db_tables,
                derived_paths: Vec::new(),
                workspace_files: Vec::new(),
            },
            mutation_contract: CommandMutationContract::supervised_jobs(
                "workspace id plus job type plus unapplied feedback set",
                "job result reports failed/skipped/cancelled work; durable changes are limited to handler-owned audited updates",
            ),
            runtime_contract: CommandRuntimeContract::supervised_jobs(),
            requires_read_snapshot: false,
            requires_audit: true,
            description,
        }
    }

    /// Override the default runtime contract for a specific command path.
    #[must_use]
    pub const fn with_runtime_contract(mut self, runtime_contract: CommandRuntimeContract) -> Self {
        self.runtime_contract = runtime_contract;
        self
    }

    /// Mark this command as requiring a read-side snapshot lease.
    #[must_use]
    pub const fn with_read_snapshot(mut self) -> Self {
        self.requires_read_snapshot = true;
        self
    }

    /// `true` if this read-only command should acquire a read-side snapshot.
    #[must_use]
    pub const fn read_snapshot(&self) -> bool {
        self.requires_read_snapshot
    }

    /// `true` if running this command is safe mid-task (no durable mutation).
    #[must_use]
    pub const fn is_safe_mid_task(&self) -> bool {
        matches!(self.default_effect, EffectClass::ReadOnly)
    }
}

/// Mutating commands whose parser accepts `--dry-run`. bd-3j6l3 / bd-mmbzu.
///
/// THIS LIST EXISTS BECAUSE A DEFAULT THAT CLAIMS A CAPABILITY IS THE DEFECT.
/// Every mutating constructor used to hardcode
/// `dry_run_effect: Some(EffectClass::ReadOnly)`, so a command that had never
/// once considered dry-run still asserted it offered a read-only preview. That
/// was not a stale value; it was a claim no author made, produced by a
/// constructor. Measured against the real Clap tree, it was false for 90 of the
/// 191 mutating commands that carried it.
///
/// Silence now means "no `--dry-run`". Support is declared HERE and nowhere
/// else, and `EffectManifest::build` applies it last.
///
/// THIS ROSTER CANNOT DRIFT UNDETECTED, which is the whole reason it is
/// allowed to be a hand-maintained list.
/// `cli::tests::mutating_dry_run_declarations_match_the_parser` resolves every
/// declared mutating path against the real command tree and fails if:
///   - a name here has no `--dry-run` in Clap        (the roster over-claims)
///   - a command has `--dry-run` and is missing here (the roster under-claims)
///   - a declared manifest path is not a real command path at all
/// and `dry_run_capable_roster_names_only_real_commands` fails on a name that
/// matches no manifest entry, because `build` skips those silently.
///
/// MEASURED, NOT CURATED. These 96 names came from walking `Cli::command()` at
/// ddc11bb21; the run partitioned 192 mutating commands into 96 here, 90 that
/// claimed the flag without having it, 5 declared paths Clap cannot address,
/// and `ask`, which declares no preview and has none. The four sets are
/// disjoint and sum to 192.
///
/// Adding a name to silence a failure is the one way to weaken this. The gate
/// asks Clap, so a name added without a flag fails immediately in the other
/// direction.
pub(crate) const DRY_RUN_CAPABLE: &[&str] = &[
    "artifact register",
    "backup create",
    "backup keys export",
    "backup keys import",
    "backup restore",
    "causal promote-plan",
    "config set",
    "coordination evidence ingest",
    "curate accept",
    "curate apply",
    "curate auto-promote",
    "curate merge",
    "curate propose-derived",
    "curate reject",
    "curate retire",
    "curate snooze",
    "curate tombstone",
    "curate untombstone",
    "daemon",
    "decide record",
    "demo run",
    "diag pack-record",
    "export",
    "export agentsmd",
    "graph centrality-refresh",
    "graph feature-enrichment",
    "graph snapshot refresh",
    "handoff create",
    "import agentsmd",
    "import cass",
    "import eidetic-legacy",
    "import jsonl",
    "index backfill-tags",
    "index rebuild",
    "index reembed",
    "init",
    "job run",
    "journal distill",
    "lab capture",
    "lab swarm replay",
    "learn close",
    "learn experiment run",
    "learn observe",
    "maintenance graph-snapshot-prune",
    "maintenance graph-witnesses-prune",
    "maintenance run",
    "maintenance wal-checkpoint",
    "memory demote-global",
    "memory expire",
    "memory level",
    "memory link",
    "memory outcome-global",
    "memory promote-global",
    "memory revise",
    "memory tags",
    "mesh auto-enroll",
    "mesh disable",
    "mesh import",
    "mesh reenable",
    "migrate run",
    "migrate shard-fanout",
    "note",
    "outcome",
    "plan recipe save",
    "playbook export",
    "playbook extract",
    "preflight close",
    "preflight run",
    "procedure promote",
    "procedure propose",
    "profile config apply",
    "rationale attach",
    "recorder event",
    "recorder finish",
    "recorder start",
    "reflect ingest",
    "reflect propose",
    "remember",
    "review session --propose",
    "review workspace",
    "rule add",
    "rule mark",
    "rule protect",
    "rule update",
    "shadow demote",
    "shadow promote",
    "support bundle",
    "task-frame close",
    "task-frame create",
    "task-frame subgoal add",
    "task-frame update",
    "tripwire check",
    "verification provenance",
    "verify provenance",
    "workflow create",
    "workspace alias",
];

/// The complete command effect manifest.
#[derive(Clone, Debug)]
pub struct EffectManifest {
    entries: HashMap<&'static str, CommandEffect>,
}

impl EffectManifest {
    /// Build the manifest from the canonical command list.
    ///
    /// Panics if any `command_path` appears in more than one of the nine
    /// category vectors. The contract is "a command must be classified
    /// in exactly one effect class"; a duplicate would mean the second
    /// category silently overwrites the first via `HashMap::insert`,
    /// and a command's declared effect would depend on the order
    /// `build()` walks the category functions. The mid-task safety
    /// classifier (`is_safe_mid_task`), the doctor capability surface,
    /// and the audit log all key off this manifest, so a silent
    /// miscategorization here would route a mutating command through a
    /// safe-read code path. The duplicate check turns that drift into a
    /// loud, immediate failure at startup.
    #[must_use]
    pub fn build() -> Self {
        let mut entries = HashMap::new();

        // Read-only commands
        for entry in Self::read_only_commands() {
            Self::insert_unique(&mut entries, entry);
        }

        // Explicitly unavailable commands that must not mutate.
        for entry in Self::degraded_unavailable_commands() {
            Self::insert_unique(&mut entries, entry);
        }

        // Derived artifact write commands
        for entry in Self::derived_write_commands() {
            Self::insert_unique(&mut entries, entry);
        }

        // Audited external command execution surfaces
        for entry in Self::external_io_write_commands() {
            Self::insert_unique(&mut entries, entry);
        }

        // Supervised steward jobs
        for entry in Self::supervised_job_commands() {
            Self::insert_unique(&mut entries, entry);
        }

        // Append-only write commands
        for entry in Self::append_only_write_commands() {
            Self::insert_unique(&mut entries, entry);
        }

        // Durable write commands
        for entry in Self::durable_write_commands() {
            Self::insert_unique(&mut entries, entry);
        }

        // Config write commands
        for entry in Self::config_write_commands() {
            Self::insert_unique(&mut entries, entry);
        }

        // Workspace file write commands
        for entry in Self::workspace_file_write_commands() {
            Self::insert_unique(&mut entries, entry);
        }

        // OPT-IN, APPLIED LAST. Constructors no longer guess whether a command
        // accepts `--dry-run`; DRY_RUN_CAPABLE names the ones that do.
        //
        // A name here that matches no entry is SKIPPED SILENTLY by this loop,
        // which is why `dry_run_capable_roster_names_only_real_commands` exists
        // to fail on it. Do not add an `else` that panics: build() is called
        // from the CLI hot path and a typo should red a gate, not abort a user
        // command.
        // ONLY FILLS A HOLE, NEVER OVERWRITES. `daemon` sets
        // `Some(EffectClass::WorkspaceFileWrite)` deliberately because its
        // foreground dry-run still records planned job rows; an unconditional
        // assignment here would silently downgrade that to ReadOnly and make
        // the manifest understate what `daemon --dry-run` writes.
        for path in DRY_RUN_CAPABLE {
            if let Some(entry) = entries.get_mut(*path)
                && entry.dry_run_effect.is_none()
            {
                entry.dry_run_effect = Some(EffectClass::ReadOnly);
            }
        }

        Self { entries }
    }

    fn insert_unique(entries: &mut HashMap<&'static str, CommandEffect>, entry: CommandEffect) {
        let path = entry.command_path;
        let previous_class = entries.get(path).map(|prior| prior.default_effect);
        if entries.insert(path, entry).is_some() {
            panic!(
                "EffectManifest::build: duplicate command path `{path}` registered in two \
                 categories (previous default_effect = {previous_class:?}). A command must \
                 appear in exactly one of {{read_only, degraded_unavailable, derived_write, \
                 external_io_write, supervised_job, append_only_write, durable_write, \
                 config_write, workspace_file_write}} — the second registration would \
                 silently overwrite the first and the command's declared effect class \
                 would depend on the build()-walk order."
            );
        }
    }

    fn read_only_commands() -> Vec<CommandEffect> {
        vec![
            CommandEffect::read_only("agent detect", "Detect installed coding agents"),
            CommandEffect::read_only("agent scan", "Scan agent probe paths"),
            CommandEffect::read_only("agent sources", "List known agent source connectors"),
            CommandEffect::read_only("agent status", "Report local agent inventory status"),
            CommandEffect::read_only(
                "analyze clustering",
                "Analyze consolidation clustering posture",
            ),
            CommandEffect::read_only(
                "analyze drift",
                "Analyze drift between evaluation snapshots",
            ),
            CommandEffect::read_only("analyze science-status", "Report science readiness"),
            CommandEffect::read_only("agent-docs", "Display agent documentation"),
            CommandEffect::read_only_db("audit diff", "Show audit log mutations in a time window"),
            CommandEffect::read_only_db("audit show", "Show one audit log row"),
            CommandEffect::read_only_db("audit timeline", "List audit log rows"),
            CommandEffect::read_only_db("audit verify", "Verify audit hash-chain integrity"),
            CommandEffect::read_only_db("backup inspect", "Inspect backup manifest"),
            CommandEffect::read_only_db("backup list", "List backup manifests"),
            CommandEffect::read_only_db("backup verify", "Verify backup manifest and contents"),
            CommandEffect::read_only_db(
                "bootstrap docs",
                "Compile allowlisted workspace docs into dry-run bootstrap candidates (no durable mutation)",
            ),
            CommandEffect::read_only("capabilities", "Report feature availability"),
            CommandEffect::read_only(
                "cache hotset-manifest",
                "Collect a read-only hotset manifest from bounded coordination evidence",
            ),
            CommandEffect::read_only(
                "cache prewarm",
                "Plan explicit cache prewarm admission from a hotset manifest",
            ),
            CommandEffect::read_only_db("certificate list", "List persisted certificate records"),
            CommandEffect::read_only_db(
                "certificate show",
                "Inspect a persisted certificate record",
            ),
            CommandEffect::read_only_db(
                "certificate verify",
                "Verify persisted certificate hash and signature evidence",
            ),
            CommandEffect::read_only(
                "certificate sign",
                "Compute a certificate signature from local key material without persisting it",
            ),
            CommandEffect::read_only_db("causal trace", "Trace persisted causal evidence chains"),
            CommandEffect::read_only_db(
                "causal compare",
                "Compare persisted causal evidence chains and scoped causal evidence",
            ),
            CommandEffect::read_only_db(
                "causal estimate",
                "Estimate causal uplift from persisted or scoped causal evidence",
            ),
            CommandEffect::read_only(
                "regress explain",
                "Build a regression-causality capsule from explicit structured artifacts",
            ),
            CommandEffect::read_only("check", "Quick posture summary"),
            CommandEffect::read_only("claim list", "List executable claims from claims.yaml"),
            CommandEffect::read_only("claim show", "Inspect one executable claim"),
            CommandEffect::read_only(
                "claim verify",
                "Verify executable claim evidence without mutating source records",
            ),
            CommandEffect::read_only_db(
                "capture suggest",
                "Suggest ambient capture candidates from session evidence without durable mutation",
            ),
            CommandEffect::read_only("config get", "Read one merged config key"),
            CommandEffect::read_only("config show", "Show merged config values"),
            CommandEffect::read_only_db(
                "conflict cluster",
                "Cluster persisted contradiction/conflict evidence without mutation",
            ),
            CommandEffect::read_only_db(
                "conflict explain",
                "Explain persisted contradiction/conflict evidence",
            ),
            CommandEffect::read_only_db(
                "conflict list",
                "List persisted contradiction/conflict evidence",
            ),
            CommandEffect::read_only_db("context", "Assemble context pack (reads only)"),
            CommandEffect::read_only_db("context-show", "Show a persisted context pack"),
            CommandEffect::read_only_db(
                "decide list",
                "List durable decision memories and supersede-chain heads",
            ),
            CommandEffect::read_only_db(
                "decide revisit",
                "List due or near-due decision revisit reminders",
            ),
            CommandEffect::read_only_db("orient", "Assemble read-only agent orientation bundle"),
            CommandEffect::read_only_db(
                "orient decisions",
                "Read due decision revisit reminders for orientation output",
            ),
            CommandEffect::read_only("completion", "Generate shell completion scripts"),
            CommandEffect::read_only_db("db status", "Report database status"),
            CommandEffect::read_only_db("db check", "Check database integrity"),
            CommandEffect::read_only_db(
                "db inspect",
                "Inspect rows from one database table without mutation",
            ),
            CommandEffect::read_only_db(
                "db reindex",
                "Preview database-derived index rebuild work",
            ),
            CommandEffect::read_only_db("db migrations", "List database migrations"),
            CommandEffect::read_only_db("curate candidates", "List curation candidates"),
            CommandEffect::read_only_db(
                "curate doctor",
                "Diagnose memory debt from persisted memory state",
            ),
            CommandEffect::read_only_db(
                "health scorecard",
                "Summarize memory-store health from debt, gaps, trust, redundancy, and graph signals",
            ),
            CommandEffect::read_only_db(
                "curate show",
                "Inspect a single curation candidate read-only with apply preview",
            ),
            CommandEffect::read_only_db("curate validate", "Validate curation candidate"),
            CommandEffect::read_only("demo list", "List demo manifests"),
            CommandEffect::read_only_db("demo show", "Show persisted demo audit rows"),
            CommandEffect::read_only_db("demo verify", "Verify demo artifacts"),
            CommandEffect::read_only_db("diag advisory-lock", "Inspect advisory-lock diagnostics"),
            CommandEffect::read_only_db(
                "diag agentsmd-drift",
                "Report AGENTS.md bridge drift: stale export, file-vs-memory contradictions, missing rules",
            ),
            CommandEffect::read_only_db("diag artifacts", "Inspect artifact diagnostics"),
            CommandEffect::read_only_db(
                "diag build-admission",
                "Inspect build-admission diagnostics",
            ),
            CommandEffect::read_only_db("diag causal-edge", "Inspect causal-edge diagnostics"),
            CommandEffect::read_only_db("diag claims", "Inspect claim diagnostics"),
            CommandEffect::read_only_db("diag contention", "Inspect contention diagnostics"),
            CommandEffect::read_only_db(
                "diag curation-candidate",
                "Inspect curation-candidate diagnostics",
            ),
            CommandEffect::read_only_db("diag database-skew", "Inspect database-skew diagnostics"),
            CommandEffect::read_only_db("diag dependencies", "Inspect dependency diagnostics"),
            CommandEffect::read_only_db("diag disk-pressure", "Inspect disk-pressure diagnostics"),
            CommandEffect::read_only_db(
                "diag environment-attestation",
                "Inspect environment attestation diagnostics",
            ),
            CommandEffect::read_only_db("diag graph", "Inspect graph diagnostics"),
            CommandEffect::read_only_db(
                "diag graph-snapshot",
                "Inspect graph-snapshot diagnostics",
            ),
            CommandEffect::read_only_db("diag host-profile", "Inspect host-profile diagnostics"),
            CommandEffect::read_only_db("diag incident", "Inspect incident diagnostics"),
            CommandEffect::read_only_db("diag integrity", "Inspect storage integrity diagnostics"),
            CommandEffect::read_only_db(
                "diag memory-validity",
                "Inspect memory-validity diagnostics",
            ),
            CommandEffect::read_only_db(
                "diag model-registry",
                "Inspect model-registry diagnostics",
            ),
            CommandEffect::read_only_db("diag pack-latest", "Inspect latest pack diagnostics"),
            CommandEffect::read_only(
                "diag resource-admission",
                "Preview a resource admission decision from explicit inputs",
            ),
            CommandEffect::read_only_db("diag plan-cache", "Inspect plan-cache diagnostics"),
            CommandEffect::read_only_db(
                "diag provenance",
                "Inspect live provenance freshness diagnostics",
            ),
            CommandEffect::read_only_db("diag quarantine list", "List quarantine entries"),
            CommandEffect::read_only_db("diag quarantine show", "Show single quarantine entry"),
            CommandEffect::read_only_db("diag search", "Inspect search diagnostics"),
            CommandEffect::read_only(
                "diag store-integrity",
                "Inspect explicit read-fence and write-immune diagnostics",
            ),
            CommandEffect::read_only_db("diag streams", "Show streams status"),
            CommandEffect::read_only(
                "diag toolchain-provenance",
                "Inspect observed toolchain provenance without mutating state",
            ),
            CommandEffect::read_only_db("diag tripwire", "Inspect tripwire diagnostics"),
            CommandEffect::read_only_db("diag write-owner", "Inspect write-owner diagnostics"),
            CommandEffect::read_only_db("diag write-spool", "Inspect write-spool diagnostics"),
            CommandEffect::read_only_db("doctor", "Run health checks"),
            CommandEffect::read_only("eval list", "List evaluation scenarios"),
            CommandEffect::read_only("eval report", "Summarize evaluation fixture reports"),
            CommandEffect::read_only("eval run", "Run evaluation (reads fixtures)"),
            CommandEffect::read_only_db(
                "economy report",
                "Report DB-backed memory economy metrics without mutation",
            ),
            CommandEffect::read_only_db(
                "economy score",
                "Score one persisted memory economy artifact without mutation",
            ),
            CommandEffect::read_only_db(
                "economy simulate",
                "Simulate attention budgets from persisted economy metrics without mutation",
            ),
            CommandEffect::read_only_db(
                "economy prune-plan",
                "Plan report-only memory economy pruning without mutation",
            ),
            CommandEffect::read_only_db(
                "focus explain",
                "Explain passive active-memory focus state",
            ),
            CommandEffect::read_only_db("focus show", "Show passive active-memory focus state"),
            CommandEffect::read_only_db(
                "focus suggest",
                "Suggest focus areas from recent CASS spans and graph centrality (bd-sg5si Phase 1: schema scaffold)",
            ),
            CommandEffect::read_only_db("graph articulation", "List graph articulation points"),
            CommandEffect::read_only_db(
                "graph betweenness",
                "Compute graph betweenness centrality",
            ),
            CommandEffect::read_only_db("graph centrality", "Compute graph centrality metrics"),
            CommandEffect::read_only_db("graph communities", "Compute graph communities"),
            CommandEffect::read_only_db(
                "graph diff",
                "Temporal structural diff between two persisted graph snapshots: add/remove sets, fingerprint-matched community deltas, persisted-centrality movers (ADR 0066); never recomputes centrality inline",
            ),
            CommandEffect::read_only_db("graph explain-link", "Explain graph link evidence"),
            CommandEffect::read_only_db("graph export", "Export graph projection report"),
            CommandEffect::read_only_db("graph hits", "Compute graph HITS centrality"),
            CommandEffect::read_only_db("graph k-core", "Compute graph k-core decomposition"),
            CommandEffect::read_only_db("graph louvain", "Compute graph Louvain communities"),
            CommandEffect::read_only_db("graph neighborhood", "Inspect graph neighborhood"),
            CommandEffect::read_only_db("graph pagerank", "Compute graph PageRank scores"),
            CommandEffect::read_only_db("graph path", "Find graph shortest path"),
            CommandEffect::read_only(
                "handoff completion-audit",
                "Audit objective completion evidence without mutation",
            ),
            CommandEffect::read_only("handoff inspect", "Inspect handoff capsule"),
            CommandEffect::read_only(
                "handoff preview",
                "Plan handoff capsule contents without writing",
            ),
            CommandEffect::read_only_db("artifact inspect", "Inspect artifact metadata"),
            CommandEffect::read_only_db("artifact list", "List registered artifacts"),
            CommandEffect::read_only_db(
                "attest memory",
                "Inspect memory attestation inputs and verdicts",
            ),
            CommandEffect::read_only_db(
                "attest pack",
                "Inspect context-pack attestation inputs and verdicts",
            ),
            CommandEffect::read_only_db(
                "attest query",
                "Inspect query attestation inputs and verdicts",
            ),
            CommandEffect::read_only_db("health", "Quick health check"),
            CommandEffect::read_only("help", "Print help"),
            CommandEffect::read_only_db("history", "Show persisted memory history summary"),
            CommandEffect::read_only(
                "hook git-readiness",
                "Inspect local Git hook-chain readiness without mutation",
            ),
            CommandEffect::read_only(
                "hook claude-code",
                "Print Claude Code recall/journal harness hook plan by default",
            ),
            CommandEffect::read_only(
                "hook codex",
                "Print Codex recall/journal harness hook plan by default",
            ),
            CommandEffect::read_only(
                "hook gemini",
                "Report Gemini harness hook support posture without guessing",
            ),
            CommandEffect::read_only(
                "hook status",
                "Inspect managed harness hook posture without mutating settings",
            ),
            CommandEffect::read_only_db(
                "impact",
                "Estimate impact from persisted graph and memory state",
            ),
            CommandEffect::read_only_db("index status", "Show index status"),
            CommandEffect::read_only(
                "index vacuum",
                "Preview reclaimable derived index artifacts without mutation; --apply reclaims retained generations beyond the retention bound under the publication lease",
            ),
            CommandEffect::read_only_db("insights", "Render persisted insight summaries"),
            CommandEffect::read_only("install check", "Inspect install posture"),
            CommandEffect::read_only("install plan", "Plan install without mutation"),
            CommandEffect::read_only("introspect", "Introspect ee metadata"),
            CommandEffect::read_only_db("job list", "List available steward job types"),
            CommandEffect::read_only_db("job show", "Show steward job row details"),
            CommandEffect::read_only(
                "lab replay",
                "Re-assembles a pack against a previously captured frozen episode and reports whether the captured inputs still produce a matching pack hash (N15.4 / bd-17c65.14.15.5)",
            ),
            CommandEffect::read_only(
                "lab counterfactual",
                "Replays a frozen episode with single-input swaps and surfaces the pack diff between the captured pack and the counterfactual pack (N15.5 / bd-17c65.14.15.6)",
            ),
            CommandEffect::read_only(
                "lab generate-workload",
                "Generate a deterministic workload report without persisting it",
            ),
            CommandEffect::read_only(
                "lab promote-workload",
                "Preview workload promotion and admission without persisting it",
            ),
            CommandEffect::read_only_db(
                "journal list",
                "List append-only journal entries newest-first",
            ),
            CommandEffect::read_only_db(
                "journal show",
                "Show one journal entry with structured sidecar and redaction report",
            ),
            CommandEffect::read_only_db(
                "ask --read-only",
                "Extractive question answering without audit writes or automatic migrations",
            ),
            CommandEffect::read_only_db(
                "recall",
                "Code-anchored reverse lookup from paths, symbols, or a git diff to anchored memories",
            ),
            // `similar` moved to append_only_write_commands (bd-czj3e): it
            // opens a WRITABLE audit handle and appends a retrieval audit row.
            CommandEffect::read_only_db(
                "learn agenda",
                "Show learning agenda with prioritized gaps",
            ),
            CommandEffect::read_only_db(
                "learn cluster",
                "Cluster learning evidence without durable mutation",
            ),
            CommandEffect::read_only_db(
                "learn gaps",
                "Mine query-miss demand into learning gap templates",
            ),
            CommandEffect::read_only_db("learn summary", "Show learning summary statistics"),
            CommandEffect::read_only_db("learn uncertainty", "Show uncertainty estimates"),
            CommandEffect::read_only_db("lens explain", "Explain a persisted lens projection"),
            CommandEffect::read_only_db("lens list", "List available persisted lens projections"),
            CommandEffect::read_only_db(
                "maintenance status",
                "Report maintenance job availability",
            ),
            CommandEffect::read_only_db("migrate status", "Report pending schema migrations"),
            CommandEffect::read_only("mcp manifest", "Inspect optional MCP adapter manifest"),
            CommandEffect::read_only("mcp validate", "Validate optional MCP adapter contracts"),
            CommandEffect::read_only_db(
                "shadow run",
                "Execute a shadowable policy evaluator offline (ADR 0070): label extraction, seeded read-only query replay, deterministic candidate sweep, report emission; writes nothing",
            ),
            CommandEffect::read_only_db("memory drift", "Report read-only memory provenance drift"),
            CommandEffect::read_only_db("memory history", "Show memory revision history"),
            CommandEffect::read_only_db("memory list", "List memories"),
            CommandEffect::read_only_db("memory show", "Show memory details"),
            // The CLI emits "mesh hello-responder status"; the bare parent path
            // was never produced by `extract_command_path`, so this row matched
            // nothing while the real status path went undeclared (bd-qked7).
            // The sibling run/register/unregister paths are declared in
            // `external_io_write_commands`.
            CommandEffect::read_only_db(
                "mesh hello-responder status",
                "Inspect mesh hello-responder status",
            ),
            CommandEffect::read_only_db("mesh init", "Preview mesh initialization state"),
            CommandEffect::read_only_db(
                "mesh ledger",
                "Inspect receiver-local mesh import decisions",
            ),
            CommandEffect::read_only_db("mesh peer list", "List mesh peers"),
            CommandEffect::read_only_db("mesh peer show", "Show one mesh peer"),
            CommandEffect::read_only_db(
                "mesh peer unknown-attempt",
                "Inspect unknown mesh-peer admission attempts",
            ),
            CommandEffect::read_only_db("mesh peers", "List mesh peers"),
            CommandEffect::read_only_db(
                "mesh preview-grant",
                "Preview a mesh sharing grant without persisting it",
            ),
            CommandEffect::read_only_db("mesh status", "Inspect mesh status"),
            CommandEffect::read_only_db("model list", "List model registry entries"),
            CommandEffect::read_only_db("model status", "Inspect model registry status"),
            CommandEffect::read_only_db("outcome quarantine list", "List feedback quarantine rows"),
            CommandEffect::read_only_db("pack diff", "Compare persisted pack ledgers"),
            CommandEffect::read_only_db("pack replay", "Inspect persisted pack ledger"),
            CommandEffect::read_only(
                "perf budget check",
                "Check normalized performance artifact budget posture",
            ),
            CommandEffect::read_only(
                "perf compare",
                "Compare normalized performance artifact summaries",
            ),
            CommandEffect::read_only(
                "perf explain-latency",
                "Explain latency stages for a normalized performance artifact",
            ),
            CommandEffect::read_only(
                "perf live",
                "Stream read-only performance snapshots for swarm observability",
            ),
            CommandEffect::read_only(
                "perf prompt-budget",
                "Estimate prompt budget posture without durable mutation",
            ),
            CommandEffect::read_only(
                "perf snapshot",
                "Emit one read-only performance snapshot for swarm observability",
            ),
            CommandEffect::read_only_db(
                "plan recipe list",
                "List built-in and stored workspace recipes",
            ),
            CommandEffect::read_only_db(
                "plan recipe show",
                "Show a built-in or stored workspace recipe",
            ),
            CommandEffect::read_only(
                "preflight show",
                "Read a persisted preflight run from the workspace-local store",
            ),
            CommandEffect::read_only_db(
                "preflight check",
                "Retrieve advisory command-risk memories without granting or revoking execution authority",
            ),
            CommandEffect::read_only_db(
                "preflight guard",
                "Alias for read-only advisory command-risk memory retrieval",
            ),
            CommandEffect::read_only("plan goal", "Recommends recipes for goals"),
            CommandEffect::read_only_db(
                "plan explain",
                "Explains recipe provenance and applicability",
            ),
            CommandEffect::read_only("plan recommend", "Recommends recipes for tasks"),
            CommandEffect::read_only_db("playbook list", "List procedural rules in playbook form"),
            CommandEffect::read_only_db(
                "procedure drift",
                "Inspect procedure maturity and feedback drift signals",
            ),
            CommandEffect::read_only_db(
                "procedure export",
                "Render a persisted procedure artifact",
            ),
            CommandEffect::read_only_db("procedure list", "List persisted procedures"),
            CommandEffect::read_only_db("procedure show", "Show persisted procedure details"),
            CommandEffect::read_only_db(
                "procedure verify",
                "Verify a persisted procedure against evidence sources",
            ),
            CommandEffect::read_only_db("rationale list", "List safe rationale traces"),
            CommandEffect::read_only_db("rationale show", "Show a safe rationale trace"),
            CommandEffect::read_only_db(
                "reflect request-ledger diagnostics",
                "Inspect reflection request ledger diagnostics without exposing secret payloads",
            ),
            CommandEffect::read_only(
                "profile config plan",
                "Plan operating profile configuration without writing files",
            ),
            CommandEffect::read_only_db("proof admit", "Preview proof admission status"),
            CommandEffect::read_only_db("proof status", "Inspect proof status"),
            CommandEffect::read_only_db(
                "proximity",
                "Compute memory proximity from persisted graph state",
            ),
            CommandEffect::read_only(
                "recorder tail",
                "Recorder tail reads persisted recorder events without mutation",
            ),
            CommandEffect::read_only(
                "recorder follow",
                "Recorder follow streams persisted recorder events without mutation",
            ),
            CommandEffect::read_only(
                "recorder import",
                "Recorder import planning is read-only unless explicitly promoted to execution",
            ),
            CommandEffect::read_only(
                "recorder events list",
                "List persisted recorder events without mutation",
            ),
            CommandEffect::read_only(
                "recorder flight replay",
                "Replay a flight-recorder trace without mutating source state",
            ),
            CommandEffect::read_only(
                "rehearse plan",
                "Rehearsal planning validates command specs and estimates side-path artifacts",
            ),
            CommandEffect::read_only(
                "rehearse inspect",
                "Rehearsal inspection reads a prior manifest and verifies hashes",
            ),
            CommandEffect::read_only(
                "rehearse promote-plan",
                "Rehearsal promotion planning reads a manifest and emits a conservative checklist",
            ),
            CommandEffect::read_only(
                "review session",
                "Analyze session evidence spans for curation candidates",
            ),
            CommandEffect::read_only_db(
                "resume",
                "Assemble the session-resume bundle: recent episodic sessions, revisit-conditioned decisions, queued-tag items, staleness flags, nearby stores (reads only)",
            ),
            CommandEffect::read_only(
                "sandbox diff",
                "Compare sandbox state with workspace state without applying changes",
            ),
            CommandEffect::read_only_db("rule list", "List procedural rules"),
            CommandEffect::read_only_db(
                "rule provenance",
                "Inspect the rule-to-memory provenance ego graph",
            ),
            CommandEffect::read_only_db("rule show", "Show procedural rule"),
            CommandEffect::read_only("schema export", "Export public response schemas"),
            CommandEffect::read_only("schema list", "List response schemas"),
            // `search`, `search --all-workspaces` and `similar` all moved to
            // append_only_write_commands (bd-czj3e): every user retrieval
            // appends a retrieval audit row.
            CommandEffect::read_only_db(
                "sentinel explain",
                "Explain sentinel specifications and prior results",
            ),
            CommandEffect::read_only_db(
                "share preview",
                "Preview outbound mesh sharing without exporting data",
            ),
            CommandEffect::read_only_db("show", "Show a persisted memory or artifact"),
            CommandEffect::read_only(
                "situation classify",
                "Classify task into situation category",
            ),
            CommandEffect::read_only("situation compare", "Compare two situations (dry-run)"),
            CommandEffect::read_only_db("situation explain", "Explain a stored situation"),
            CommandEffect::read_only("situation link", "Plan situation link (dry-run)"),
            CommandEffect::read_only_db("situation show", "Show stored situation details"),
            CommandEffect::read_only_db("status", "Report workspace status"),
            CommandEffect::read_only_db("subscribe poll", "Poll subscription state"),
            CommandEffect::read_only_db(
                "subscribe stream",
                "Stream subscription state without writing durable records",
            ),
            CommandEffect::read_only(
                "support inspect",
                "Verify and inspect a redacted support bundle manifest",
            ),
            CommandEffect::read_only(
                "session-budget plan",
                "Advisory deterministic plan for cheapest useful next command given ledger and posture",
            ),
            CommandEffect::read_only_db("swarm brief", "Report read-only swarm coordination brief"),
            CommandEffect::read_only_db(
                "swarm next-action",
                "Recommend the next swarm action without claiming work",
            ),
            CommandEffect::read_only_db(
                "swarm repair-plan",
                "Render an advisory degraded-stack repair plan without executing repairs",
            ),
            CommandEffect::read_only_db(
                "swarm work-packet",
                "Render a swarm work packet without mutating coordination state",
            ),
            CommandEffect::read_only_db("task-frame show", "Show passive task-frame state"),
            // The six read-only members of the `team` surface (bd-qked7).
            // Each was established by reading its delegate, not by name:
            // `team activity` -> mesh::team::list_team_activity,
            // `team doctor` -> inspect_team_health (reads the origin stream and
            // probes the key-store PLATFORM constant; it never opens the store
            // for create), `team idp status` -> team_idp_status,
            // `team members list` -> handle_team_status (a cli->cli delegate),
            // `team projects list` -> list_team_projects,
            // `team status` -> local_team_status + collect_team_member_freshness.
            // None of those bodies reaches a SQL writer or a filesystem write;
            // the only `insert(` in the group is a local HashSet in
            // list_team_activity. A false read_only here is worse than the
            // missing declaration was, so treat a change to any of these six
            // delegates as a reason to re-derive the class.
            CommandEffect::read_only_db(
                "team activity",
                "List team activity projected from the local mesh origin stream",
            ),
            CommandEffect::read_only_db(
                "team doctor",
                "Report team genesis, key-store platform, and membership health checks",
            ),
            CommandEffect::read_only_db(
                "team idp status",
                "Report the team IdP policy generation and pinned OIDC provider",
            ),
            CommandEffect::read_only_db(
                "team members list",
                "List team members through the read-only team status projection",
            ),
            CommandEffect::read_only_db(
                "team port show",
                "Report the folded team hello port and genesis hash without mutation",
            ),
            CommandEffect::read_only_db(
                "team projects list",
                "List team projects recorded in the local team store",
            ),
            CommandEffect::read_only_db(
                "team status",
                "Report local team status and per-member freshness",
            ),
            CommandEffect::read_only_db(
                "timeline",
                "Reconstruct read-only memory state for a topic at a historical timestamp",
            ),
            CommandEffect::read_only_db(
                "trust report",
                "Audit confidence calibration and outcome-backed reliability",
            ),
            CommandEffect::read_only_db("tripwire list", "List persisted tripwire rules"),
            CommandEffect::read_only("update", "Plan update without mutation"),
            CommandEffect::read_only_db(
                "verification broker lookup",
                "Look up verification broker state",
            ),
            CommandEffect::read_only_db(
                "verification closeout capsule",
                "Render a verification closeout capsule",
            ),
            CommandEffect::read_only_db(
                "verification closure-guidance",
                "Render verification closure guidance",
            ),
            CommandEffect::read_only_db("verification proofs", "List verification proofs"),
            CommandEffect::read_only_db(
                "verification rch blockers",
                "List RCH verification blockers",
            ),
            CommandEffect::read_only_db("verification rch runs", "List RCH verification runs"),
            CommandEffect::read_only_db(
                "verification rch topology-audit",
                "Audit RCH topology closure for path-dep and crate-graph gaps",
            ),
            CommandEffect::read_only_db(
                "verify broker lookup",
                "Look up verification broker state",
            ),
            CommandEffect::read_only_db(
                "verify closeout capsule",
                "Render a verification closeout capsule",
            ),
            CommandEffect::read_only_db(
                "verify closure-guidance",
                "Render verification closure guidance",
            ),
            CommandEffect::read_only_db("verify proofs", "List verification proofs"),
            CommandEffect::read_only_db("verify rch blockers", "List RCH verification blockers"),
            CommandEffect::read_only_db("verify rch runs", "List RCH verification runs"),
            CommandEffect::read_only_db(
                "verify rch topology-audit",
                "Audit RCH topology closure for path-dep and crate-graph gaps",
            ),
            CommandEffect::read_only("version", "Print version"),
            CommandEffect::read_only_db(
                "workspace hygiene",
                "Inspect workspace hygiene and coordination state",
            ),
            CommandEffect::read_only("workspace list", "List workspace aliases"),
            CommandEffect::read_only("workspace resolve", "Resolve workspace identity"),
            CommandEffect::read_only_db("why", "Explain memory selection"),
            CommandEffect::read_only_db("why-not", "Explain why a memory was not selected"),
        ]
    }

    fn derived_write_commands() -> Vec<CommandEffect> {
        vec![
            CommandEffect::derived_write(
                "index rebuild",
                vec![".ee/index/"],
                "Rebuild search indexes from database",
            ),
            CommandEffect::derived_write(
                "primer",
                vec!["primer_cache (db table)"],
                "Assemble the cached workspace primer; cache rows are derived and rebuildable (--no-persist is read-only)",
            ),
            CommandEffect::derived_write(
                "index reembed",
                vec![".ee/index/embeddings/"],
                "Rebuild semantic embeddings from database records",
            ),
            CommandEffect::derived_write(
                "search --recalibrate-now",
                vec![".ee/search/calibration.jsonl"],
                "Rewrite the derived search score calibration artifact from persisted feedback",
            ),
            CommandEffect::derived_write(
                "graph centrality-refresh",
                vec![".ee/graph/"],
                "Refresh derived graph centrality metrics",
            ),
            CommandEffect::derived_write(
                "graph feature-enrichment",
                vec![".ee/graph/"],
                "Refresh derived graph feature enrichments",
            ),
            CommandEffect::derived_write(
                "graph snapshot refresh",
                vec![".ee/graph/"],
                "Refresh derived graph snapshots from source database state",
            ),
        ]
    }

    fn degraded_unavailable_commands() -> Vec<CommandEffect> {
        Vec::new()
    }

    fn daemon_command_effect() -> CommandEffect {
        let mut effect = CommandEffect::external_io_write(
            "daemon",
            vec!["memories", "feedback_events", "audit_log"],
            vec![
                "$XDG_RUNTIME_DIR/ee/daemon.sock or ${TMPDIR:-/tmp}/ee-<uid>/daemon.sock",
                ".ee/daemon-jobs.jsonl",
            ],
            "daemon subcommand plus socket path plus workspace plus job type",
            "Run daemon status, foreground steward jobs, background steward scheduling, or UDS hot-mode lifecycle operations",
        );
        effect.idempotency = IdempotencyClass::DryRunAvailable;
        effect.dry_run_effect = Some(EffectClass::WorkspaceFileWrite);
        effect.mutation_contract = CommandMutationContract {
            side_effect_class: SideEffectClass::Mixed,
            transaction_scope: Some("daemon subcommand-specific operation"),
            idempotency_key: Some(
                "daemon subcommand plus socket path plus workspace plus job type",
            ),
            audit_surface: Some(
                "daemon job ledger or audit_log when a selected subcommand mutates",
            ),
            db_generation_effect: "subcommand-specific: status is read-only; foreground jobs may advance handler-owned state",
            index_generation_effect: "subcommand-specific: unchanged unless a selected steward job processes index work",
            dry_run_behavior: Some(
                "foreground --dry-run records planned daemon job rows and reports handler plans without committing handler mutations",
            ),
            recovery_behavior: "foreground and background job rows are persisted and recovered on restart",
            no_overwrite_behavior: Some(
                "daemon socket paths are guarded by same-UID socket checks; stop refuses regular files and stale unauthenticated sockets",
            ),
            degraded_code: None,
        };
        effect.runtime_contract = CommandRuntimeContract {
            runtime_class: RuntimeClass::MultiStage,
            default_budget_ms: Some(300_000),
            cancellation_points: &[
                "before_daemon_mode_dispatch",
                "before_socket_publish_or_probe",
                "before_foreground_job_schedule",
                "before_steward_handler",
                "before_daemon_job_row_commit",
            ],
            partial_progress_policy: "option-specific: status is read-only; foreground jobs persist planned rows before handler execution and terminal rows after completion",
            outcome_mapping: "success, usage_error, storage_error, policy_denied, or supervised job failure",
        };
        effect
    }

    fn external_io_write_commands() -> Vec<CommandEffect> {
        vec![
            Self::daemon_command_effect(),
            // Writes an encrypted recovery envelope OUTSIDE the workspace, so it
            // belongs here rather than with the workspace-file writers. It
            // implements its own `--dry-run` and honours it (`src/core/backup.rs`,
            // the `if !options.dry_run` guards). The constructor NO LONGER
            // supplies `dry_run_effect` (bd-3j6l3 -- it used to supply
            // `Some(ReadOnly)` to every command whether or not the flag
            // existed); this command is on `DRY_RUN_CAPABLE` instead. Removing
            // it from that roster would make `dry_run_refusal_message` refuse a
            // `--dry-run` that works today, and would also red
            // `mutating_dry_run_declarations_match_the_parser`, which asks Clap.
            CommandEffect::external_io_write(
                "backup keys export",
                Vec::new(),
                vec!["<--output-dir>/store-auth.recovery.json"],
                "output dir plus exported key ids",
                "Write an encrypted store-auth recovery envelope outside the workspace",
            ),
            CommandEffect::external_io_write(
                "demo run",
                vec!["audit_log"],
                vec![
                    "demo evidence root",
                    "manifest-declared demo artifact paths",
                ],
                "demo id plus manifest hash plus generated run id",
                "Execute safe demo manifest steps with audit ledger rows and evidence artifacts",
            ),
            CommandEffect::external_io_write(
                "lab swarm replay",
                vec!["audit_log"],
                vec![".ee/lab/swarm-replay/"],
                "workload id plus replay host profile plus generated run id",
                "Replay a swarm workload through subprocess execution and write replay evidence artifacts",
            ),
            CommandEffect::external_io_write(
                "mcp serve-stdio",
                Vec::new(),
                vec!["stdio JSON-RPC stream"],
                "process id plus stdio session",
                "Serve the optional MCP stdio adapter over process I/O",
            ),
            CommandEffect::external_io_write(
                "mesh auto-enroll",
                vec!["mesh_peers", "audit_log"],
                vec![
                    ".ee/auto_enroll_overrides.toml",
                    ".ee/discovery_denylist.toml",
                ],
                "tailscale peer set hash plus workspace id",
                "Probe mesh peers and persist reviewed auto-enrollment state",
            ),
            CommandEffect::external_io_write(
                "mesh hello-responder register",
                Vec::new(),
                vec!["user-scoped mesh-responder control socket"],
                "request nonce plus team id plus workspace id plus peer handle",
                "Register exact team routes with the same-EUID user-scoped responder owner",
            ),
            CommandEffect::external_io_write(
                "mesh hello-responder run",
                Vec::new(),
                vec![
                    "verified Tailscale listener",
                    "user-scoped mesh-responder control socket",
                ],
                "process id plus committed port plus control socket path",
                "Own the user-scoped Tailscale responder listener and same-EUID control channel",
            ),
            CommandEffect::external_io_write(
                "mesh hello-responder unregister",
                Vec::new(),
                vec!["user-scoped mesh-responder control socket"],
                "request nonce plus team id plus workspace id plus peer handle",
                "Remove exact team routes from the same-EUID user-scoped responder owner",
            ),
            CommandEffect::external_io_write(
                "model fetch",
                vec!["model_registry", "audit_log"],
                vec!["~/.local/share/ee/models/"],
                "model alias plus artifact content hash",
                "Fetch or import a model artifact and update the model registry",
            ),
            CommandEffect::external_io_write(
                "serve",
                Vec::new(),
                vec!["localhost HTTP/SSE listener"],
                "process id plus listener address",
                "Serve the optional localhost adapter",
            ),
            // The `team` paths that leave the machine (bd-qked7). These are
            // external I/O first and durable writers second, so they are
            // declared here rather than with the pure DB mutations: an agent
            // reading `durable_write` would not learn that the command dials a
            // peer, spawns `curl`, or binds a listener.
            //
            //   team fetch body          peer body fetch on a cache miss
            //                            (foreground_cli::fetch_pending_team_bodies_from_paths)
            //   team invite --wait       binds a TcpListener and serves one
            //                            bootstrap join plus first sync
            //   team join                TcpStream::connect_timeout to the inviter
            //   team members revalidate  tailscale socket + CLI probe
            //   team idp attest          curl HTTPS GET of the pinned JWKS
            //   team idp device --execute  curl HTTPS POST to the token endpoint
            //   team steward run-once    peer body fetch plus a delegated mesh sync
            CommandEffect::external_io_write(
                "team fetch body",
                vec!["mesh_body_cache_metadata"],
                vec!["mesh body cache"],
                "team id plus body cache key",
                "Fetch a represented team body, falling back to a peer fetch that fills the local body cache",
            ),
            CommandEffect::external_io_write(
                "team idp attest",
                vec![
                    "mesh_origin_events",
                    "mesh_origin_event_nonces",
                    "team_idp_token_replay",
                    "team_member_identity",
                ],
                vec!["<workspace>/.ee/keys/mesh/"],
                "id token jti plus team id",
                "Verify a local OIDC id token against the pinned JWKS over constrained HTTPS and attest the member identity",
            ),
            CommandEffect::external_io_write(
                "team idp device",
                Vec::new(),
                vec!["constrained curl HTTPS device-code exchange"],
                "team id plus device code",
                "Plan, or with --execute run, the RFC 8628 device-code ceremony through a constrained curl invocation",
            ),
            CommandEffect::external_io_write(
                "team invite",
                vec![
                    "team_pending_invites",
                    "team_invite_auth_floor",
                    "team_members",
                    "team_member_nodes",
                    "team_member_signing_keys",
                ],
                vec![
                    "<workspace>/.ee/keys/mesh/",
                    "bootstrap join TCP listener (--wait)",
                ],
                "team id plus invite endpoint plus invite expiry",
                "Mint a team invite, and with --wait serve one bootstrap join and first sync over a bound TCP listener",
            ),
            CommandEffect::external_io_write(
                "team join",
                vec![
                    "team_members",
                    "team_member_nodes",
                    "team_member_signing_keys",
                    "team_join_attempts",
                    "team_invite_auth_floor",
                    "mesh_peers",
                ],
                vec!["<workspace>/.ee/keys/mesh/"],
                "invite id plus joining node id",
                "Redeem an invite code against the inviter over TCP and persist the resulting membership and peer locator",
            ),
            CommandEffect::external_io_write(
                "team members revalidate",
                vec!["team_member_identity"],
                Vec::new(),
                "team id plus member identity snapshot",
                "Re-probe local tailnet identity and refresh the team member identity projection",
            ),
            CommandEffect::external_io_write(
                "team steward run-once",
                vec![
                    "team_members",
                    "team_member_nodes",
                    "team_member_signing_keys",
                    "team_projects",
                    "team_removal_acknowledgements",
                    "mesh_body_cache_metadata",
                    "mesh_peers",
                    "mesh_import_ledger",
                    "search_index_jobs",
                ],
                vec!["<workspace>/.ee/keys/mesh/"],
                "team id plus steward tick",
                "Run one team steward pass, then either fetch pending bodies from peers or delegate a single foreground mesh sync",
            ),
        ]
    }

    fn supervised_job_commands() -> Vec<CommandEffect> {
        vec![
            CommandEffect::supervised_job(
                "daemon foreground decay_sweep",
                vec!["memories", "feedback_events", "audit_log"],
                "Run the real score-decay steward handler in a bounded foreground daemon tick",
            ),
            CommandEffect::supervised_job(
                "daemon background",
                vec!["memories", "feedback_events", "audit_log"],
                "Run configured steward handlers on the daemon background scheduler",
            ),
            CommandEffect::supervised_job(
                "daemon foreground non-decay",
                vec!["memories", "feedback_events", "audit_log"],
                "Run real non-decay steward handlers in a bounded foreground daemon tick",
            ),
            CommandEffect::supervised_job(
                "job run",
                vec!["memories", "feedback_events", "audit_log"],
                "Run a steward job directly through the job interface",
            ),
            CommandEffect::supervised_job(
                "maintenance run",
                vec!["memories", "feedback_events", "audit_log"],
                "Run an explicit bounded maintenance job through the steward backend",
            ),
            CommandEffect::supervised_job(
                "maintenance graph-snapshot-prune",
                vec!["graph_snapshots", "audit_log"],
                "Prune expired graph snapshots through a bounded steward job",
            ),
        ]
    }

    fn append_only_write_commands() -> Vec<CommandEffect> {
        let mut ask = CommandEffect::append_only_write(
            "ask",
            vec!["audit_log"],
            "audit row id",
            "Answer from a read-only snapshot, then best-effort append cited retrieval or query-miss audit records",
        )
        .with_read_snapshot();
        // Ask has an explicit read-only mode, not a dry-run argument.
        ask.dry_run_effect = None;
        ask.mutation_contract.dry_run_behavior = None;
        ask.mutation_contract.recovery_behavior = "answers remain available when optional audit writes fail; --read-only suppresses those writes";
        let diag_pack_record = CommandEffect::append_only_write(
            "diag pack-record",
            vec!["pack_records", "audit_log"],
            "pack id",
            "Append one audited diagnostic pack record without overwriting an existing ID",
        );
        vec![
            ask,
            diag_pack_record,
            CommandEffect::append_only_write(
                "db check-integrity",
                vec!["audit_log"],
                "audit row id",
                "Run full database integrity verification and append an audit row",
            ),
            // Resume is read-only on a well-formed capsule, but a capsule that
            // fails HMAC verification -- or is loaded with verification bypassed
            // -- durably appends an audit row before the command returns
            // (`record_handoff_hmac_verify_failure_audit`,
            // `record_handoff_insecure_load_audit`). A security record of a
            // bypass is exactly what must survive, so the declaration follows
            // the write rather than the write being suppressed to fit a
            // read-only declaration. Parallel to `db check-integrity` (bd-czj3e).
            CommandEffect::append_only_write(
                "handoff resume",
                vec!["audit_log"],
                "audit row id",
                "Render handoff resume payload, appending an audit row when capsule HMAC verification fails or is bypassed",
            ),
            // `search`, `search --all-workspaces` and `similar` are RETRIEVAL
            // surfaces that nevertheless append to the audit hash-chain on
            // every user-facing invocation, so they are declared for what they
            // do rather than for what they look like (bd-czj3e).
            //
            // src/core/search.rs:7652-7669, in production
            // run_search_with_performance_and_filters_with_cx_and_reconcile_timeout:
            //
            //     // bd-l8dn0. Record this retrieval. ...
            //     // the audit write takes the flock write gate, ...
            //     if record_retrieval_audit && !retrieval_only
            //         && let Some(facts) = &run.audit_facts
            //         && let Ok(connection) = DbConnection::open_file(&database_path)
            //
            // `DbConnection::open_file` is DatabaseConfig::file = ReadWrite, and
            // the write-owner gate is taken for exactly (File, ReadWrite) at
            // OPEN (src/db/mod.rs:1130-1137) -- so the flock is taken even
            // before a row is written. run_search_with_performance_and_filters
            // passes record_retrieval_audit: true, so a bare `ee search`
            // reaches it with no flag. `ee similar` does the same at :8169,
            // deliberately: 76991fe96 "give ee similar a writable audit handle
            // so its rows are not lost".
            //
            // The rows land in `audit_log` via SearchAuditBatch ->
            // DbConnection::insert_audit_batch -> AUDIT_INSERT_SQL_PREFIX
            // (src/db/mod.rs:28410), inside one transaction that extends the
            // audit hash chain.
            //
            // NOT FIXED BY SUPPRESSING THE WRITE. Retrieval audit feeds
            // `never_retrieved`, decay and trust (ADR 0071); bd-b9dmp records
            // that passing a read-only handle here silently lost every row and
            // scored retrieved memories as never-retrieved. The declaration was
            // the wrong half.
            //
            // `search --all-workspaces` is the same writer, fanned out:
            // handle_search_all_workspaces (src/cli/mod.rs:49345) calls the
            // recording `run_search` once per registered workspace, up to
            // ALL_WORKSPACES_SCAN_CAP = 16 of them, so a single invocation
            // appends an audit row to as many as sixteen databases. Its old
            // description called it a "Read-only diagnostic ... inspection
            // surface"; the read-only handle it opens at :49282 is only for
            // enumerating the workspaces, not for the searches it then runs.
            CommandEffect::append_only_write(
                "search",
                vec!["audit_log"],
                "retrieval audit row id",
                "Search memories, appending one retrieval audit row per user-facing query",
            ),
            CommandEffect::append_only_write(
                "search --all-workspaces",
                vec!["audit_log"],
                "retrieval audit row id",
                "Run one query across every workspace registered in the addressed database plus the global lane, rows labeled per workspace, appending one retrieval audit row to each workspace searched; inspection surface, never a pack input",
            ),
            CommandEffect::append_only_write(
                "similar",
                vec!["audit_log"],
                "retrieval audit row id",
                "Find embedding-native nearest-neighbor memories for a persisted seed memory, appending one retrieval audit row",
            ),
            CommandEffect::append_only_write(
                "artifact register",
                vec!["artifacts", "artifact_links", "audit_log"],
                "content hash",
                "Register artifact metadata keyed by content hash",
            ),
            // Measured per triggering input (bd-kmsd6 probe v4, c10175): a first
            // import writes sessions, evidence_spans, search_index_jobs,
            // import_ledger and workspace_generations; any re-import writes
            // import_ledger; audit_log comes from a refresh after the session
            // grew, or from a redacted span. No path writes memories. A
            // storeless address is refused (bd-hin8m), so workspaces is not
            // written.
            CommandEffect::append_only_write(
                "import cass",
                vec![
                    "sessions",
                    "evidence_spans",
                    "search_index_jobs",
                    "import_ledger",
                    "workspace_generations",
                    "audit_log",
                ],
                "source hash",
                "Import from CASS sessions",
            ),
            CommandEffect::append_only_write(
                "import jsonl",
                vec!["memories", "audit_log"],
                "source hash",
                "Import from JSONL export",
            ),
            CommandEffect::append_only_write(
                "import eidetic-legacy",
                vec!["memories", "audit_log"],
                "source hash",
                "Import from legacy Eidetic export",
            ),
            CommandEffect::append_only_write(
                "import agentsmd",
                vec![
                    "curation_candidates",
                    "evidence_spans",
                    "sessions",
                    "audit_log",
                ],
                "deterministic candidate id over (file, statement text)",
                "Import rule-like AGENTS.md statements as pending curation candidates",
            ),
            // Measured per triggering input (bd-hb5os): pack_omissions under a
            // tight budget, pack_baselines from the `ee pack <query>` shim when
            // an agent name is set, pack_evidence_items from imported CASS
            // evidence.
            CommandEffect::append_only_write(
                "pack build",
                vec![
                    "pack_records",
                    "pack_items",
                    "pack_candidate_impressions",
                    "pack_omissions",
                    "pack_evidence_items",
                    "pack_rule_items",
                    "pack_baselines",
                    "audit_log",
                ],
                "pack hash",
                "Persist a context pack keyed by deterministic pack hash",
            ),
            CommandEffect::append_only_write(
                "reflect propose",
                vec!["reflection_request_ledger"],
                "requestHash",
                "Create an external reflection request artifact and non-secret replay ledger row",
            ),
            CommandEffect::append_only_write(
                "mesh import",
                vec!["mesh_peers", "mesh_import_ledger", "search_index_jobs"],
                "origin peer cursor plus event content hash",
                "Import a mesh artifact by replaying idempotent peer events",
            ),
            // Honest classification (bd-6dmhw): the production sync transport
            // is a deliberate no-op until M1 real transport lands
            // (bd-tc-epic-qzk7o.3.x); the command performs no peer network
            // I/O today. Restore external_io_write with the transport.
            CommandEffect::append_only_write(
                "mesh sync",
                vec!["mesh_peers", "mesh_import_ledger", "search_index_jobs"],
                "origin peer cursor plus event content hash",
                "Run one foreground sync cycle over locally available peer state; network transport is deferred",
            ),
            // `ee team sync` checks team posture and then calls
            // `handle_mesh_sync` directly (src/cli/team.rs), so its blast
            // radius is `mesh sync`'s, not a separate one. Declared next to it
            // deliberately: if the `mesh sync` surface changes, this row has to
            // change with it (bd-qked7).
            CommandEffect::append_only_write(
                "team sync",
                vec!["mesh_peers", "mesh_import_ledger", "search_index_jobs"],
                "origin peer cursor plus event content hash",
                "Refuse while the team is paused, then delegate one foreground mesh sync cycle",
            ),
            CommandEffect::append_only_write(
                "verification ingest",
                vec!["audit_log"],
                "verification evidence content hash",
                "Ingest verification evidence into the audit ledger",
            ),
            CommandEffect::append_only_write(
                "verification rch ingest",
                vec!["rch_verify_runs"],
                "rch proof command hash plus run id",
                "Ingest RCH verification run evidence",
            ),
            CommandEffect::append_only_write(
                "verification record",
                vec!["audit_log"],
                "verification record content hash",
                "Record verification evidence in the audit ledger",
            ),
            CommandEffect::append_only_write(
                "verify ingest",
                vec!["audit_log"],
                "verification evidence content hash",
                "Ingest verification evidence into the audit ledger",
            ),
            CommandEffect::append_only_write(
                "verify rch ingest",
                vec!["rch_verify_runs"],
                "rch proof command hash plus run id",
                "Ingest RCH verification run evidence",
            ),
            CommandEffect::append_only_write(
                "verify record",
                vec!["audit_log"],
                "verification record content hash",
                "Record verification evidence in the audit ledger",
            ),
        ]
    }

    fn durable_write_commands() -> Vec<CommandEffect> {
        vec![
            CommandEffect::durable_write(
                "graph suggest-links",
                vec!["curation_candidates"],
                "Predict missing memory links with bounded, typed, explained blended scoring (ADR 0066); default is a read-only report, --propose writes link/contradiction-review curation candidates (dedup on re-propose), never links directly",
            ),
            CommandEffect::durable_write(
                "conflict resolve",
                vec![
                    "memories",
                    "memory_links",
                    "memory_tags",
                    "search_index_jobs",
                    "audit_log",
                ],
                "Resolve one live conflict pair via verb-mapped EXISTING audited atoms (decide record / memory expire / memory link / memory tags); dry-run plan by default, --apply executes (ADR 0066)",
            ),
            CommandEffect::durable_write(
                "diagnose-error",
                vec!["error_fingerprints"],
                "Diagnose a tool error against the fingerprint recall store; --record persists its fingerprint",
            ),
            CommandEffect::durable_write(
                "bootstrap apply",
                vec![
                    "curation_candidates",
                    "memories",
                    "procedural_rules",
                    "rule_source_memories",
                    "rule_tags",
                    "search_index_jobs",
                    "audit_log",
                ],
                "Apply an approved docs bootstrap run through curation (routes through curate apply with audit)",
            ),
            CommandEffect::durable_write(
                "causal promote-plan",
                vec!["curation_candidates", "audit_log"],
                "Plan causal promotion and persist reviewed curation candidates when evidence clears thresholds",
            ),
            CommandEffect::durable_write(
                "curate accept",
                vec!["curation_candidates", "procedural_rules", "audit_log"],
                "Accept a curation candidate",
            ),
            CommandEffect::durable_write(
                "curate auto-promote",
                vec!["memories", "search_index_jobs", "audit_log"],
                "Threshold-based memory level promotion; dry-run by default, --apply routes through memory.level_transition",
            ),
            CommandEffect::durable_write(
                "curate apply",
                vec![
                    "curation_candidates",
                    "plan_recipes",
                    "memories",
                    "procedural_rules",
                    "rule_source_memories",
                    "rule_tags",
                    "search_index_jobs",
                    "audit_log",
                ],
                "Apply a curation candidate",
            ),
            CommandEffect::durable_write(
                "curate disposition",
                vec!["curation_candidates", "audit_log"],
                "Record curation disposition",
            ),
            CommandEffect::durable_write(
                "curate merge",
                vec!["curation_candidates", "memories", "audit_log"],
                "Merge curation candidates",
            ),
            CommandEffect::durable_write(
                "curate propose-derived",
                vec!["curation_candidates", "audit_log"],
                "Persist derived curation proposals for explicit review",
            ),
            CommandEffect::durable_write(
                "curate reject",
                vec!["curation_candidates", "audit_log"],
                "Reject a curation candidate",
            ),
            CommandEffect::durable_write(
                "curate retire",
                vec![
                    "curation_candidates",
                    "memories",
                    "search_index_jobs",
                    "audit_log",
                ],
                "Retire an accepted curation artifact through audited memory/index updates",
            ),
            CommandEffect::durable_write(
                "curate snooze",
                vec!["curation_candidates", "audit_log"],
                "Snooze a curation candidate",
            ),
            CommandEffect::durable_write(
                "curate tombstone",
                vec![
                    "curation_candidates",
                    "memories",
                    "search_index_jobs",
                    "audit_log",
                ],
                "Tombstone a curation candidate and related memory state without deleting records",
            ),
            CommandEffect::durable_write(
                "curate untombstone",
                vec![
                    "curation_candidates",
                    "memories",
                    "search_index_jobs",
                    "audit_log",
                ],
                "Restore a tombstoned curation candidate through audited memory/index updates",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "handoff rotate-key",
                vec!["audit_log"],
                vec!["<handoff capsule path>"],
                "Rotate a handoff capsule HMAC key and rewrite the signed capsule body",
            ),
            CommandEffect {
                command_path: "health scorecard --record-snapshot",
                default_effect: EffectClass::DurableMemoryWrite,
                dry_run_effect: None,
                idempotency: IdempotencyClass::Idempotent,
                write_surfaces: WriteSurfaces {
                    db_tables: vec!["debt_snapshots"],
                    derived_paths: Vec::new(),
                    workspace_files: Vec::new(),
                },
                mutation_contract: CommandMutationContract {
                    side_effect_class: SideEffectClass::AuditedMutation,
                    transaction_scope: Some("single DB insert-or-ignore for memory debt snapshot"),
                    idempotency_key: Some("workspace id plus snapshot day plus generation"),
                    audit_surface: Some("debt_snapshots"),
                    db_generation_effect: "advances only when a new debt snapshot row commits; unchanged on duplicate",
                    index_generation_effect: "none",
                    dry_run_behavior: Some(
                        "omit --record-snapshot to render the same scorecard without writing debt_snapshots",
                    ),
                    recovery_behavior: "insert-or-ignore leaves at most one complete snapshot row per workspace/day/generation",
                    no_overwrite_behavior: None,
                    degraded_code: None,
                },
                runtime_contract: CommandRuntimeContract::transactional(),
                requires_read_snapshot: false,
                requires_audit: true,
                description: "Record a memory-debt trend snapshot before rendering the health scorecard",
            },
            CommandEffect::durable_write(
                "playbook extract",
                vec!["curation_candidates", "audit_log"],
                "Extract procedural-rule candidates from repeated semantic memories",
            ),
            CommandEffect::durable_write(
                "playbook import",
                vec![
                    "procedural_rules",
                    "rule_source_memories",
                    "rule_tags",
                    "search_index_jobs",
                    "audit_log",
                ],
                "Import portable playbook rules through audited procedural-rule writes",
            ),
            CommandEffect::durable_write(
                "journal append",
                vec!["journal_entries"],
                "Append a redaction-screened observation to the agent journal",
            ),
            CommandEffect::durable_write(
                "journal distill",
                vec![
                    "journal_entries",
                    "curation_candidates",
                    "evidence_spans",
                    "sessions",
                    "audit_log",
                ],
                "Distill journal entries into pending curation candidates; dry-run by default, --apply writes",
            ),
            // No learning_experiments or evaluation_reports table exists
            // (bd-hb5os). `learn close` lists the tables it was measured to
            // write; `learn experiment run` was measured to write none.
            CommandEffect::durable_write(
                "learn close",
                vec!["feedback_events", "learning_observations", "audit_log"],
                "Close a learning experiment",
            ),
            // bd-acxw6: the bd-hb5os comment above already recorded that this
            // command was measured to write no table, but the declaration was
            // left naming `audit_log`. Execution is refused without
            // `--dry-run` (src/core/learn.rs:3200, src/cli/mod.rs:26064) and a
            // dry run persists nothing, so the measured write set is empty.
            // The class stays `DurableMemoryWrite`: that is what the command
            // becomes once experiment execution is backed by persisted
            // ledgers, and the typed refusal is what stands in until then.
            CommandEffect::durable_write(
                "learn experiment run",
                Vec::new(),
                "Record a learning experiment run",
            ),
            CommandEffect::durable_write(
                "learn observe",
                vec!["learning_observations", "audit_log"],
                "Record a learning observation",
            ),
            CommandEffect::durable_write(
                "learn experiment propose",
                vec!["curation_candidates", "audit_log"],
                "Persist experiment proposals to curation queue",
            ),
            CommandEffect::durable_write(
                "outcome",
                vec!["feedback_events", "audit_log"],
                "Record observed outcome feedback",
            ),
            CommandEffect::durable_write(
                "procedure promote",
                vec!["procedures", "procedure_events", "audit_log"],
                "Promote a persisted procedure maturity level",
            ),
            CommandEffect::durable_write(
                "procedure propose",
                vec!["procedures", "procedure_events", "audit_log"],
                "Persist a procedure candidate from explicit evidence",
            ),
            CommandEffect::durable_write(
                "procedure retire",
                vec!["procedures", "procedure_events", "audit_log"],
                "Retire a persisted procedure with an audited reason",
            ),
            CommandEffect::durable_write(
                "outcome quarantine release",
                vec!["feedback_quarantine", "audit_log"],
                "Release feedback from quarantine",
            ),
            CommandEffect::durable_write(
                "rationale attach",
                vec!["rationale_traces", "audit_log"],
                "Attach a safe rationale trace with audit provenance",
            ),
            // Measured per triggering input (bd-hb5os): anchors from anchorable
            // content, sentinel specs from --sentinel/--revive-when, links from
            // workflow and co-tag auto-linking, seals from --seal.
            CommandEffect::durable_write(
                "remember",
                vec![
                    "memories",
                    "memory_tags",
                    "memory_anchors",
                    "memory_anchor_index",
                    "memory_links",
                    "memory_sentinel_specs",
                    "memory_seals",
                    "search_index_jobs",
                    "workspace_generations",
                    "audit_log",
                ],
                "Store a new memory with direct or audit-lane-backed audit_log provenance",
            ),
            CommandEffect::durable_write(
                "plan recipe save",
                vec!["plan_recipes", "audit_log"],
                "Save explicit instructions as an audited draft recipe without execution",
            ),
            CommandEffect::durable_write(
                "decide record",
                vec![
                    "memories",
                    "memory_tags",
                    "memory_links",
                    "search_index_jobs",
                    "audit_log",
                ],
                "Record a durable decision memory and optionally supersede the prior decision head",
            ),
            CommandEffect::durable_write(
                "memory revise",
                vec!["memories", "search_index_jobs", "audit_log"],
                "Inserts a new memory row with the same logical_id as the original, sets the prior row's valid_to, emits a memory.revise audit entry, atomically enqueues the new live row in search_index_jobs, then reconciles that durable job post-commit with truthful immediate-or-queued status (N15.2 / bd-17c65.14.15.3 / bd-index-auto-freshness-m5kwf)",
            ),
            CommandEffect::durable_write(
                "memory expire",
                vec!["memories", "search_index_jobs", "audit_log"],
                "Expire a memory through an audited tombstone without deleting data",
            ),
            CommandEffect::durable_write(
                "memory reveal",
                vec!["memories", "memory_seals", "search_index_jobs", "audit_log"],
                "Verify supplied bytes against a sealed memory's commitment; on match publish the content through the revise path, mark the seal revealed, and audit memory.reveal — a mismatch mutates nothing and audits memory.reveal_failed (bd-sealed-preregistration-memory-b67be)",
            ),
            // The promoted/demoted config is a workspace file, not a table
            // (bd-hb5os).
            CommandEffect::durable_write_with_workspace_files(
                "shadow promote",
                vec!["audit_log"],
                vec![".ee/config.toml"],
                "Apply the persisted promotable tuning report's [search] fusion-weight overlay to <workspace>/.ee/config.toml via toml_edit, recording the full prior config bytes in the promotion audit; refusals (missing/stale/abstained/non-promotable report) are typed exit-7 policy denials and dry-run writes nothing (ADR 0070 §5)",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "shadow demote",
                vec!["audit_log"],
                vec![".ee/config.toml"],
                "Restore the pre-promotion config.toml bytes from the promotion audit (byte-identical; an absent prior restores as empty, never a deletion) and record the demotion",
            ),
            CommandEffect::durable_write(
                "memory promote-global",
                vec!["memories", "audit_log"],
                "Promote a workspace memory into the user-global store through the bd-1bfwa.2 policy core (evidence gate, secret re-screen, duplicate merge); refusals are typed exit-7 policy denials and dry-run writes nothing",
            ),
            CommandEffect::durable_write(
                "memory demote-global",
                vec!["memories", "audit_log"],
                "Tombstone a promoted user-global memory (withdraw the global copy without touching the origin workspace row), origin parsed from promotion provenance",
            ),
            CommandEffect::durable_write(
                "memory outcome-global",
                vec!["memories", "audit_log"],
                "Record helpful/harmful feedback on a user-global memory and backflow a clamped confidence adjustment to the origin workspace row (bd-1bfwa.2 engine); dry-run writes nothing",
            ),
            CommandEffect::durable_write(
                "memory level",
                vec!["memories", "search_index_jobs", "audit_log"],
                "Apply a canonical manual memory-level transition with audit provenance",
            ),
            CommandEffect::durable_write(
                "memory link",
                vec!["memory_links", "audit_log"],
                "List or create explicit memory links with deterministic idempotent audits",
            ),
            CommandEffect::durable_write(
                "index backfill-tags",
                vec!["memory_tags", "search_index_jobs", "audit_log"],
                "Derive tags for historical tag-less memories through the audited memory-tags patch path",
            ),
            CommandEffect::durable_write(
                "memory tags",
                vec!["memory_tags", "search_index_jobs", "audit_log"],
                "List or mutate memory tags with deterministic idempotent audits",
            ),
            CommandEffect::durable_write(
                "link",
                vec!["memory_links", "audit_log"],
                "Create or inspect explicit memory links with audited mutation when requested",
            ),
            {
                let mut checkpoint = CommandEffect::durable_state_write(
                    "maintenance wal-checkpoint",
                    Vec::new(),
                    "database path plus checkpoint mode",
                    "database WAL checkpoint",
                    "Checkpoint the workspace database WAL without changing logical memory records",
                );
                // The checkpoint rewrites database files, never a logical row:
                // passive mode was measured to change the database file,
                // truncate mode the WAL file (bd-hb5os).
                checkpoint.write_surfaces.workspace_files = vec![".ee/ee.db", ".ee/ee.db-wal"];
                checkpoint
            },
            CommandEffect::durable_write_with_workspace_files(
                "mesh discovery-policy",
                vec!["audit_log"],
                vec![
                    ".ee/discovery_policy.toml",
                    ".ee/discovery_allowlist.toml",
                    ".ee/discovery_denylist.toml",
                ],
                "Persist mesh discovery policy files and audit the policy change",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "mesh export",
                vec!["audit_log"],
                vec!["<--out path>"],
                "Export authorized mesh material to an explicit side-path artifact",
            ),
            CommandEffect::durable_write(
                "mesh peer add",
                vec!["mesh_peers", "audit_log"],
                "Add a mesh peer with audited policy metadata",
            ),
            CommandEffect::durable_write(
                "mesh peer revoke",
                vec!["mesh_peers", "audit_log"],
                "Revoke a mesh peer with audited policy metadata",
            ),
            CommandEffect::durable_write(
                "mesh peer rotate",
                vec!["mesh_peers", "audit_log"],
                "Rotate mesh peer credentials with audited policy metadata",
            ),
            CommandEffect::durable_write(
                "mesh grant",
                vec!["mesh_lane_grant_states", "audit_log"],
                "Apply an authenticated mesh lane consent grant with generation fencing and audit provenance",
            ),
            CommandEffect::durable_write(
                "mesh revoke-lane",
                vec!["mesh_lane_grant_states", "audit_log"],
                "Narrow mesh lane consent with generation fencing and audit provenance",
            ),
            CommandEffect::durable_write(
                "note",
                vec!["memories", "memory_tags", "audit_log"],
                "Store a note as a memory with optional tags",
            ),
            CommandEffect::durable_write(
                "reflect ingest",
                vec![
                    "reflection_request_ledger",
                    "curation_candidates",
                    "audit_log",
                ],
                "Ingest reflection evidence and propose reviewed curation candidates",
            ),
            CommandEffect::durable_write(
                "review workspace",
                vec!["curation_candidates", "audit_log"],
                "Review workspace evidence and persist curation candidates",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "sandbox apply",
                vec!["memories", "memory_tags", "audit_log"],
                vec![".ee/sandbox/<session>.json"],
                "Apply reviewed sandbox memories and record sandbox session state",
            ),
            CommandEffect::durable_state_write(
                "sentinel check",
                vec!["memory_sentinel_results"],
                "sentinel spec hash plus observed result hash",
                "memory sentinel results",
                "Evaluate sentinel specs and persist checked results",
            ),
            CommandEffect::durable_write(
                "tag",
                vec!["memory_tags", "search_index_jobs", "audit_log"],
                "Add or remove memory tags through audited metadata updates",
            ),
            // The `team` mutation surface (bd-qked7). Every table below was
            // read out of the delegate's own SQL, not inferred from the command
            // name: each handler in `src/cli/team.rs` calls exactly one or two
            // `src/mesh/team.rs` entry points, and those reach the
            // `DbConnection` methods named in the comments. Commands that also
            // touch `<workspace>/.ee/keys/mesh/` do so through
            // `Ed25519OriginSigner::load_or_create`, which mints and persists a
            // signing seed on first use — that is why they carry a
            // workspace-file surface and their siblings do not. Commands that
            // additionally speak to the network or spawn `curl` are declared in
            // `external_io_write_commands`, not here.
            CommandEffect::durable_write_with_workspace_files(
                "team create",
                vec![
                    "mesh_origin_events",
                    "mesh_origin_event_nonces",
                    "team_members",
                    "team_member_nodes",
                    "team_member_signing_keys",
                    "team_invite_auth_floor",
                ],
                vec!["<workspace>/.ee/keys/mesh/"],
                "Append the teamCreated genesis origin event, persist the self member, and raise the invite auth floor",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "team idp require",
                vec![
                    "mesh_origin_events",
                    "mesh_origin_event_nonces",
                    "team_idp_policy",
                ],
                vec!["<workspace>/.ee/keys/mesh/"],
                "Pin the team IdP policy to tailnet-attested identities through a signed origin event",
            ),
            CommandEffect::durable_write(
                "team idp set",
                vec!["team_idp_oidc"],
                "Pin a secretless-public OIDC issuer from a local discovery document",
            ),
            CommandEffect::durable_write(
                "team leave",
                vec![
                    "team_members",
                    "team_member_nodes",
                    "team_invite_auth_floor",
                    "team_removal_acknowledgements",
                ],
                "Retire the local member and its nodes, raise the invite auth floor, and record the removal acknowledgement",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "team members add-node",
                vec![
                    "mesh_origin_events",
                    "mesh_origin_event_nonces",
                    "team_members",
                    "team_member_nodes",
                    "team_member_signing_keys",
                ],
                vec!["<workspace>/.ee/keys/mesh/"],
                "Bind an additional local node to the team through a signed origin event and a new signing key generation",
            ),
            CommandEffect::durable_write(
                "team members reconcile",
                vec![
                    "team_members",
                    "team_member_nodes",
                    "team_member_signing_keys",
                ],
                "Reconcile the member projection against the local origin stream",
            ),
            CommandEffect::durable_write(
                "team members remove",
                vec![
                    "team_members",
                    "team_member_nodes",
                    "team_invite_auth_floor",
                    "team_removal_acknowledgements",
                ],
                "Retire a member and its nodes, raise the invite auth floor, and record the removal acknowledgement",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "team members rotate-key",
                vec!["team_member_signing_keys"],
                vec!["<workspace>/.ee/keys/mesh/"],
                "Mint the next local signing key generation and persist the seed in the hardened key store",
            ),
            CommandEffect::durable_write(
                "team pause",
                vec!["team_posture"],
                "Pause local team posture so sync refuses to run",
            ),
            CommandEffect::durable_write(
                "team port migrate",
                vec!["mesh_origin_events", "mesh_peers"],
                "Append a versioned teamPortMigrated origin event and rewrite enrolled peer locators without touching pair keys or grants",
            ),
            CommandEffect::durable_write(
                "team projects adopt",
                vec!["team_projects"],
                "Bind a shared team project to a local workspace path",
            ),
            CommandEffect::durable_write(
                "team projects reconcile",
                vec!["team_projects"],
                "Reconcile the team project projection against the local origin stream",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "team projects share",
                vec![
                    "mesh_origin_events",
                    "mesh_origin_event_nonces",
                    "team_projects",
                ],
                vec!["<workspace>/.ee/keys/mesh/"],
                "Share a project with the team through a signed origin event and a project projection row",
            ),
            CommandEffect::durable_write(
                "team resume",
                vec!["team_posture"],
                "Clear the paused team posture so sync can run again",
            ),
            CommandEffect::durable_write(
                "team revoke",
                vec!["team_pending_invites", "team_invite_auth_floor"],
                "Revoke pending team invites and raise the invite auth floor",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "team share bodies",
                vec![
                    "mesh_origin_events",
                    "mesh_origin_event_nonces",
                    "mesh_body_cache_metadata",
                ],
                vec!["<workspace>/.ee/keys/mesh/"],
                "Represent memory bodies to the team through a signed origin event and body-cache metadata",
            ),
            CommandEffect::durable_write_with_workspace_files(
                "team share history",
                vec![
                    "mesh_origin_events",
                    "mesh_origin_event_nonces",
                    "team_history_projections",
                ],
                vec!["<workspace>/.ee/keys/mesh/"],
                "Share memory history with the team through a signed origin event and a history projection row",
            ),
            CommandEffect::durable_write(
                "team unshare bodies",
                vec!["mesh_body_cache_metadata"],
                "Withdraw represented memory bodies by clearing their body-cache metadata",
            ),
            CommandEffect::durable_write(
                "verification provenance",
                vec!["memories", "curation_candidates", "audit_log"],
                "Verify provenance and persist reviewed revalidation candidates",
            ),
            CommandEffect::durable_write(
                "verify provenance",
                vec!["memories", "curation_candidates", "audit_log"],
                "Verify provenance and persist reviewed revalidation candidates",
            ),
            CommandEffect::durable_state_write(
                "recorder start",
                vec!["recorder_runs"],
                "generated recorder run id",
                "recorder run store",
                "Persist a recorder run start row",
            ),
            CommandEffect::durable_state_write(
                "recorder event",
                vec!["recorder_events"],
                "run id plus next recorder sequence",
                "recorder event spine",
                "Append a redacted recorder event row",
            ),
            CommandEffect::durable_state_write(
                "recorder finish",
                vec!["recorder_runs"],
                "recorder run id",
                "recorder run store",
                "Mark a recorder run finished and persist rolled-up counts",
            ),
            CommandEffect::durable_write(
                "review session --propose",
                vec!["curation_candidates", "audit_log"],
                "Persist session-derived curation candidates",
            ),
            CommandEffect::durable_write(
                "workflow close",
                vec!["memories", "audit_log"],
                "Promote eligible workflow working memories to episodic records",
            ),
            CommandEffect::durable_write(
                "workflow create",
                vec!["memories", "audit_log"],
                "Create workflow working memory and audit provenance",
            ),
            CommandEffect::durable_write(
                "rule add",
                vec![
                    "procedural_rules",
                    "rule_source_memories",
                    "rule_tags",
                    "audit_log",
                    "search_index_jobs",
                ],
                "Store a procedural rule",
            ),
            CommandEffect::durable_write(
                "rule mark",
                vec!["procedural_rules", "audit_log", "search_index_jobs"],
                "Record lifecycle evidence for a procedural rule",
            ),
            CommandEffect::durable_write(
                "rule protect",
                vec!["procedural_rules", "audit_log"],
                "Protect or unprotect a procedural rule",
            ),
            CommandEffect::durable_write(
                "rule update",
                vec![
                    "procedural_rules",
                    "rule_source_memories",
                    "rule_tags",
                    "audit_log",
                    "search_index_jobs",
                ],
                "Update procedural rule metadata",
            ),
            CommandEffect::durable_write(
                "situation adopt",
                vec!["situation_records"],
                "Adopt task text as a persisted situation record via the idempotent fingerprint",
            ),
            CommandEffect::durable_state_write(
                "tripwire check",
                vec!["tripwires", "tripwire_check_events"],
                "tripwire id plus checked_at plus event payload hash",
                "tripwire check event store",
                "Evaluate a persisted tripwire and record the check event unless --dry-run is used",
            ),
            CommandEffect::durable_state_write(
                "maintenance graph-witnesses-prune",
                vec!["graph_algorithm_witnesses"],
                "workspace id plus retention policy plus witness row identity",
                "graph_algorithm_witnesses",
                "Classify graph algorithm witnesses and delete only rows older than policy TTL that are not tied to active snapshots",
            ),
            CommandEffect::schema_migration_run(),
            {
                let mut rebind = CommandEffect::durable_write(
                    "workspace rebind",
                    vec!["workspaces", "audit_log"],
                    "Preview a moved local store; apply only an authenticated exact binding plan",
                )
                .with_read_snapshot();
                rebind.mutation_contract.idempotency_key =
                    Some("authenticated plan commitment and exact previous workspace binding");
                rebind.mutation_contract.dry_run_behavior = Some(
                    "default preview reads existing store and keys; only --apply-plan can commit",
                );
                rebind.mutation_contract.recovery_behavior = "binding and audit commit together; audit and report retain previous addressing for an authenticated reverse rebind";
                rebind
            },
        ]
    }

    fn config_write_commands() -> Vec<CommandEffect> {
        vec![
            // bd-acxw6: both lists are what `ee init` actually writes, measured
            // rather than intended (bd-hb5os probe v2).
            //
            // Files: `ee.toml` was declared and never written — the string
            // occurred nowhere else in the tree, and the real config file is
            // `.ee/config.toml`, already covered by the `.ee/` prefix. The two
            // guidance files ARE written, at src/core/init.rs:874 and :904, and
            // are suppressed by `--skip-boilerplate`; a surface that a flag
            // exists to turn off is a surface the manifest has to name. The
            // store-auth root key falls under `.ee/` but is named in full
            // anyway: it is key material, `backup keys import` already names
            // the identical path, and "does init create key material?" should
            // not require a reader to reason about prefixes.
            //
            // Tables: the six tables init leaves non-empty. `audit_log` is NOT
            // among them — init creates the table and writes no row into it.
            CommandEffect::config_write(
                "init",
                vec![
                    "curation_ttl_policies",
                    "ee_schema_migrations",
                    "memory_timestamp_spelling_repair_v124",
                    "model_registry",
                    "workspace_generations",
                    "workspaces",
                ],
                vec![
                    ".ee/",
                    ".ee/keys/store_auth_root.json",
                    "AGENTS.md",
                    "CLAUDE.md",
                ],
                "workspace root",
                "Initialize workspace-local ee configuration and storage",
            ),
            {
                // bd-acxw6: measured to touch no table of the workspace store
                // and no workspace file. `.ee/workspaces.toml` was declared and
                // never written; the alias lands in the registry database only,
                // which is the derived path below.
                let mut alias = CommandEffect::config_write(
                    "workspace alias",
                    Vec::new(),
                    Vec::new(),
                    "alias name and workspace root",
                    "Create or update a workspace alias",
                );
                // The registry database outside the workspace: EE_WORKSPACE_REGISTRY,
                // else $XDG_DATA_HOME/ee/workspaces.db
                // (core::workspace::registry_database_path_override).
                alias.write_surfaces.derived_paths = vec!["<workspace-registry-db>"];
                alias
            },
            CommandEffect::config_file_write(
                "profile config apply",
                vec![".ee/config.toml"],
                "workspace profile config path plus requested profile",
                "Apply operating profile configuration to the workspace config file",
            ),
            CommandEffect::config_file_write(
                "config set",
                vec![".ee/config.toml"],
                "workspace config path plus exact config key and scalar value",
                "Set a supported workspace configuration key",
            ),
            CommandEffect::harness_hook_settings_write(
                "hook claude-code --install",
                vec![
                    "~/.claude/settings.json",
                    "~/.claude/settings.json.ee-backup",
                ],
                "Claude Code settings path plus generated managed hook snippets",
                "Install ee-managed Claude Code recall and journal hooks into harness settings",
            ),
            CommandEffect::harness_hook_settings_write(
                "hook claude-code --undo",
                vec![
                    "~/.claude/settings.json",
                    "~/.claude/settings.json.ee-backup",
                ],
                "Claude Code settings backup path",
                "Restore Claude Code harness settings from the deterministic ee backup",
            ),
            CommandEffect::harness_hook_settings_write(
                "hook codex --install",
                vec![".codex/hooks.json", ".codex/hooks.json.ee-backup"],
                "Codex hooks path plus generated managed hook snippets",
                "Install ee-managed Codex recall and journal hooks into harness settings",
            ),
            CommandEffect::harness_hook_settings_write(
                "hook codex --undo",
                vec![".codex/hooks.json", ".codex/hooks.json.ee-backup"],
                "Codex hooks backup path",
                "Restore Codex harness settings from the deterministic ee backup",
            ),
            CommandEffect::certificate_key_file_write(
                "certificate keygen",
                vec!["~/.config/ee/keys/<workspace>.ed25519"],
                "workspace key path plus --show/--force mode",
                "Generate or inspect a local certificate signing key",
            ),
            // bd-acxw6: both were measured to change `.ee/config.toml` and
            // nothing else — no table of the workspace store, no registry
            // entry, no `.ee/audit.jsonl` record. That is by construction:
            // src/mesh/emergency_disable.rs is "deliberately local-only" so
            // containment still works when the store is unreachable. The
            // declared tables are corrected here to the measured set; the fact
            // that an incident-containment command with `--reason` and
            // `--temporary-for` leaves no audit record anywhere is a product
            // gap, filed separately, not a declaration to quietly soften.
            CommandEffect::config_write(
                "mesh disable",
                Vec::new(),
                vec![".ee/config.toml"],
                "workspace id plus mesh disable reason",
                "Disable mesh synchronization for a workspace",
            ),
            CommandEffect::config_write(
                "mesh reenable",
                Vec::new(),
                vec![".ee/config.toml"],
                "workspace id plus mesh reenable reason",
                "Re-enable mesh synchronization for a workspace",
            ),
        ]
    }

    fn workspace_file_write_commands() -> Vec<CommandEffect> {
        vec![
            CommandEffect::shard_fanout_migration(),
            CommandEffect::workspace_file_write(
                "backup create",
                vec![".ee/backups/<backup-id>/"],
                "Create redacted backup artifacts in the workspace",
            ),
            // Writes the root key file INSIDE `.ee/`, so it is a workspace-file
            // write rather than external I/O. Same `--dry-run` note as `backup
            // keys export`: the command honours the flag and is declared on
            // `DRY_RUN_CAPABLE`, not by a constructor default.
            CommandEffect::workspace_file_write(
                "backup keys import",
                vec![".ee/keys/store_auth_root.json"],
                "Write the store-auth root key file from an encrypted recovery envelope",
            ),
            CommandEffect::workspace_file_write(
                "backup restore",
                vec!["<side-path>/"],
                "Restore backup contents into an explicit side path",
            ),
            CommandEffect::workspace_file_write(
                "export",
                vec![".ee/backups/<backup-id>/ or <--output-dir>/<backup-id>/"],
                "Export redacted JSONL records as side-path artifacts",
            ),
            CommandEffect::workspace_file_write(
                "export agentsmd",
                vec!["AGENTS.md (or --file target) managed block plus .ee-backup sibling"],
                "Render the primer rules+warnings sections into the AGENTS.md managed block",
            ),
            CommandEffect::workspace_file_write(
                "artifact relocate",
                vec![
                    "artifact relocation destination paths",
                    "artifact relocation manifest path",
                ],
                "Copy preserved artifacts to explicit relocation destinations without deleting originals",
            ),
            CommandEffect::workspace_state_write(
                "coordination evidence ingest",
                vec![".ee/coordination-fallback-evidence.jsonl"],
                "coordination evidence content hash",
                "Append coordination fallback evidence to the workspace-local evidence log",
            ),
            CommandEffect::workspace_state_write(
                "daemon start",
                vec!["$XDG_RUNTIME_DIR/ee/daemon.sock or ${TMPDIR:-/tmp}/ee-<uid>/daemon.sock"],
                "daemon socket path plus process id",
                "Bind the optional UDS RPC socket outside the workspace",
            ),
            CommandEffect::workspace_state_write(
                "daemon stop",
                vec!["$XDG_RUNTIME_DIR/ee/daemon.sock or ${TMPDIR:-/tmp}/ee-<uid>/daemon.sock"],
                "daemon socket path plus process id",
                "Stop dialing the optional UDS RPC socket and update daemon state",
            ),
            CommandEffect::workspace_file_write(
                "lab capture",
                vec![".ee/lab/episodes"],
                "Writes a frozen episode artifact under .ee/lab/episodes/<EPISODE_ID>/ — task input, policy ids, evidence ids, pack hash, repository fingerprint (N15.3 / bd-17c65.14.15.4)",
            ),
            CommandEffect::workspace_file_write(
                "focus add",
                vec![".ee/focus/state.json"],
                "Add explicit memories to passive focus state without eviction",
            ),
            CommandEffect::workspace_file_write(
                "handoff create",
                vec!["<--out path>"],
                "Write a redacted continuity capsule to a user-specified output path",
            ),
            CommandEffect::workspace_file_write(
                "playbook export",
                vec!["<--out path>"],
                "Write portable procedural rules to a no-overwrite playbook artifact",
            ),
            CommandEffect::workspace_state_write(
                "preflight close",
                vec![".ee/preflight_runs.json"],
                "preflight run id",
                "Close a persisted preflight run in the workspace-local run store",
            ),
            CommandEffect::workspace_state_write(
                "preflight run",
                vec![".ee/preflight_runs.json"],
                "generated preflight run id",
                "Persist an evidence-backed preflight run in the workspace-local run store",
            ),
            CommandEffect::workspace_file_write(
                "support bundle",
                vec!["<--out path>/"],
                "Create a redacted support bundle side-path artifact",
            ),
            CommandEffect::workspace_file_write(
                "team credentials backup",
                vec![".ee/keys/mesh-credential-backup/"],
                "Write an encrypted mesh credential-backup envelope through the hardened keys tree",
            ),
            CommandEffect::workspace_file_write(
                "team credentials restore",
                vec![".ee/keys/mesh/"],
                "Restore pair keys and signing seeds from an encrypted credential-backup envelope",
            ),
            CommandEffect::workspace_file_write(
                "rehearse run",
                vec![
                    "rehearsal artifact root",
                    "tempfile-backed sandbox workspace",
                ],
                "Rehearsal execution writes side-path sandbox artifacts without mutating the source workspace",
            ),
            CommandEffect::workspace_state_write(
                "recorder flight append",
                vec!["flight recorder trace directory"],
                "flight recorder event hash plus sequence",
                "Append an event to a flight-recorder trace",
            ),
            CommandEffect::workspace_state_write(
                "sandbox curate",
                vec![".ee/sandbox/<session>.json"],
                "sandbox session id plus curation event hash",
                "Update sandbox curation state without applying it to durable memories",
            ),
            CommandEffect::workspace_state_write(
                "sandbox import",
                vec![".ee/sandbox/<session>.json"],
                "sandbox session id plus import source hash",
                "Import memories into sandbox state without applying them to durable storage",
            ),
            CommandEffect::workspace_state_write(
                "sandbox remember",
                vec![".ee/sandbox/<session>.json"],
                "sandbox session id plus memory content hash",
                "Record a sandbox memory without applying it to durable storage",
            ),
            CommandEffect::workspace_file_write(
                "focus clear",
                vec![".ee/focus/state.json"],
                "Clear passive focus state by writing an empty state artifact",
            ),
            CommandEffect::workspace_file_write(
                "focus remove",
                vec![".ee/focus/state.json"],
                "Remove explicit memories from passive focus state",
            ),
            CommandEffect::workspace_file_write(
                "focus set",
                vec![".ee/focus/state.json"],
                "Replace passive focus state from explicit command arguments",
            ),
            CommandEffect::workspace_file_write(
                "task-frame create",
                vec![".ee/task_frames.json"],
                "Create a passive task frame without executing commands",
            ),
            CommandEffect::workspace_file_write(
                "task-frame update",
                vec![".ee/task_frames.json"],
                "Update passive task-frame state without executing commands",
            ),
            CommandEffect::workspace_file_write(
                "task-frame close",
                vec![".ee/task_frames.json"],
                "Close a passive task frame without executing commands",
            ),
            CommandEffect::workspace_file_write(
                "task-frame subgoal add",
                vec![".ee/task_frames.json"],
                "Add a passive task-frame subgoal without executing commands",
            ),
        ]
    }

    /// Get the effect entry for a command path.
    #[must_use]
    pub fn get(&self, command_path: &str) -> Option<&CommandEffect> {
        self.entries.get(command_path)
    }

    /// All command paths in the manifest.
    #[must_use]
    pub fn command_paths(&self) -> Vec<&'static str> {
        let mut paths: Vec<_> = self.entries.keys().copied().collect();
        paths.sort_unstable();
        paths
    }

    /// Commands that are safe to call mid-task (read-only).
    #[must_use]
    pub fn safe_mid_task_commands(&self) -> Vec<&CommandEffect> {
        self.entries
            .values()
            .filter(|e| e.is_safe_mid_task())
            .collect()
    }

    /// Commands that perform durable mutations.
    #[must_use]
    pub fn mutating_commands(&self) -> Vec<&CommandEffect> {
        self.entries
            .values()
            .filter(|e| e.default_effect.is_mutating())
            .collect()
    }

    /// Number of commands in the manifest.
    #[must_use]
    pub fn len(&self) -> usize {
        self.entries.len()
    }

    /// `true` if the manifest is empty.
    #[must_use]
    pub fn is_empty(&self) -> bool {
        self.entries.is_empty()
    }
}

impl Default for EffectManifest {
    fn default() -> Self {
        Self::build()
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    type TestResult = Result<(), String>;

    fn ensure<T: std::fmt::Debug + PartialEq>(actual: T, expected: T, ctx: &str) -> TestResult {
        if actual == expected {
            Ok(())
        } else {
            Err(format!("{ctx}: expected {expected:?}, got {actual:?}"))
        }
    }

    fn ensure_at_least<T: std::fmt::Debug + PartialOrd>(
        actual: T,
        minimum: T,
        ctx: &str,
    ) -> TestResult {
        if actual >= minimum {
            Ok(())
        } else {
            Err(format!(
                "{ctx}: expected at least {minimum:?}, got {actual:?}"
            ))
        }
    }

    fn ensure_category<F>(category: &str, entries: Vec<CommandEffect>, accepts: F) -> TestResult
    where
        F: Fn(&CommandEffect) -> bool,
    {
        for entry in entries {
            if !accepts(&entry) {
                return Err(format!(
                    "{} builder contains `{}` with default_effect={} side_effect_class={}",
                    category,
                    entry.command_path,
                    entry.default_effect.as_str(),
                    entry.mutation_contract.side_effect_class.as_str()
                ));
            }
        }
        Ok(())
    }

    #[test]
    fn effect_class_strings_are_stable() -> TestResult {
        ensure(EffectClass::ReadOnly.as_str(), "read_only", "read_only")?;
        ensure(
            EffectClass::DerivedArtifactWrite.as_str(),
            "derived_artifact_write",
            "derived_artifact_write",
        )?;
        ensure(
            EffectClass::DurableMemoryWrite.as_str(),
            "durable_memory_write",
            "durable_memory_write",
        )?;
        ensure(
            EffectClass::WorkspaceFileWrite.as_str(),
            "workspace_file_write",
            "workspace_file_write",
        )
    }

    #[test]
    fn side_effect_class_strings_match_matrix_vocabulary() -> TestResult {
        ensure(
            SideEffectClass::ReadOnly.as_str(),
            "class=read_only",
            "read_only",
        )?;
        ensure(
            SideEffectClass::AppendOnly.as_str(),
            "class=append_only",
            "append_only",
        )?;
        ensure(
            SideEffectClass::ReadOnlyNow.as_str(),
            "class=read_only_now",
            "read_only_now",
        )?;
        ensure(
            SideEffectClass::ReportOnly.as_str(),
            "class=report_only",
            "report_only",
        )?;
        ensure(
            SideEffectClass::ReadOnlyOrUnavailable.as_str(),
            "class=read_only_or_unavailable",
            "read_only_or_unavailable",
        )?;
        ensure(
            SideEffectClass::AuditedMutation.as_str(),
            "class=audited_mutation",
            "audited_mutation",
        )?;
        ensure(
            SideEffectClass::DerivedAssetRebuild.as_str(),
            "class=derived_asset_rebuild",
            "derived_asset_rebuild",
        )?;
        ensure(
            SideEffectClass::SidePathArtifact.as_str(),
            "class=side_path_artifact",
            "side_path_artifact",
        )?;
        ensure(
            SideEffectClass::SupervisedJobs.as_str(),
            "class=supervised_jobs",
            "supervised_jobs",
        )?;
        ensure(SideEffectClass::Mixed.as_str(), "class=mixed", "mixed")?;
        ensure(
            SideEffectClass::DegradedUnavailable.as_str(),
            "class=degraded_unavailable",
            "degraded_unavailable",
        )?;
        ensure(
            SideEffectClass::ReportOnlyOrAppend.as_str(),
            "class=report_only_or_append",
            "report_only_or_append",
        )?;
        ensure(
            SideEffectClass::ReportOnlyOrAuditedMutation.as_str(),
            "class=report_only_or_audited_mutation",
            "report_only_or_audited_mutation",
        )
    }

    #[test]
    fn effect_class_is_mutating_classifies_correctly() -> TestResult {
        ensure(EffectClass::ReadOnly.is_mutating(), false, "read_only")?;
        ensure(
            EffectClass::DerivedArtifactWrite.is_mutating(),
            true,
            "derived_artifact_write",
        )?;
        ensure(
            EffectClass::DurableMemoryWrite.is_mutating(),
            true,
            "durable_memory_write",
        )?;
        ensure(
            EffectClass::WorkspaceFileWrite.is_mutating(),
            true,
            "workspace_file_write",
        )
    }

    #[test]
    fn idempotency_strings_are_stable() -> TestResult {
        ensure(
            IdempotencyClass::Idempotent.as_str(),
            "idempotent",
            "idempotent",
        )?;
        ensure(
            IdempotencyClass::NonIdempotent.as_str(),
            "non_idempotent",
            "non_idempotent",
        )?;
        ensure(
            IdempotencyClass::DryRunAvailable.as_str(),
            "dry_run_available",
            "dry_run_available",
        )
    }

    #[test]
    fn runtime_class_strings_are_stable() -> TestResult {
        ensure(RuntimeClass::Immediate.as_str(), "immediate", "immediate")?;
        ensure(RuntimeClass::Bounded.as_str(), "bounded", "bounded")?;
        ensure(
            RuntimeClass::LongRunning.as_str(),
            "long_running",
            "long_running",
        )?;
        ensure(
            RuntimeClass::MultiStage.as_str(),
            "multi_stage",
            "multi_stage",
        )?;
        ensure(
            RuntimeClass::Supervised.as_str(),
            "supervised",
            "supervised",
        )
    }

    #[test]
    fn runtime_budget_math_is_deterministic() -> TestResult {
        let long = CommandRuntimeContract::long_running_derived();
        ensure(
            long.requires_budget(),
            true,
            "long-running work requires budget",
        )?;
        ensure(
            long.effective_budget_ms(None),
            Ok(Some(300_000)),
            "default long-running budget",
        )?;
        ensure(
            long.effective_budget_ms(Some(42)),
            Ok(Some(42)),
            "explicit budget overrides default",
        )?;
        ensure(
            long.effective_budget_ms(Some(0)),
            Err("runtime budget must be greater than zero"),
            "zero budget rejected",
        )?;

        let immediate = CommandRuntimeContract::immediate();
        ensure(
            immediate.requires_budget(),
            false,
            "immediate work does not require budget",
        )?;
        ensure(
            immediate.effective_budget_ms(None),
            Ok(None),
            "immediate work is unbounded by default",
        )
    }

    #[test]
    fn command_effect_read_only_is_safe_mid_task() -> TestResult {
        let effect = CommandEffect::read_only("status", "Report status");
        ensure(
            effect.is_safe_mid_task(),
            true,
            "read_only is safe mid-task",
        )?;
        ensure(
            effect.default_effect,
            EffectClass::ReadOnly,
            "default effect",
        )?;
        ensure(
            effect.runtime_contract.runtime_class,
            RuntimeClass::Bounded,
            "read-only work is bounded",
        )?;
        ensure(effect.requires_audit, false, "no audit required")
    }

    #[test]
    fn command_effect_durable_write_requires_audit() -> TestResult {
        let effect = CommandEffect::durable_write("remember", vec!["memories"], "Store memory");
        ensure(
            effect.is_safe_mid_task(),
            false,
            "durable write not safe mid-task",
        )?;
        ensure(
            effect.default_effect,
            EffectClass::DurableMemoryWrite,
            "default effect",
        )?;
        ensure(effect.requires_audit, true, "audit required")?;
        // bd-3j6l3. This builds a CommandEffect DIRECTLY, so it sees what the
        // constructor decides and never reaches DRY_RUN_CAPABLE. It used to
        // assert Some(ReadOnly) here, which pinned the very default that made
        // 90 commands claim a flag they do not have. The constructor no longer
        // decides, so the truthful constructor-level value is None.
        ensure(
            effect.dry_run_effect,
            None,
            "the constructor no longer claims --dry-run support",
        )?;
        // The capability is a property of the MANIFEST, not the constructor,
        // and asserting only the line above would leave this test unable to
        // tell "opt-in works" from "opt-in was deleted". `remember` is on the
        // roster and really does take --dry-run.
        ensure(
            EffectManifest::build()
                .get("remember")
                .and_then(|entry| entry.dry_run_effect),
            Some(EffectClass::ReadOnly),
            "the manifest grants remember its --dry-run via DRY_RUN_CAPABLE",
        )?;
        ensure(
            effect.mutation_contract.side_effect_class,
            SideEffectClass::AuditedMutation,
            "durable write has audited contract",
        )?;
        ensure(
            effect.mutation_contract.audit_surface,
            Some("audit_log"),
            "durable write names audit surface",
        )?;
        ensure(
            effect.runtime_contract.runtime_class,
            RuntimeClass::MultiStage,
            "durable write has multi-stage runtime",
        )
    }

    #[test]
    fn command_effect_append_only_write_uses_idempotency_key() -> TestResult {
        let effect = CommandEffect::append_only_write(
            "artifact register",
            vec!["artifacts", "audit_log"],
            "content hash",
            "Register artifact",
        );
        ensure(
            effect.default_effect,
            EffectClass::DurableMemoryWrite,
            "append-only writes durable records",
        )?;
        ensure(
            effect.idempotency,
            IdempotencyClass::Idempotent,
            "append-only retries are idempotent",
        )?;
        ensure(
            effect.mutation_contract.side_effect_class,
            SideEffectClass::AppendOnly,
            "append-only class",
        )?;
        ensure(
            effect.mutation_contract.idempotency_key,
            Some("content hash"),
            "idempotency key",
        )?;
        ensure(
            effect.runtime_contract.runtime_class,
            RuntimeClass::MultiStage,
            "append-only write has multi-stage runtime",
        )?;
        ensure(effect.requires_audit, true, "append-only requires audit")
    }

    #[test]
    fn command_effect_degraded_unavailable_is_read_only_with_code() -> TestResult {
        let effect = CommandEffect::degraded_unavailable(
            "lab replay",
            "lab_replay_unavailable",
            "Lab replay abstains until replay evidence exists",
        );
        ensure(
            effect.default_effect,
            EffectClass::ReadOnly,
            "degraded unavailable is read-only",
        )?;
        ensure(
            effect.mutation_contract.side_effect_class,
            SideEffectClass::DegradedUnavailable,
            "degraded side-effect class",
        )?;
        ensure(
            effect.write_surfaces.is_empty(),
            true,
            "degraded command has no write surfaces",
        )?;
        ensure(
            effect.mutation_contract.degraded_code,
            Some("lab_replay_unavailable"),
            "degraded code is explicit",
        )?;
        ensure(
            effect.runtime_contract.runtime_class,
            RuntimeClass::Immediate,
            "degraded unavailable returns immediately",
        )?;
        ensure(
            effect.requires_audit,
            false,
            "degraded command does not audit",
        )
    }

    #[test]
    fn command_effect_workspace_file_write_uses_side_path_no_delete_contract() -> TestResult {
        let effect = CommandEffect::workspace_file_write(
            "backup restore",
            vec!["<side-path>/"],
            "Restore backup into an explicit side path",
        );

        ensure(
            effect.default_effect,
            EffectClass::WorkspaceFileWrite,
            "workspace write effect",
        )?;
        // bd-3j6l3, same reason as command_effect_durable_write_requires_audit:
        // built directly, so this is the constructor's value and the
        // constructor no longer claims a flag on the command's behalf.
        ensure(
            effect.dry_run_effect,
            None,
            "the constructor no longer claims --dry-run support",
        )?;
        ensure(
            EffectManifest::build()
                .get("backup restore")
                .and_then(|entry| entry.dry_run_effect),
            Some(EffectClass::ReadOnly),
            "the manifest grants backup restore its --dry-run via DRY_RUN_CAPABLE",
        )?;
        ensure(
            effect.mutation_contract.side_effect_class,
            SideEffectClass::SidePathArtifact,
            "workspace writes are side-path artifacts",
        )?;
        ensure(
            effect.requires_audit,
            true,
            "side-path artifacts require manifest audit",
        )?;
        ensure(
            effect.runtime_contract.runtime_class,
            RuntimeClass::MultiStage,
            "side-path artifacts are multi-stage",
        )?;
        ensure(
            effect
                .mutation_contract
                .no_overwrite_behavior
                .is_some_and(|policy| {
                    policy.contains("no-overwrite") && policy.contains("no-delete")
                }),
            true,
            "side-path policy names no-overwrite and no-delete",
        )?;
        ensure(
            effect
                .mutation_contract
                .recovery_behavior
                .contains("never deleted by ee"),
            true,
            "side-path recovery never deletes partial output",
        )
    }

    #[test]
    fn command_effect_derived_write_is_idempotent() -> TestResult {
        let effect =
            CommandEffect::derived_write("index rebuild", vec![".ee/index/"], "Rebuild indexes");
        ensure(
            effect.idempotency,
            IdempotencyClass::Idempotent,
            "derived write is idempotent",
        )?;
        ensure(
            effect.default_effect,
            EffectClass::DerivedArtifactWrite,
            "default effect",
        )?;
        ensure(
            effect.mutation_contract.side_effect_class,
            SideEffectClass::DerivedAssetRebuild,
            "derived write uses derived rebuild contract",
        )?;
        ensure(
            effect.mutation_contract.declares_no_source_mutation(),
            true,
            "derived write leaves source DB unchanged",
        )?;
        ensure(
            effect.runtime_contract.runtime_class,
            RuntimeClass::LongRunning,
            "derived writes are long-running",
        )
    }

    #[test]
    fn manifest_build_includes_all_command_classes() -> TestResult {
        let manifest = EffectManifest::build();

        ensure_at_least(manifest.len(), 20, "at least 20 commands")?;

        let safe = manifest.safe_mid_task_commands();
        ensure_at_least(safe.len(), 15, "at least 15 safe commands")?;

        let mutating = manifest.mutating_commands();
        ensure_at_least(mutating.len(), 2, "at least 2 mutating commands")
    }

    #[test]
    fn manifest_build_has_no_duplicate_command_paths() -> TestResult {
        // Pin the no-duplicate invariant directly via the category-
        // vector unions rather than only via `build()` (which already
        // panics on duplicate via `insert_unique`). The vector unions
        // exhibit the same drift surface but with a clearer test-
        // failure message identifying the offending category pair —
        // and the test also fails cleanly in release builds (where
        // `debug_assert!` would be a no-op).
        use std::collections::HashMap;
        let mut origins: HashMap<&'static str, &'static str> = HashMap::new();
        let category_vectors: &[(&'static str, Vec<CommandEffect>)] = &[
            ("read_only", EffectManifest::read_only_commands()),
            (
                "degraded_unavailable",
                EffectManifest::degraded_unavailable_commands(),
            ),
            ("derived_write", EffectManifest::derived_write_commands()),
            (
                "external_io_write",
                EffectManifest::external_io_write_commands(),
            ),
            ("supervised_job", EffectManifest::supervised_job_commands()),
            (
                "append_only_write",
                EffectManifest::append_only_write_commands(),
            ),
            ("durable_write", EffectManifest::durable_write_commands()),
            ("config_write", EffectManifest::config_write_commands()),
            (
                "workspace_file_write",
                EffectManifest::workspace_file_write_commands(),
            ),
        ];
        for (category, entries) in category_vectors {
            // Iterating `&[(K, Vec<V>)]` makes `category: &&'static str`
            // and `entries: &Vec<CommandEffect>`; dereference `category`
            // so the HashMap value type matches and the error message
            // shows the bare category name, not a `&str` debug form.
            let category = *category;
            for entry in entries {
                if let Some(previous_category) = origins.insert(entry.command_path, category) {
                    return Err(format!(
                        "command path `{}` is declared in both `{previous_category}` and `{category}` categories; a command must appear in exactly one",
                        entry.command_path
                    ));
                }
            }
        }
        Ok(())
    }

    #[test]
    fn manifest_category_builders_match_declared_effect_classes() -> TestResult {
        ensure_category("read_only", EffectManifest::read_only_commands(), |entry| {
            entry.default_effect == EffectClass::ReadOnly
                && entry.mutation_contract.side_effect_class != SideEffectClass::DegradedUnavailable
        })?;
        ensure_category(
            "degraded_unavailable",
            EffectManifest::degraded_unavailable_commands(),
            |entry| {
                entry.default_effect == EffectClass::ReadOnly
                    && entry.mutation_contract.side_effect_class
                        == SideEffectClass::DegradedUnavailable
                    && entry.write_surfaces.is_empty()
                    && !entry.requires_audit
                    && entry.mutation_contract.degraded_code.is_some()
            },
        )?;
        ensure_category(
            "derived_write",
            EffectManifest::derived_write_commands(),
            |entry| entry.default_effect == EffectClass::DerivedArtifactWrite,
        )?;
        ensure_category(
            "external_io_write",
            EffectManifest::external_io_write_commands(),
            |entry| entry.default_effect == EffectClass::ExternalIo,
        )?;
        ensure_category(
            "supervised_job",
            EffectManifest::supervised_job_commands(),
            |entry| entry.mutation_contract.side_effect_class == SideEffectClass::SupervisedJobs,
        )?;
        ensure_category(
            "append_only_write",
            EffectManifest::append_only_write_commands(),
            |entry| entry.mutation_contract.side_effect_class == SideEffectClass::AppendOnly,
        )?;
        ensure_category(
            "durable_write",
            EffectManifest::durable_write_commands(),
            |entry| {
                entry.default_effect == EffectClass::DurableMemoryWrite
                    && entry.mutation_contract.side_effect_class == SideEffectClass::AuditedMutation
            },
        )?;
        ensure_category(
            "config_write",
            EffectManifest::config_write_commands(),
            |entry| entry.default_effect == EffectClass::ConfigWrite,
        )?;
        ensure_category(
            "workspace_file_write",
            EffectManifest::workspace_file_write_commands(),
            |entry| entry.default_effect == EffectClass::WorkspaceFileWrite,
        )
    }

    #[test]
    fn manifest_get_returns_correct_entry() -> TestResult {
        let manifest = EffectManifest::build();

        let recipe = manifest
            .get("plan recipe save")
            .ok_or("recipe save effect missing")?;
        ensure(
            recipe.default_effect,
            EffectClass::DurableMemoryWrite,
            "recipe save writes durable state",
        )?;
        ensure(
            recipe.dry_run_effect,
            Some(EffectClass::ReadOnly),
            "recipe preview is read-only",
        )?;
        ensure(recipe.requires_audit, true, "recipe save requires an audit")?;
        ensure(
            recipe.write_surfaces.db_tables.clone(),
            vec!["plan_recipes", "audit_log"],
            "recipe mutation surfaces",
        )?;

        let status = manifest.get("status");
        ensure(status.is_some(), true, "status exists")?;
        ensure(
            status.map(|e| e.default_effect),
            Some(EffectClass::ReadOnly),
            "status is read_only",
        )?;

        let remember = manifest.get("remember");
        ensure(remember.is_some(), true, "remember exists")?;
        ensure(
            remember.map(|e| e.default_effect),
            Some(EffectClass::DurableMemoryWrite),
            "remember is durable_memory_write",
        )?;

        let outcome = manifest.get("outcome");
        ensure(outcome.is_some(), true, "outcome exists")?;
        ensure(
            outcome.map(|e| e.write_surfaces.db_tables.clone()),
            Some(vec!["feedback_events", "audit_log"]),
            "outcome writes feedback and audit",
        )?;

        let preflight_check = manifest
            .get("preflight check")
            .ok_or_else(|| "preflight check not found".to_owned())?;
        let preflight_guard = manifest
            .get("preflight guard")
            .ok_or_else(|| "preflight guard alias not found".to_owned())?;
        ensure(
            preflight_guard.default_effect,
            preflight_check.default_effect,
            "preflight guard alias matches preflight check effect",
        )?;
        ensure(
            preflight_guard.mutation_contract.side_effect_class,
            preflight_check.mutation_contract.side_effect_class,
            "preflight guard alias matches preflight check mutation class",
        )?;
        ensure(
            preflight_guard.requires_audit,
            false,
            "preflight guard alias is read-only and needs no audit",
        )?;
        ensure(
            preflight_guard.write_surfaces.is_empty(),
            true,
            "preflight guard alias declares no write surfaces",
        )?;

        let backup = manifest.get("backup create");
        ensure(backup.is_some(), true, "backup create exists")?;
        ensure(
            backup.map(|e| e.default_effect),
            Some(EffectClass::WorkspaceFileWrite),
            "backup create writes workspace files",
        )?;

        let decay_sweep = manifest.get("daemon foreground decay_sweep");
        ensure(
            decay_sweep.is_some(),
            true,
            "daemon foreground decay_sweep exists",
        )?;
        ensure(
            decay_sweep.map(|e| e.mutation_contract.side_effect_class),
            Some(SideEffectClass::SupervisedJobs),
            "daemon decay sweep uses supervised job contract",
        )?;
        ensure(
            decay_sweep.map(|e| e.runtime_contract.runtime_class),
            Some(RuntimeClass::Supervised),
            "daemon decay sweep runtime",
        )
    }

    #[test]
    fn manifest_classifies_migrate_command_paths() -> TestResult {
        let manifest = EffectManifest::build();

        let status = manifest
            .get("migrate status")
            .ok_or_else(|| "migrate status not found".to_owned())?;
        ensure(
            status.default_effect,
            EffectClass::ReadOnly,
            "migrate status is read-only",
        )?;
        ensure(
            status.read_snapshot(),
            true,
            "migrate status reads through a DB snapshot",
        )?;

        let run = manifest
            .get("migrate run")
            .ok_or_else(|| "migrate run not found".to_owned())?;
        ensure(
            run.default_effect,
            EffectClass::DurableMemoryWrite,
            "migrate run writes durable schema state",
        )?;
        ensure(
            run.dry_run_effect,
            Some(EffectClass::ReadOnly),
            "migrate run dry-run is read-only",
        )?;
        ensure(
            run.mutation_contract.side_effect_class,
            SideEffectClass::AuditedMutation,
            "migrate run has audited mutation contract",
        )?;
        ensure(
            run.write_surfaces
                .db_tables
                .contains(&"ee_schema_migrations"),
            true,
            "migrate run names schema migration table",
        )?;
        ensure(
            run.write_surfaces.derived_paths.contains(&".ee/index/"),
            true,
            "migrate run names post-migration index rebuild",
        )?;

        let shard = manifest
            .get("migrate shard-fanout")
            .ok_or_else(|| "migrate shard-fanout not found".to_owned())?;
        ensure(
            shard.default_effect,
            EffectClass::WorkspaceFileWrite,
            "migrate shard-fanout writes shard files",
        )?;
        ensure(
            shard.dry_run_effect,
            Some(EffectClass::ReadOnly),
            "migrate shard-fanout dry-run is read-only",
        )?;
        ensure(
            shard.mutation_contract.side_effect_class,
            SideEffectClass::AuditedMutation,
            "migrate shard-fanout has audited mutation contract",
        )?;
        ensure(
            shard
                .write_surfaces
                .workspace_files
                .contains(&"<shards-dir>/catalog.db"),
            true,
            "migrate shard-fanout names shard catalog file",
        )
    }

    #[test]
    fn manifest_distinguishes_append_only_imports_from_audited_mutations() -> TestResult {
        let manifest = EffectManifest::build();

        for command in [
            "artifact register",
            "import cass",
            "import jsonl",
            "import eidetic-legacy",
        ] {
            let effect = manifest
                .get(command)
                .ok_or_else(|| format!("{command} not found"))?;
            ensure(
                effect.mutation_contract.side_effect_class,
                SideEffectClass::AppendOnly,
                &format!("{command} is append-only"),
            )?;
            ensure(
                effect.idempotency,
                IdempotencyClass::Idempotent,
                &format!("{command} retries by idempotency key"),
            )?;
        }

        let remember = manifest
            .get("remember")
            .ok_or_else(|| "remember not found".to_owned())?;
        ensure(
            remember.mutation_contract.side_effect_class,
            SideEffectClass::AuditedMutation,
            "remember remains an audited mutation",
        )
    }

    #[test]
    fn manifest_tracks_lab_replay_as_available_read_only_path() -> TestResult {
        let manifest = EffectManifest::build();

        let command = "lab replay";
        let effect = manifest
            .get(command)
            .ok_or_else(|| format!("{command} not found"))?;
        ensure(
            effect.default_effect,
            EffectClass::ReadOnly,
            &format!("{command} is read-only"),
        )?;
        ensure(
            effect.mutation_contract.side_effect_class,
            SideEffectClass::ReadOnly,
            &format!("{command} uses read-only class"),
        )?;
        ensure(
            effect.write_surfaces.is_empty(),
            true,
            &format!("{command} has no write surfaces"),
        )?;
        ensure(
            effect.mutation_contract.degraded_code,
            None,
            &format!("{command} has no unavailable degraded code"),
        )?;

        Ok(())
    }

    #[test]
    fn manifest_classifies_mesh_lane_mutations_as_audited_durable_writes() -> TestResult {
        let manifest = EffectManifest::build();
        for command in ["mesh grant", "mesh revoke-lane"] {
            let effect = manifest
                .get(command)
                .ok_or_else(|| format!("{command} not found"))?;
            ensure(
                effect.default_effect,
                EffectClass::DurableMemoryWrite,
                &format!("{command} is a durable write"),
            )?;
            ensure(
                effect.requires_audit,
                true,
                &format!("{command} requires audit"),
            )?;
            ensure(
                effect.write_surfaces.db_tables.clone(),
                vec!["mesh_lane_grant_states", "audit_log"],
                &format!("{command} declares exact write surfaces"),
            )?;
        }
        Ok(())
    }

    #[test]
    fn manifest_tracks_demo_run_as_audited_external_io() -> TestResult {
        let manifest = EffectManifest::build();
        let effect = manifest
            .get("demo run")
            .ok_or_else(|| "demo run not found".to_owned())?;

        ensure(
            effect.default_effect,
            EffectClass::ExternalIo,
            "demo run executes manifest commands",
        )?;
        ensure(
            effect.dry_run_effect,
            Some(EffectClass::ReadOnly),
            "demo run --dry-run is read-only",
        )?;
        ensure(
            effect.mutation_contract.side_effect_class,
            SideEffectClass::AuditedMutation,
            "demo run writes audit rows",
        )?;
        ensure(
            effect.write_surfaces.db_tables.contains(&"audit_log"),
            true,
            "demo run writes audit_log",
        )?;
        ensure(
            effect.write_surfaces.workspace_files.is_empty(),
            false,
            "demo run names evidence/artifact write surfaces",
        )?;
        ensure(
            effect.mutation_contract.degraded_code,
            None,
            "demo run has no unavailable sentinel",
        )
    }

    #[test]
    fn manifest_command_paths_are_sorted() -> TestResult {
        let manifest = EffectManifest::build();
        let paths = manifest.command_paths();

        let mut sorted = paths.clone();
        sorted.sort_unstable();
        ensure(paths, sorted, "paths are sorted")
    }

    #[test]
    fn write_surfaces_none_is_empty() -> TestResult {
        let surfaces = WriteSurfaces::none();
        ensure(surfaces.is_empty(), true, "none is empty")
    }

    #[test]
    fn remember_command_writes_to_memory_tables() -> TestResult {
        let manifest = EffectManifest::build();
        let remember = manifest
            .get("remember")
            .ok_or_else(|| "remember not found".to_string())?;

        let has_memories = remember.write_surfaces.db_tables.contains(&"memories");
        ensure(has_memories, true, "writes to memories table")?;

        let has_audit = remember.write_surfaces.db_tables.contains(&"audit_log");
        ensure(has_audit, true, "writes to audit_log")
    }

    #[test]
    fn index_rebuild_writes_to_index_path() -> TestResult {
        let manifest = EffectManifest::build();
        let rebuild = manifest
            .get("index rebuild")
            .ok_or_else(|| "index rebuild not found".to_string())?;

        let has_index = rebuild.write_surfaces.derived_paths.contains(&".ee/index/");
        ensure(has_index, true, "writes to .ee/index/")
    }

    // ========================================================================
    // No-Mutation Contract Tests
    // ========================================================================

    #[test]
    fn all_read_only_commands_have_empty_write_surfaces() -> TestResult {
        let manifest = EffectManifest::build();

        for effect in manifest.safe_mid_task_commands() {
            if !effect.write_surfaces.is_empty() {
                return Err(format!(
                    "Read-only command '{}' has non-empty write surfaces",
                    effect.command_path
                ));
            }
        }
        Ok(())
    }

    #[test]
    fn db_backed_read_only_commands_declare_read_snapshot_requirement() -> TestResult {
        let manifest = EffectManifest::build();
        // `search` and `similar` are deliberately ABSENT from this list: both
        // append a retrieval audit row, so they are append-only writers, not
        // db-backed reads (bd-czj3e). They are re-pinned POSITIVELY by
        // retrieval_surfaces_declare_the_audit_row_they_append below -- removing
        // a name from a read-only list without asserting what it became is the
        // silencing this repo forbids.
        let db_backed = [
            "context",
            "orient",
            "why",
            "status",
            "doctor",
            "memory drift",
            "memory list",
            "memory show",
            "pack replay",
            "graph pagerank",
            "db status",
            "audit timeline",
            "curate candidates",
            "diag provenance",
            "trust report",
            "swarm brief",
        ];

        for command in db_backed {
            let effect = manifest
                .get(command)
                .ok_or_else(|| format!("{command} not found"))?;
            ensure(
                effect.default_effect,
                EffectClass::ReadOnly,
                &format!("{command} remains read-only"),
            )?;
            ensure(
                effect.read_snapshot(),
                true,
                &format!("{command} declares read snapshot requirement"),
            )?;
        }

        for command in ["help", "version", "completion"] {
            let effect = manifest
                .get(command)
                .ok_or_else(|| format!("{command} not found"))?;
            ensure(
                effect.read_snapshot(),
                false,
                &format!("{command} does not require a DB read snapshot"),
            )?;
        }

        Ok(())
    }

    /// The positive half of moving the three retrieval surfaces out of the
    /// read-only lists (bd-czj3e).
    ///
    /// All three open a WRITABLE handle after the read pin is released and
    /// append a retrieval audit row on every user-facing invocation
    /// (`src/core/search.rs:7652-7669` and `:8169`; the `--all-workspaces`
    /// fan-out reaches the same writer once per workspace through
    /// `run_search`). A declaration of `read_only_db` told agents and the
    /// boundary inventory that they take no write gate and leave the database
    /// byte-identical; both statements were false. Asserting what they ARE,
    /// rather than only deleting them from a list, is what keeps the next
    /// read-only sweep from quietly putting them back.
    #[test]
    fn retrieval_surfaces_declare_the_audit_row_they_append() -> TestResult {
        let manifest = EffectManifest::build();

        for command in ["search", "search --all-workspaces", "similar"] {
            let effect = manifest
                .get(command)
                .ok_or_else(|| format!("{command} not in manifest"))?;
            ensure(
                effect.default_effect,
                EffectClass::DurableMemoryWrite,
                &format!("{command} appends a durable audit row"),
            )?;
            ensure(
                effect.mutation_contract.side_effect_class,
                SideEffectClass::AppendOnly,
                &format!("{command} appends, never rewrites"),
            )?;
            ensure(
                effect.write_surfaces.db_tables.as_slice(),
                ["audit_log"].as_slice(),
                &format!("{command} writes exactly the audit log"),
            )?;
            ensure(
                effect.write_surfaces.derived_paths.is_empty()
                    && effect.write_surfaces.workspace_files.is_empty(),
                true,
                &format!("{command} touches no derived path or workspace file"),
            )?;
            ensure(
                effect.requires_audit,
                true,
                &format!("{command} requires audit"),
            )?;
            // A retrieval that cannot be re-run is not a retrieval surface: the
            // audit row is keyed, so a repeated query does not multiply rows.
            ensure(
                effect.idempotency,
                IdempotencyClass::Idempotent,
                &format!("{command} stays idempotent under retry"),
            )?;
            // The read-only lists these three came from all carry
            // `requires_read_snapshot`. Append-only writers must not, or
            // `db_backed_read_only_commands_declare_read_snapshot_requirement`
            // and this test could both pass with the entry in both places.
            ensure(
                effect.read_snapshot(),
                false,
                &format!("{command} is no longer a read-snapshot surface"),
            )?;
        }

        // The neighbouring recalibration path is a DIFFERENT write, not the
        // same one: it rewrites a derived artifact and touches no DB table.
        // Pinning the contrast is what makes the two classifications mean
        // something rather than both collapsing to "mutates".
        let recalibrate = manifest
            .get("search --recalibrate-now")
            .ok_or_else(|| "search --recalibrate-now not in manifest".to_owned())?;
        ensure(
            recalibrate.write_surfaces.db_tables.is_empty(),
            true,
            "search --recalibrate-now writes no DB table",
        )?;
        ensure(
            recalibrate.write_surfaces.derived_paths.as_slice(),
            [".ee/search/calibration.jsonl"].as_slice(),
            "search --recalibrate-now rewrites the derived calibration artifact",
        )?;

        Ok(())
    }

    #[test]
    fn all_read_only_commands_do_not_require_audit() -> TestResult {
        let manifest = EffectManifest::build();

        for effect in manifest.safe_mid_task_commands() {
            if effect.requires_audit {
                return Err(format!(
                    "Read-only command '{}' requires audit, but should not",
                    effect.command_path
                ));
            }
        }
        Ok(())
    }

    #[test]
    fn all_durable_write_commands_require_audit() -> TestResult {
        let manifest = EffectManifest::build();

        for effect in manifest.entries.values() {
            if effect.default_effect == EffectClass::DurableMemoryWrite && !effect.requires_audit {
                return Err(format!(
                    "Durable-write command '{}' does not require audit, but should",
                    effect.command_path
                ));
            }
        }
        Ok(())
    }

    #[test]
    fn all_mutating_commands_have_s43e_contract_metadata() -> TestResult {
        let manifest = EffectManifest::build();

        for effect in manifest.mutating_commands() {
            let contract = &effect.mutation_contract;
            if contract.transaction_scope.is_none() {
                return Err(format!(
                    "Mutating command '{}' has no transaction scope",
                    effect.command_path
                ));
            }
            if contract.idempotency_key.is_none() {
                return Err(format!(
                    "Mutating command '{}' has no idempotency key",
                    effect.command_path
                ));
            }
            // NARROWED (bd-3j6l3): this clause demanded dry_run_behavior from
            // every mutating command. LOST COVERAGE, stated rather than
            // implied: nothing now requires preview prose from a command that
            // has no preview. That is intended -- prose describing a preview
            // that does not exist is the same false claim in a second field,
            // and this clause could only be satisfied by writing one. The
            // surviving implication runs the other way and is asserted by
            // `every_mutating_command_declares_a_dry_run_behavior_when_it_claims_the_flag`:
            // a command claiming `dry_run_effect: Some(_)` must describe it.
            // The other four clauses of this gate are untouched.
            if contract.recovery_behavior.is_empty()
                || contract.db_generation_effect.is_empty()
                || contract.index_generation_effect.is_empty()
            {
                return Err(format!(
                    "Mutating command '{}' has incomplete recovery/generation effects",
                    effect.command_path
                ));
            }
        }
        Ok(())
    }

    #[test]
    fn side_path_artifact_commands_name_no_overwrite_behavior() -> TestResult {
        let manifest = EffectManifest::build();

        for effect in manifest.mutating_commands() {
            let contract = &effect.mutation_contract;
            if contract.side_effect_class.requires_no_overwrite_contract() {
                let Some(policy) = contract.no_overwrite_behavior else {
                    return Err(format!(
                        "Side-path command '{}' has no no-overwrite behavior",
                        effect.command_path
                    ));
                };
                if !policy.contains("no-overwrite") || !policy.contains("no-delete") {
                    return Err(format!(
                        "Side-path command '{}' must name no-overwrite and no-delete behavior",
                        effect.command_path
                    ));
                }
                if !contract.recovery_behavior.contains("never deleted by ee") {
                    return Err(format!(
                        "Side-path command '{}' must never delete partial output during recovery",
                        effect.command_path
                    ));
                }
            }
        }
        Ok(())
    }

    /// `EffectManifest::build` applies `DRY_RUN_CAPABLE` with `if let Some`,
    /// so a name that matches no manifest entry -- a typo, or a command that
    /// was renamed or removed -- is skipped WITHOUT ANY SIGNAL. The roster
    /// would silently stop granting a flag that still exists, and
    /// `dry_run_refusal_message` would begin refusing a working `--dry-run`.
    /// This is the gate that makes that loud. It is core-only: it needs no
    /// parser, just the manifest's own key set.
    ///
    /// It also guards the empty world. An empty or drastically shortened
    /// roster is not a clean bill of health, it is the instrument failing.
    #[test]
    fn dry_run_capable_roster_names_only_real_commands() -> TestResult {
        let manifest = EffectManifest::build();
        let mut unmatched = Vec::new();
        for path in DRY_RUN_CAPABLE {
            if manifest.get(path).is_none() {
                unmatched.push(*path);
            }
        }
        if !unmatched.is_empty() {
            return Err(format!(
                "{} DRY_RUN_CAPABLE name(s) match no manifest command, and build() skips \
                 them silently: {}",
                unmatched.len(),
                unmatched.join(", ")
            ));
        }

        let mut duplicates: Vec<&str> = Vec::new();
        let mut seen = std::collections::BTreeSet::new();
        for path in DRY_RUN_CAPABLE {
            if !seen.insert(*path) {
                duplicates.push(*path);
            }
        }
        if !duplicates.is_empty() {
            return Err(format!(
                "DRY_RUN_CAPABLE lists {} name(s) twice: {}",
                duplicates.len(),
                duplicates.join(", ")
            ));
        }

        if DRY_RUN_CAPABLE.len() < 80 {
            return Err(format!(
                "DRY_RUN_CAPABLE holds {} names; it held 96 when measured against Clap at \
                 ddc11bb21. A collapse this large is the roster being emptied, not 16 \
                 commands dropping `--dry-run`. Confirm against the parser before lowering \
                 this floor",
                DRY_RUN_CAPABLE.len()
            ));
        }
        Ok(())
    }

    /// NARROWED FROM `all_mutating_commands_have_dry_run_option` (bd-3j6l3).
    /// This is a deliberate reduction in what is asserted and the lost coverage
    /// is named below, because a narrowing whose lost coverage is not stated is
    /// indistinguishable from a weakening.
    ///
    /// WHAT IT ASSERTED BEFORE: that every mutating command declares
    /// `dry_run_effect: Some(_)`.
    ///
    /// WHY THAT WAS NOT A PROPERTY WORTH ASSERTING: `dry_run_effect` is read at
    /// runtime by `dry_run_refusal_message` to decide whether `--dry-run` is
    /// honoured, so it is a claim about the PARSER. This gate never consulted
    /// the parser; it asked the manifest about itself, and every mutating
    /// constructor hardcoded `Some(EffectClass::ReadOnly)`, so it was satisfied
    /// by a value no author chose. It passed for 191 commands of which 90 could
    /// not accept `--dry-run` at all, and the single command it ever failed --
    /// `ask`, which took `--read-only` instead -- was the one telling the truth.
    /// A gate that only fails on the truthful entry is not measuring the defect.
    ///
    /// WHAT IT ASSERTS NOW: the one implication that is checkable without the
    /// parser -- a command claiming a preview must say what the preview does.
    ///
    /// WHAT IT CAN NO LONGER CATCH: a mutating command that supports
    /// `--dry-run` and forgets to declare it. Nothing in core can catch that,
    /// because core cannot see Clap. That case is now covered, in both
    /// directions and by name, by
    /// `cli::tests::mutating_dry_run_declarations_match_the_parser`, which
    /// resolves each declared path against the real command tree. If that test
    /// is ever deleted, this coverage is gone and not merely relocated.
    #[test]
    fn every_mutating_command_declares_a_dry_run_behavior_when_it_claims_the_flag() -> TestResult {
        let manifest = EffectManifest::build();

        for effect in manifest.mutating_commands() {
            if effect.dry_run_effect.is_some()
                && effect.mutation_contract.dry_run_behavior.is_none()
            {
                return Err(format!(
                    "Mutating command '{}' declares dry_run_effect: Some(_) but no \
                     mutation_contract.dry_run_behavior, so it claims a preview and does not \
                     say what the preview does",
                    effect.command_path
                ));
            }
        }
        Ok(())
    }

    #[test]
    fn effect_class_ordering_is_monotone() -> TestResult {
        // ReadOnly < DerivedArtifactWrite < DurableMemoryWrite < ...
        ensure(
            EffectClass::ReadOnly < EffectClass::DerivedArtifactWrite,
            true,
            "read_only < derived_artifact_write",
        )?;
        ensure(
            EffectClass::DerivedArtifactWrite < EffectClass::DurableMemoryWrite,
            true,
            "derived_artifact_write < durable_memory_write",
        )?;
        ensure(
            EffectClass::DurableMemoryWrite < EffectClass::WorkspaceFileWrite,
            true,
            "durable_memory_write < workspace_file_write",
        )
    }

    /// Every declared `db_tables` name, with the command paths declaring it.
    fn declared_db_tables(
        manifest: &EffectManifest,
    ) -> std::collections::BTreeMap<&'static str, Vec<&'static str>> {
        let mut declared = std::collections::BTreeMap::new();
        for path in manifest.command_paths() {
            if let Some(effect) = manifest.get(path) {
                for table in &effect.write_surfaces.db_tables {
                    declared
                        .entry(*table)
                        .or_insert_with(Vec::new)
                        .push(effect.command_path);
                }
            }
        }
        declared
    }

    /// The table set of a freshly migrated store (the same set `ee init`
    /// creates, measured in bd-hb5os).
    fn migrated_schema_tables() -> Result<std::collections::BTreeSet<String>, String> {
        let connection =
            crate::db::DbConnection::open_memory().map_err(|error| error.to_string())?;
        connection.migrate().map_err(|error| error.to_string())?;
        Ok(connection
            .list_user_tables()
            .map_err(|error| error.to_string())?
            .into_iter()
            .collect())
    }

    /// bd-hb5os: a `db_tables` name is admissible only if it is a table of
    /// the migrated schema. An empty world is red, not a vacuous pass: no
    /// declared name, or a schema with no table, fails.
    fn declared_tables_missing_from_schema(
        declared: &std::collections::BTreeMap<&'static str, Vec<&'static str>>,
        schema: &std::collections::BTreeSet<String>,
    ) -> TestResult {
        if declared.is_empty() {
            return Err("empty world: the manifest declares no db_tables name".to_owned());
        }
        if schema.is_empty() {
            return Err("empty world: the migrated schema has no table".to_owned());
        }
        let missing: Vec<(&&str, &Vec<&str>)> = declared
            .iter()
            .filter(|(table, _)| !schema.contains(**table))
            .collect();
        if missing.is_empty() {
            Ok(())
        } else {
            Err(format!(
                "db_tables names with no table in the migrated schema (name, declared by): {missing:?}"
            ))
        }
    }

    #[test]
    fn every_declared_db_table_exists_in_the_migrated_schema() -> TestResult {
        let manifest = EffectManifest::build();
        declared_tables_missing_from_schema(
            &declared_db_tables(&manifest),
            &migrated_schema_tables()?,
        )
    }

    #[test]
    fn declared_table_guard_is_red_on_an_empty_world_and_on_a_phantom() -> TestResult {
        let declared = declared_db_tables(&EffectManifest::build());
        let schema = migrated_schema_tables()?;
        let no_names =
            declared_tables_missing_from_schema(&std::collections::BTreeMap::new(), &schema);
        ensure(
            no_names
                .as_ref()
                .err()
                .is_some_and(|error| error.contains("declares no db_tables name")),
            true,
            "a manifest with no declared name is red for that reason",
        )?;
        let no_tables =
            declared_tables_missing_from_schema(&declared, &std::collections::BTreeSet::new());
        ensure(
            no_tables
                .as_ref()
                .err()
                .is_some_and(|error| error.contains("schema has no table")),
            true,
            "an empty schema is red for that reason",
        )?;
        let mut phantom = declared.clone();
        phantom.insert("hb5os_phantom_table", vec!["phantom control"]);
        let with_phantom = declared_tables_missing_from_schema(&phantom, &schema);
        ensure(
            with_phantom
                .as_ref()
                .err()
                .is_some_and(|error| error.contains("hb5os_phantom_table")),
            true,
            "a declared name outside the schema is red and named",
        )
    }
}