@wjarka/cezarion 0.10.0-pr30.54

package/dist/workflows/run.js

@@ -0,0 +1,3389 @@
+import { randomUUID } from 'node:crypto';
+import { spawn } from 'node:child_process';
+import { existsSync, mkdirSync, readFileSync, readdirSync, rmSync, writeFileSync } from 'node:fs';
+import { dirname, join } from 'node:path';
+import { parseAskMarkerResult, stripAskMarker, } from '../core/ask.js';
+import { onUsage, registerRunProcess, unregisterRunProcess } from '../core/process-usage.js';
+import { parseUsageLimit } from '../core/usage-limit.js';
+import { createRunner } from '../core/runner-factory.js';
+import { modelConflictsWithRunner } from '../core/model-presets.js';
+import { AGENT_MODELS_LOCKED_ERROR, agentModelsLocked } from '../core/agent-model-policy.js';
+import { ModelIdentityError, formatModelIdentity, normalizeModelForBackend, } from '../core/model-identity.js';
+import { HANDOFF_ONLY_INSTRUCTIONS, HANDOFF_INSTRUCTIONS, appendHandoffHeartbeat, followupsEnabled, handoffPath, seedHandoffFile, } from '../handoff.js';
+import { todosPath } from '../todos.js';
+import { discoverSkills } from '../skills.js';
+import { materializeSkillDir } from '../skills-remote.js';
+import { seedAgentConfigLocalLayer } from '../agent-config/seed.js';
+import { readAgentModelProvider } from '../agent-config/models.js';
+import { loadConfig, resolveWorktreeRetention } from '../config.js';
+import { autosaveCommit, createWorktree, resolveBaseRef, worktreeDiff, worktreeShortstat } from '../git-worktree.js';
+import { getHeadCommit, getRepoInfo } from '../server/git.js';
+import { loadWorkflows } from './load.js';
+import { reclaimWorktrees, rematerializeReclaimedWorktree } from '../runs/retention.js';
+import { AgentTempDirError, agentTmpEnv, removeAgentTmpDir, sweepAgentTmpDirs, } from '../runs/agent-tmpdir.js';
+import { extractTaskRefs, refineTaskRefs, titleRefNumber } from '../runs/task-refs.js';
+import { parseTaskMarkers, stripTaskMarkers } from '../runs/task-markers.js';
+import { autoNamingActive, generateRunName, liveTitleUpdatesEnabled, postValidateTitle } from '../runs/auto-name.js';
+import { reviewGateEnabled } from '../runs/review-gate.js';
+import { resolveProfileEnvForRoot } from '../workspace/agent-profiles.js';
+import { DEFAULT_AGENT_ACCOUNT_ID } from '../workspace/agent-accounts.js';
+import { WorkspaceSemaphore } from '../workspace/semaphore.js';
+import { UiEventSink } from '../runs/ui-event-sink.js';
+import { allowedToolsForStep, chainStepNote, stepKind, } from './types.js';
+const CHECK_OUTPUT_CAP = 20_000;
+async function configuredModelProvider(backend, repoRoot) {
+    return readAgentModelProvider(backend, repoRoot).catch(() => undefined);
+}
+/** An interactive session that hears nothing from the user closes itself. */
+export const IDLE_TIMEOUT_MS = 15 * 60_000;
+/**
+ * Task-completion marker from the agent contract (HANDOFF_INSTRUCTIONS): a
+ * turn whose text ends with `CEZ:DONE` means "goal achieved, nothing to ask" —
+ * the session is closed right away instead of parking at `waiting` (#347).
+ * Detection runs on the accumulated turn text so delta-streaming backends
+ * (codex, opencode) can't split the marker across text events.
+ */
+const DONE_MARKER_RE = /CEZ:DONE\s*$/;
+/**
+ * Still-working marker from the agent contract (spec
+ * 2026-07-18-subagent-monitoring-status, #490): a turn whose text ends with
+ * `CEZ:MONITORING` means "I ended this turn but I'm still working on my own
+ * downstream work (a sub-agent / a command I'm monitoring), not waiting on the
+ * user" — cezar parks it as `running`/`activity:'monitoring'` instead of
+ * `waiting`, so the cockpit shows a non-attention state. `CEZ:DONE` wins if both
+ * appear. Detected on accumulated turn text (like `CEZ:DONE`) so delta-streaming
+ * backends can't split the marker across text events.
+ */
+const MONITORING_MARKER_RE = /CEZ:MONITORING\s*$/;
+/**
+ * Preserve boundaries between complete assistant text blocks while a turn is
+ * accumulated for marker parsing. The runners join these same v1 blocks with
+ * newlines in `AgentRunResult`; matching that contract here prevents a
+ * trailing `CEZ:TITLE=` block from absorbing later commentary (#623).
+ */
+export function appendTurnText(current, next) {
+    if (!current)
+        return next;
+    if (!next)
+        return current;
+    return `${current}\n${next}`;
+}
+/** Strip a trailing marker from one text event so transcripts stay free of
+ *  protocol noise. Delta backends may split the marker across events — then
+ *  it stays visible; detection above is unaffected. */
+function stripDoneMarker(text) {
+    return text.replace(/\s*CEZ:DONE\s*$/, '');
+}
+/** Strip a trailing `CEZ:MONITORING` marker from one text event (see
+ *  `stripDoneMarker`; same delta-backend caveat). */
+function stripMonitoringMarker(text) {
+    return text.replace(/\s*CEZ:MONITORING\s*$/, '');
+}
+/** Emit the v2 `ask.requested` event for a parsed marker (the cockpit renders
+ *  it as an ask card, #473). Returns the minted request id. */
+function emitAskRequested(sink, ask) {
+    const requestId = randomUUID();
+    sink.handle({ type: 'ask.requested', requestId, questions: ask.questions });
+    return requestId;
+}
+/** A persisted, non-fatal explanation for protocol-shaped text that could not
+ * become an ask card. Never include the raw payload in this diagnostic. */
+function askMarkerRejection(result) {
+    if (result.kind === 'invalid-json') {
+        return 'structured question ignored — CEZ:ASK payload is not valid JSON';
+    }
+    if (result.kind !== 'invalid-structure')
+        return undefined;
+    const issue = result.issues[0];
+    const location = issue?.path.length ? ` at ${issue.path.join('.')}` : '';
+    return `structured question ignored — CEZ:ASK payload failed validation${location}${issue ? `: ${issue.message}` : ''}`;
+}
+/** Periodic "cezar autosave" commit in the task worktree (spec 006). */
+export const AUTOSAVE_INTERVAL_MS = 90_000;
+/** The periodic autosave timer is opt-in (#471): off, a task branch carries only the
+ *  agent's own commits plus the turn-end/pre-PR flushes — no mid-run "cezar autosave"
+ *  noise interleaving PR history. The flushes (`autosaveCommit` at turn end and before
+ *  a draft PR) are NOT gated: the branch must still end holding the finished state. */
+export function periodicAutosaveEnabled(env = process.env) {
+    return env.CEZ_AUTOSAVE === '1';
+}
+/**
+ * Explicitly opt out of the repository-root lease for runs that execute in the
+ * current checkout. This covers explicit worktree opt-out, non-Git degradation,
+ * and continuations whose worktree cannot be restored (spec 006 hardening, #438).
+ * This is intentionally unsafe: concurrent agents may overwrite each other's
+ * files or Git state. Isolated worktree runs are unaffected.
+ */
+export function repositoryRootLockDisabled(env = process.env) {
+    return env.CEZ_DISABLE_REPO_LOCK === '1';
+}
+const REPOSITORY_ROOT_LOCK_DISABLED_NOTE = 'repository-root lock disabled by CEZ_DISABLE_REPO_LOCK=1 (shared checkout is unsafe)';
+/** Safety cap on autonomous auto-continues per run — stops a stuck agent from nudging forever. */
+const MAX_AUTO_CONTINUES = 40;
+const AUTONOMOUS_NUDGE = 'Continue working autonomously until the task is fully complete. Do not ask me for confirmation or clarification — make reasonable assumptions and proceed. When everything is done, end the session with your done signal.';
+const MONITORING_WAKE_NUDGE = 'Re-check the downstream work you were monitoring. Continue toward the task goal; emit CEZ:MONITORING again only if it is still pending.';
+/**
+ * Auto-resume after a provider usage limit (spec 2026-08-03-auto-resume-after-usage-limit).
+ *
+ * The wait is the provider's own reset instant plus this grace: resuming AT the boundary races the
+ * provider's clock (and its rounding), and one failed resume costs the whole window over again.
+ * Thirty seconds is cheap next to five hours and long enough to be past any sane skew.
+ */
+export const AUTO_RESUME_GRACE_MS = 30_000;
+/**
+ * Consecutive automatic resumes allowed without a human turn. A resume can only fire after a real
+ * reset instant, so this is not a throttle — it is the backstop for the pathological case (a
+ * provider that answers "limit reached, retry now" in a loop), and it is deliberately generous
+ * enough to sit through a couple of days of five-hour windows.
+ */
+export const MAX_AUTO_RESUMES = 12;
+/**
+ * How long a missed deadline stays worth acting on. The promise is "we pick this up when the
+ * window reopens" — kept across a restart or an overnight close, which is the case the feature
+ * exists for. A day later it is no longer that promise: the user has moved on, and a task
+ * springing back to life is a surprise rather than a service. Such a deadline is retired with a
+ * note instead of fired, so the only tasks a sweep can revive are ones someone is still waiting on.
+ */
+export const AUTO_RESUME_MISSED_WINDOW_MS = 24 * 60 * 60_000;
+/**
+ * How often the queue checks that it is not wedged.
+ *
+ * A hold is the only thing in the engine that can make an idle queue CORRECT, so it is also the
+ * only thing that can make a wedged one look correct. This tick is the way out: cheap (a few
+ * in-memory checks), unref'd, and it only ever acts when idling has no justification left.
+ */
+export const QUEUE_WATCHDOG_MS = 60_000;
+/** Shared empty holds for the common "nothing is held" pump — avoids allocating per sweep. */
+const NO_HOLDS = { deadline: new Set(), inFlight: new Set() };
+/**
+ * May this run start, given what its account is holding?
+ *
+ * The two kinds of hold bind different work, and getting that wrong has produced a bug in each
+ * direction (spec 2026-08-03-auto-resume-after-usage-limit):
+ *
+ *  - a `deadline` hold means the window is KNOWN shut until an instant, so it blocks everything
+ *    on that account — resumes included. Exempting them let four resumes fire at once and
+ *    re-limit one after another, which is the stampede wearing a different hat.
+ *  - an `inFlight` hold means a resume is testing the window right now and nothing is proven, so
+ *    it blocks fresh work but not other resumes. Blocking those deadlocked a live workspace.
+ */
+function accountHeldFor(run, holds, fallbackRunner) {
+    const key = runAccountKey(run, fallbackRunner);
+    if (holds.deadline.has(key))
+        return true;
+    return holds.inFlight.has(key) && !resumeInFlight(run);
+}
+/**
+ * Which agent ACCOUNT a run's work runs on — the thing a provider usage limit actually closes
+ * (spec 2026-08-03-auto-resume-after-usage-limit).
+ *
+ * Backend plus agent account, because those are the two axes a limit is scoped to: a Claude
+ * limit must never stall a Codex task, and a second Claude login is a second budget. A record
+ * that names no runner has not started yet and will take the configured default, which is what
+ * `fallbackRunner` carries; a run that HAS started always carries its resolved runner (execute
+ * persists it), and only started runs can be holding.
+ */
+export function runAccountKey(run, fallbackRunner) {
+    return `${run.runner ?? fallbackRunner}:${run.agentProfile ?? 'default'}`;
+}
+/**
+ * Is this run an automatic resume that has not completed a turn yet?
+ *
+ * Such a run is the work the reopened window is FOR, so the hold must never apply to it — not
+ * its own, and not another resume's. Two resumes that hold each other is a deadlock the queue
+ * cannot recover from: both sit `queued` with a counter and no deadline, each waiting for the
+ * other to prove a window neither will ever get to test. That is the shape a live run produced
+ * — two scheduled tasks fired, both went `queued`, and nothing in the workspace moved again.
+ *
+ * The hold exists to stop NEW work walking into a closed window. A resume is not new work.
+ */
+function resumeInFlight(run) {
+    return (run.autoResumeAttempts !== undefined && (run.status === 'queued' || run.status === 'running'));
+}
+const AUTO_RESUME_PROMPT = 'The provider usage limit that interrupted this task has reset. Read the handoff file (CEZ_HANDOFF_FILE) to recover context, then continue the task from where you left off.';
+/**
+ * The wake instant as a human reads it — local, to the SECOND, with the zone named. The
+ * transcript line is what someone scanning a stalled task actually reads, and "18:41" is not
+ * enough to tell a wait that is nearly over from one that just started; the machine-readable ISO
+ * copy lives on `RunRecord.autoResumeAt`. Server-side formatting is honest here because cezar is
+ * local-first: the process and the browser reading it are the same machine.
+ */
+function formatWakeInstant(at) {
+    return new Intl.DateTimeFormat(undefined, { dateStyle: 'medium', timeStyle: 'long' }).format(at);
+}
+/**
+ * The effective "extra" system prompt for a run (spec §protocol v2, R2 2.3):
+ * the per-run override (`POST /api/runs` `systemPrompt`) REPLACES the
+ * `config.json` default — they are the same knob at two scopes, so the more
+ * specific one wins outright; they never concatenate. Whichever wins is
+ * ADDITIVE to the skill body and the handoff contract, which always ride
+ * along (see `composeSystemPrompt`). Blank strings count as unset.
+ */
+export function resolveExtraSystemPrompt(override, configDefault) {
+    return override?.trim() || configDefault?.trim() || undefined;
+}
+/**
+ * Joins the parts of one agent step's system prompt in fixed order — skill
+ * body (most task-specific), then the run's extra prompt (user guidance, can
+ * amend the skill), then the handoff contract (always last, never optional in
+ * practice). Blank parts drop out; survivors join with the same `\n\n---\n\n`
+ * divider the skill+handoff composition has always used.
+ */
+export function composeSystemPrompt(...parts) {
+    return parts
+        .map((p) => p?.trim())
+        .filter((p) => Boolean(p))
+        .join('\n\n---\n\n');
+}
+/**
+ * The directories a spawned agent may reach outside its worktree: the run-state
+ * folder that holds its handoff file, plus its own temp directory when this run
+ * got one (#785). Handing an agent a `TMPDIR` its file tools are not allowed to
+ * write would trade one silent failure for another, so the two travel together;
+ * under `CEZ_AGENT_TMPDIR=0` there is no per-run directory and the list is
+ * exactly what it always was.
+ */
+export function agentDirectories(runsDir, env) {
+    return env.TMPDIR ? [runsDir, env.TMPDIR] : [runsDir];
+}
+/**
+ * Materialized pasted attachment: the on-disk name/serving-URL pair the
+ * transcript already used, plus the absolute path that lets the agent
+ * operate on the file itself — save it, `cp` it, attach it to a GitHub
+ * issue/PR (#357). `path` is only ever an absolute path under
+ * `.ai/cezar/runs/<runId>-images/` (see `RunManager.persistImage`).
+ */
+/** Inverse of `persistImage`'s extension mapping (#472) — a persisted attachment
+ *  is re-encoded from disk at dequeue and needs its media type back. */
+export function mediaTypeFor(name) {
+    const ext = name.split('.').pop()?.toLowerCase();
+    return ext === 'jpg' ? 'image/jpeg'
+        : ext === 'webp' ? 'image/webp'
+            : ext === 'gif' ? 'image/gif'
+                : 'image/png';
+}
+/** Highest `<prefix>-<n>.<ext>` suffix already present in a run's image dir (#472).
+ *  `screenshot-*` and `pasted-*` share one numbering space, so this scans both and
+ *  returns 0 for a missing/empty directory. */
+export function highestImageSeq(dir) {
+    try {
+        return readdirSync(dir).reduce((max, name) => {
+            const m = /^(?:screenshot|pasted)-(\d+)\./.exec(name);
+            return m ? Math.max(max, Number(m[1])) : max;
+        }, 0);
+    }
+    catch {
+        return 0;
+    }
+}
+/**
+ * Plain-text note listing the absolute paths of pasted attachments, appended
+ * to the message that carries them (#357). The base64 image blocks stay in
+ * the message for the model to *view*; this note is what lets it *use* the
+ * files as files — and the only usable reference on backends (codex,
+ * opencode) whose `textOf()` drops image blocks before reaching the model.
+ */
+export function pastedAttachmentsText(attachments) {
+    const list = attachments.map((a) => `- ${a.path}`).join('\n');
+    return (`The user attached ${attachments.length} pasted file${attachments.length > 1 ? 's' : ''}, ` +
+        `also saved on disk at:\n${list}\n` +
+        `When the task involves saving, uploading, attaching, or transforming the pasted content ` +
+        `(e.g. attaching to a GitHub issue/PR, copying into the repo), operate on these files — do ` +
+        `not attempt to reconstruct them from the conversation.`);
+}
+/** Same note as `pastedAttachmentsText`, wrapped as a trailing `ContentBlock`
+ *  ready to append to a message's content array. */
+export function pastedAttachmentsNote(attachments) {
+    return { type: 'text', text: pastedAttachmentsText(attachments) };
+}
+/** Variant letters + the fixed diversification hints (spec 010). A runs the
+ *  task verbatim; B/C get one constant sentence each — zero configuration. */
+export const VARIANT_LETTERS = ['A', 'B', 'C'];
+const VARIANT_HINTS = {
+    A: undefined,
+    B: 'Approach hint: prefer the minimal, surgical change.',
+    C: 'Approach hint: prefer a thorough, structural approach.',
+};
+const RESTART_CONTINUATION_PROMPT = 'The cezar process restarted while you were working on this task. Read the handoff file (CEZ_HANDOFF_FILE) to recover context, then continue the task from where you left off.';
+/**
+ * The mini workflow engine: executes a `WorkflowDef` against a repo, one step
+ * at a time, persisting every event to the RunStore (which the SSE endpoints
+ * relay live to the GUI). No GitHub choreography — agent steps and shell
+ * checks with bounded retry loops, plus live sessions: the last agent step
+ * stays open for follow-ups (`waiting`) until "finish", idle timeout, or
+ * cancel. Runs queue behind the workspace-wide `maxParallel` slots (the shared
+ * `WorkspaceSemaphore`, spec 2026-07-20 step 2.5) and each run executes in its
+ * own git worktree on a `cez/<id8>` branch (spec 006), autosave-committed at
+ * turn end and before a draft PR — plus every 90 s when opted in via
+ * CEZ_AUTOSAVE=1 (#471). Each autosave records its trigger in the commit
+ * subject, so the always-on flushes are not mistaken for the opt-in timer.
+ * The user's working tree is never touched.
+ */
+export class RunManager {
+    store;
+    repoRoot;
+    active = new Map();
+    // Queue + `starting` set (spec 006, janitor's pump() pattern): `starting`
+    // covers the window between shifting a run off the queue and the run
+    // registering in `active`, so parallel-slot counting is never racy.
+    queue = [];
+    starting = new Set();
+    // Runs parked at `waiting` (open session, ball in the user's court). They
+    // don't consume a `maxParallel` slot (#347) — an idle claude process costs
+    // memory but no tokens, queued work progressing matters more, and the idle
+    // timeout already bounds how long a session can sit open. Invariant:
+    // `waiting ⊆ active` — always cleared together via dropActive().
+    waiting = new Set();
+    /** Durable monitoring subset. Only the configured number receives the waiting-slot exemption. */
+    monitoring = new Set();
+    pendingJobs = new Map();
+    /** Interrupted agent turns recovered after a process restart. Unlike an
+     *  explicit user Continue, these are bulk scheduler work and must re-enter
+     *  through `pump()` so both workspace and per-project caps are honored. */
+    pendingContinuations = new Map();
+    /** Per-run image counter behind `pasted-<n>` / `screenshot-<n>` (#472). Lives on
+     *  the manager rather than the `ActiveRun` so a *queued* run — which has no
+     *  `ActiveRun` at all — can persist attachments. Seeded lazily from disk. */
+    queuedImageSeq = new Map();
+    /** Messages that landed in the dequeue → session-open gap (#472), flushed as
+     *  ordinary follow-up turns the moment the session opens. In-memory only. */
+    deferredMessages = new Map();
+    /** Armed usage-limit resumes, keyed by run id (spec
+     *  2026-08-03-auto-resume-after-usage-limit). The DEADLINE itself lives on the record
+     *  (`autoResumeAt`) — this map holds only the process-local timer, so a restart rebuilds it
+     *  from the record rather than losing the wait. Runs here are `failed` and therefore NOT in
+     *  `active`, which is why the timer cannot live on an `ActiveRun` like the monitoring one. */
+    autoResumeTimers = new Map();
+    pumping = false;
+    /** A pump that arrived while one was in flight — replayed by `pump()`'s own
+     *  loop so a slot freed mid-sweep is never a lost wakeup. */
+    pumpAgain = false;
+    /**
+     * Runs normally isolate in worktrees and may execute in parallel. When that
+     * isolation is unavailable (or explicitly disabled), access to `repoRoot` is
+     * serialized by default so two agents cannot edit/revert the same files
+     * (#438). `CEZ_DISABLE_REPO_LOCK=1` deliberately bypasses this safety lease.
+     */
+    repoRootTail = Promise.resolve();
+    /** `.ai/cezar` — where the per-task handoff files and todos.json live. */
+    dataDir;
+    /** Runs currently being paused by the memory guard — dedupes the ~2 s samples so one breach
+     *  triggers one pause, not a burst. Cleared in dropActive when the run leaves the registry. */
+    memoryPausing = new Set();
+    /** Unsubscribe handle for the constructor's `onUsage` subscription — released
+     *  by dispose() so a torn-down manager stops receiving sampler ticks. */
+    offUsage;
+    /** The stalled-queue watchdog (see `rescueStalledQueue`). */
+    queueWatchdog;
+    /** Set by the watchdog for exactly one sweep: ignore the usage-limit hold and make progress. */
+    forceNextPump = false;
+    /** Runs the watchdog started despite the hold. The spawn-time gate (`requeueWhileHeld`) would
+     *  otherwise hand them straight back and the rescue would undo itself in a millisecond. */
+    forceStarted = new Set();
+    /** The workspace-wide parallel-cap semaphore + cached resource config
+     *  (spec 2026-07-20, step 2.5). Boot constructs ONE and every manager shares
+     *  it; the private fallback keeps single-manager callers and tests working. */
+    semaphore;
+    /** Unregister handle for this manager's semaphore membership — released by
+     *  dispose() so a torn-down project stops counting against the cap. */
+    offSemaphore;
+    constructor(store, repoRoot, options = {}) {
+        this.store = store;
+        this.repoRoot = repoRoot;
+        this.dataDir = join(repoRoot, '.ai/cezar');
+        this.semaphore = options.semaphore ?? new WorkspaceSemaphore();
+        this.offSemaphore = this.semaphore.register({
+            busySlots: () => this.busySlots(),
+            pump: () => this.pump(),
+            oldestQueuedAt: () => this.oldestQueuedAt(),
+            accountHolds: () => this.accountHolds(),
+        });
+        // Memory guard (#memory-guard): the shared process-tree sampler already ticks ~every 2 s for
+        // the runs table; piggyback on it to enforce the per-task memory ceiling.
+        this.offUsage = onUsage((snapshot) => void this.enforceMemoryLimit(snapshot));
+        this.queueWatchdog = setInterval(() => void this.rescueStalledQueue(), QUEUE_WATCHDOG_MS);
+        this.queueWatchdog.unref?.();
+    }
+    /**
+     * Release everything this manager owns without touching run records
+     * (multi-project workspace, spec 2026-07-20: a removed project's context is
+     * torn down while the process lives on). Unsubscribes the shared usage
+     * sampler — before dispose() existed that subscription lived for the whole
+     * process — clears every per-run idle/autosave timer, releases any held
+     * repo-root locks, and empties the queued state so nothing fires later.
+     * Live sessions are NOT ended here: run lifecycle stays the caller's policy;
+     * dispose only guarantees the manager makes no further moves on its own.
+     */
+    dispose() {
+        this.offUsage();
+        this.offSemaphore();
+        clearInterval(this.queueWatchdog);
+        for (const [runId, state] of this.active) {
+            this.clearIdleTimer(state);
+            this.clearMonitoringWakeTimer(state, runId);
+            this.clearAutosaveTimer(state);
+            state.releaseRepoRoot?.();
+            state.releaseRepoRoot = undefined;
+        }
+        for (const timer of this.autoResumeTimers.values())
+            clearTimeout(timer);
+        this.autoResumeTimers.clear();
+        this.active.clear();
+        this.waiting.clear();
+        this.starting.clear();
+        this.queue.length = 0;
+        this.pendingJobs.clear();
+        this.pendingContinuations.clear();
+        this.memoryPausing.clear();
+        this.lastNamerKey.clear();
+    }
+    /**
+     * Pause any active run whose whole process tree exceeds the WORKSPACE
+     * `resources.memoryLimitMb`, freeing its slot so the queue advances
+     * (#memory-guard). "Pause" closes the session — freeing the tree's
+     * memory — and leaves the run resumable via Continue; a loud warning explains why. No-op when
+     * no limit is set or the sampler has no data (e.g. `ps`/PowerShell unavailable).
+     */
+    async enforceMemoryLimit(snapshot) {
+        // The sampler is module-global (one `ps` for the whole process), so with
+        // multiple projects a snapshot carries EVERY project's runs. Act only on
+        // rows this manager owns (multi-project spec, step 2.4).
+        const runIds = Object.keys(snapshot).filter((runId) => this.active.has(runId));
+        if (runIds.length === 0)
+            return;
+        // Workspace limit from the shared semaphore's in-memory cache (step 2.5:
+        // refreshed at boot and on PUT /api/workspace/config — never N per-tick
+        // file reads across N projects). Legacy per-repo `memoryLimitMb` keys are
+        // ignored post-migration.
+        const limitMb = this.semaphore.memoryLimitMb();
+        if (!limitMb || limitMb <= 0)
+            return;
+        const limitBytes = limitMb * 1024 * 1024;
+        for (const runId of runIds) {
+            const usage = snapshot[runId];
+            if (!usage || usage.rssBytes <= limitBytes)
+                continue;
+            if (this.memoryPausing.has(runId))
+                continue;
+            const state = this.active.get(runId);
+            if (!state?.session?.open || state.cancelled)
+                continue;
+            this.memoryPausing.add(runId);
+            const usedMb = Math.round(usage.rssBytes / (1024 * 1024));
+            this.store.appendEvent(runId, {
+                type: 'note',
+                message: `⚠ memory limit exceeded — this task's process tree is using ${usedMb} MiB (limit ${limitMb} MiB). Pausing it and letting the next queued task run; resume it with Continue.`,
+            });
+            this.store.appendEvent(runId, {
+                type: 'lifecycle',
+                message: `paused — memory limit exceeded (${usedMb} MiB > ${limitMb} MiB)`,
+            });
+            // Closing the session frees the tree and lets the normal exit path settle the run and
+            // pump the queue. Suppress autonomous auto-continue so the pause actually holds.
+            state.autonomous = false;
+            this.clearIdleTimer(state);
+            state.session.end();
+        }
+    }
+    /** Env the spawned claude gets so the agent can find its handoff file and
+     *  the global inbox (spec 007; the inbox only when the run opted in).
+     *
+     *  `CEZ_TODOS_FILE` is set to `''` rather than omitted when follow-ups are
+     *  off: runners spawn with `{ ...process.env, ...spec.env }`, so omitting the
+     *  key would let a value inherited from *this* process through — a nested
+     *  cezar (an agent running `cez serve`/`cez run`/the test suite) would then
+     *  write follow-ups into the parent's inbox despite the opt-out. Empty is the
+     *  established "absent" spelling — consumers guard with `if (todosFile)`.
+     *
+     *  `TMPDIR`/`TEMP`/`TMP` (#785) point at this run's own scratch directory
+     *  instead of the machine-wide one every agent used to share. Created and
+     *  write-probed here, on the last common path before a spawn, so an unusable
+     *  temp directory throws `AgentTempDirError` at the caller rather than
+     *  turning into empty command output inside a running agent. */
+    agentEnv(runId, generateFollowups = true) {
+        return {
+            CEZ_HANDOFF_FILE: handoffPath(this.dataDir, runId),
+            CEZ_TASK_ID: runId,
+            CEZ_TODOS_FILE: generateFollowups ? todosPath(this.dataDir) : '',
+            ...agentTmpEnv(this.dataDir, runId),
+        };
+    }
+    /**
+     * `agentEnv` plus the agent-account variable for the profile this STEP runs under (spec
+     * 2026-07-29-agent-profiles), and the id it resolved to so the caller can record it.
+     *
+     * Resolved per step, not per run, because a workflow can mix backends: an override naming a
+     * Claude account says nothing about which Codex account a codex step should use. Resolution
+     * order, most specific first:
+     *
+     *   1. the step's ALREADY-RECORDED `profileId` — a resume or Continue must reattach to the
+     *      account that created the session, whatever the project has since been switched to;
+     *   2. the run's composer override, but only for steps on the run's own runner;
+     *   3. the project's stored selection, and failing that the discovered default.
+     *
+     * Read fresh every time. `~/.cezar/config.json` is shared by every cezar process on this
+     * machine, so a cached snapshot is a staleness bug, and one small JSON read is free next to
+     * spawning a CLI. Never throws: an unreadable home degrades to the default profile, which is
+     * exactly the behaviour that predates profiles.
+     */
+    async agentEnvForStep(runId, backend, options = {}) {
+        const run = this.store.getRun(runId);
+        const profileId = options.recordedProfileId
+            ?? (backend === (run?.runner ?? 'claude') ? run?.agentProfile : undefined);
+        const resolved = await resolveProfileEnvForRoot(this.repoRoot, backend, profileId);
+        return {
+            env: { ...this.agentEnv(runId, options.generateFollowups), ...resolved.env },
+            profileId: resolved.profile.id,
+        };
+    }
+    startRun(workflow, input, group) {
+        // Sanitize at the manager boundary so CLI runs, workflows, variants, and
+        // direct callers cannot bypass the HTTP policy.
+        const effectiveInput = agentModelsLocked(this.repoRoot)
+            ? { ...input, model: undefined }
+            : input;
+        const run = this.store.createRun({
+            title: makeRunTitle(input.task, workflow) + (group ? ` (${group.variant})` : ''),
+            workflow: workflow.name,
+            task: input.task,
+            model: effectiveInput.model,
+            runner: input.runner,
+            // The composer's per-task account (spec 2026-07-29-agent-profiles). Persisted at creation
+            // so a queued run picks it up at dequeue and every later resume reads the same answer.
+            agentProfile: input.agentProfile,
+            // The global inbox is the ceiling on the per-run flag (#471). Enforced here rather than
+            // at the HTTP route because `cezar run`, the inbox's own "▶ Run" and variants all reach
+            // startRun directly — a route-level gate would leave those writing todos.json.
+            generateFollowups: followupsEnabled() ? input.generateFollowups : false,
+            // Persist autonomy on the record (#489) so the terminal review gate
+            // (`settleSuccess`) and the group-pick winner-park can honor it — mid-run
+            // auto-nudge reads `input.autonomous` (`execute`), but the record is the
+            // only source those after-the-fact consumers have.
+            autonomous: input.autonomous === true,
+            // Persist the explicit opt-out so queued-run restart recovery and the
+            // session Git routes can distinguish it from a removed isolated worktree.
+            worktree: !group && input.worktree === false ? false : undefined,
+            groupId: group?.groupId,
+            variant: group?.variant,
+            steps: workflow.steps.map((s) => ({ id: s.id, name: s.name ?? s.id, kind: stepKind(s) })),
+        });
+        // Persist the full definition so a queued run survives a restart (#367) —
+        // ad-hoc "(planned)" chains exist nowhere else to re-resolve from.
+        this.store.updateRun(run.id, { workflowDef: workflow });
+        // Initial pasted images must be visible while the run is still queued (#612),
+        // and must survive a restart before a slot opens. Persist them before the job
+        // enters `pendingJobs`; `hydrateQueuedInput` reconstructs their content blocks
+        // from these URLs when a recovered run eventually starts.
+        if (input.images?.length) {
+            const persisted = input.images
+                .filter((b) => b.type === 'image')
+                .map((b) => this.persistImage(run.id, b.source.media_type, b.source.data, 'pasted'))
+                .filter((saved) => saved !== null);
+            if (persisted.length) {
+                this.store.updateRun(run.id, { taskImages: persisted.map((saved) => saved.url) });
+            }
+        }
+        // Step-0 reference extraction (task auto-naming spec): the regex layer's
+        // numbers persist immediately; the namer may add the kind it verified later.
+        const skillHint = workflow.steps.find((s) => stepKind(s) === 'agent' && s.skill)?.skill?.trim();
+        const refs = refineTaskRefs(extractTaskRefs(input.task), skillHint);
+        if (refs.prNumber !== undefined || refs.issueNumber !== undefined) {
+            this.store.updateRun(run.id, {
+                ...(refs.prNumber !== undefined ? { prNumber: refs.prNumber } : {}),
+                ...(refs.issueNumber !== undefined ? { issueNumber: refs.issueNumber } : {}),
+            });
+        }
+        // Fire-and-forget LLM naming (task auto-naming spec): the heuristic title
+        // above shows instantly; the namer's short title replaces it when (and if)
+        // the model answers. Never awaited, never fails the run.
+        void this.autoNameRun(run.id, skillHint, input.task);
+        this.pendingJobs.set(run.id, { workflow, input: effectiveInput });
+        this.queue.push(run.id);
+        void this.pump();
+        return run;
+    }
+    /**
+     * Parallel variants (spec 010): N runs of the same workflow on the same
+     * task, sharing a groupId. Variant A gets the task verbatim; B and C get a
+     * fixed one-line approach hint appended to the *task input* (not the step
+     * template), so diversification works with any workflow. The normal queue
+     * applies — with maxParallel=2 a third variant simply waits.
+     */
+    startVariants(workflow, input, count) {
+        const groupId = randomUUID();
+        return VARIANT_LETTERS.slice(0, Math.min(Math.max(count, 1), VARIANT_LETTERS.length)).map((variant) => {
+            const hint = VARIANT_HINTS[variant];
+            const task = hint ? `${input.task}\n\n${hint}` : input.task;
+            return this.startRun(workflow, { ...input, task, worktree: undefined }, { groupId, variant });
+        });
+    }
+    /**
+     * Slots this manager holds against the workspace-wide cap. `waiting` runs
+     * don't hold a slot (#347): an idle claude process costs memory but no
+     * tokens, queued work progressing matters more, and the idle timeout already
+     * bounds how long a session can sit open. Because the exemption lives HERE —
+     * in the count, not in any acquire path — a message into a `waiting` run
+     * (sendMessage) resumes it immediately even when that momentarily exceeds
+     * `maxParallel`, including when other projects saturate the cap.
+     */
+    busySlots() {
+        const ordinaryWaiting = this.waiting.size - this.monitoring.size;
+        const exemptMonitoring = Math.min(this.monitoring.size, this.semaphore.maxMonitoringSessions());
+        return this.active.size + this.starting.size - ordinaryWaiting - exemptMonitoring;
+    }
+    /** Epoch ms of this manager's oldest queued run (the semaphore's fairness
+     *  key when a freed slot is broadcast), or null when nothing is queued.
+     *  `queue` is FIFO — `startRun` pushes and `recover()` re-queues by
+     *  `createdAt` — so the head is the oldest. */
+    oldestQueuedAt() {
+        const head = this.queue[0];
+        if (!head)
+            return null;
+        const createdAt = this.store.getRun(head)?.createdAt;
+        const ms = createdAt ? Date.parse(createdAt) : Number.NaN;
+        return Number.isNaN(ms) ? null : ms;
+    }
+    /**
+     * A slot this manager held just came free. Pump the whole WORKSPACE, not
+     * just this manager: `maxParallel` is counted across every project, so the
+     * run that should take the slot is the workspace's oldest queued one — which
+     * usually sits in another project's queue. Pumping only `this` is what left
+     * a queued run in project B stuck at `queued` while project A's runs came
+     * and went. `release()` pumps this manager too, so it replaces the local
+     * `pump()` at every slot-freeing transition.
+     */
+    releaseSlot() {
+        void this.semaphore.release();
+    }
+    /**
+     * Start queued runs while parallel slots are free. A run starts only under
+     * BOTH ceilings: the WORKSPACE `resources.maxParallel` (default 2, counted
+     * across every manager — spec 2026-07-20, step 2.5) AND this project's own
+     * per-project `maxParallel` when the registry sets one (spec 2026-07-22,
+     * inherits the workspace cap when unset). Legacy per-repo `maxParallel` keys
+     * are ignored. A non-git directory degrades to 1 sequential run in the repo
+     * root (spec 006 degradation rule), which is always the tighter bound.
+     */
+    async pump() {
+        this.reconcileMonitoringWakeTimers();
+        this.reconcileAutoResumes();
+        // A pump requested while one is in flight can't just be dropped: the
+        // in-flight pass may already have read capacity (it awaits `getRepoInfo`
+        // before the first check), so a slot freed in that window would be lost
+        // until the next unrelated event. Re-run the sweep instead.
+        if (this.pumping) {
+            this.pumpAgain = true;
+            return;
+        }
+        this.pumping = true;
+        try {
+            do {
+                this.pumpAgain = false;
+                const repo = await getRepoInfo(this.repoRoot);
+                const maxParallel = this.semaphore.maxParallel();
+                // Per-project ceiling (spec 2026-07-22-per-project-concurrency): this
+                // project never runs more than its own configured `maxParallel`; absent
+                // an override it equals the workspace cap, so behavior is unchanged.
+                const projectMax = this.semaphore.projectMaxParallel(this.repoRoot);
+                // `waiting` runs don't hold a slot (#347) — see busySlots(). The check
+                // below is the only slot gate: resumes never pass through it. A run
+                // starts only under BOTH the workspace cap and this project's ceiling.
+                const capacity = () => this.semaphore.busy() < maxParallel &&
+                    this.busySlots() < projectMax &&
+                    (repo !== null || this.busySlots() < 1);
+                // The usage-limit hold (spec 2026-08-03-auto-resume-after-usage-limit).
+                //
+                // A limit closes an ACCOUNT, not a run — so starting the next queued task walks it into
+                // the same wall. Measured before this gate existed: eight tasks under `maxParallel: 2`
+                // all failed within 517 ms, each spawning a CLI (and, outside worktree-opt-out mode, a
+                // worktree and a branch) only to be marked `scheduled`. The cap was respected at every
+                // instant and was no brake at all, because a doomed run lives ~200 ms.
+                //
+                // So: while any run on an account is waiting out a limit, nothing new starts on THAT
+                // account. Other accounts (a second login, a different backend) keep running — the hold
+                // is keyed, not global. The set is derived from the durable records rather than tracked
+                // separately, which is what makes it survive a restart, expire on its own, and lift the
+                // instant a user cancels a resume.
+                // The watchdog's one-shot override — read and cleared here, so a forced sweep never
+                // leaks into the next ordinary one.
+                const forced = this.forceNextPump;
+                this.forceNextPump = false;
+                const holds = this.queue.length > 0 && !forced ? this.semaphore.accountHolds() : NO_HOLDS;
+                const anyHold = holds.deadline.size > 0 || holds.inFlight.size > 0;
+                // Only pay for the config read when something is actually held: a queued record may name
+                // no runner, and then the account it would use is the configured default.
+                const defaultRunner = anyHold ? (await loadConfig(this.repoRoot)).defaultRunner : undefined;
+                while (this.queue.length > 0 && capacity()) {
+                    // FIFO among the runs that CAN start; a held one keeps its place in the queue rather
+                    // than being dequeued and re-queued (which would churn its position and its record).
+                    const next = !anyHold
+                        ? 0
+                        : this.queue.findIndex((id) => {
+                            const queued = this.store.getRun(id);
+                            return !queued || !accountHeldFor(queued, holds, defaultRunner ?? 'claude');
+                        });
+                    if (next === -1)
+                        break; // everything queued is waiting on a held account
+                    const runId = this.queue.splice(next, 1)[0];
+                    if (!runId)
+                        break;
+                    // A forced sweep has to reach the spawn: the gate inside `execute` asks the same
+                    // question and would send this run straight back to the queue.
+                    if (forced)
+                        this.forceStarted.add(runId);
+                    const job = this.pendingJobs.get(runId);
+                    const continuation = this.pendingContinuations.get(runId);
+                    this.pendingJobs.delete(runId);
+                    this.pendingContinuations.delete(runId);
+                    if (!job && !continuation)
+                        continue;
+                    this.starting.add(runId);
+                    if (continuation) {
+                        const hydrated = this.hydrateQueuedContinuation(runId, continuation);
+                        void this.runContinuation(runId, hydrated.stepId, hydrated.sessionId, hydrated.backend, hydrated.prompt, hydrated.images, hydrated.persistedImages, hydrated.persistedAttachments).catch((err) => {
+                            const message = err instanceof Error ? err.message : String(err);
+                            this.store.updateRun(runId, {
+                                status: 'failed',
+                                error: `continue crashed: ${message}`,
+                                finishedAt: new Date().toISOString(),
+                            });
+                            this.starting.delete(runId);
+                            this.dropActive(runId);
+                        });
+                        continue;
+                    }
+                    if (!job)
+                        continue;
+                    // Rebuild the prompt from the store at the last instant (#472), so an edit
+                    // or a stacked message that landed while the run waited is honored. Entered
+                    // in the same synchronous tick as the `pendingJobs.delete` above, so no
+                    // handler can observe a half-dequeued run.
+                    const input = this.hydrateQueuedInput(runId, job.input);
+                    void this.execute(runId, job.workflow, input).catch((err) => {
+                        const message = err instanceof Error ? err.message : String(err);
+                        this.store.updateRun(runId, {
+                            status: 'failed',
+                            error: `engine crashed: ${message}`,
+                            finishedAt: new Date().toISOString(),
+                        });
+                        const state = this.active.get(runId);
+                        if (state) {
+                            this.clearIdleTimer(state);
+                            this.clearAutosaveTimer(state);
+                        }
+                        this.starting.delete(runId);
+                        this.dropActive(runId);
+                    });
+                }
+            } while (this.pumpAgain);
+        }
+        finally {
+            this.pumping = false;
+        }
+    }
+    /**
+     * Make one `queued` RECORD executable again — the engine half a queued run needs but does not
+     * persist (`pendingJobs` / `pendingContinuations` are process-local, the record is not).
+     *
+     * Two callers, one path: boot recovery re-adopts everything the previous process was holding,
+     * and the queue watchdog re-adopts anything the running process has somehow lost. A queued
+     * record with no work item behind it is invisible to `pump()` and would sit there for good,
+     * which is the worst failure this engine has — the task is neither running nor failed, just
+     * silently never going to happen.
+     *
+     * A continuation is reconstructed first: its executable details are gone, but the pending
+     * `continue-N` step and the session before it are durable, which is enough. Otherwise the
+     * workflow is revived from the record. A run that can be neither is failed loudly rather than
+     * left in the queue as a ghost.
+     */
+    async reviveQueuedRun(run, reason) {
+        const queuedContinuation = [...run.steps]
+            .reverse()
+            .find((step) => step.status === 'pending' && step.id.startsWith('continue-'));
+        const sessionStep = queuedContinuation
+            ? [...run.steps].reverse().find((step) => step.id !== queuedContinuation.id && step.sessionId)
+            : undefined;
+        if (queuedContinuation && sessionStep?.sessionId) {
+            const backend = run.runner ?? 'claude';
+            const sessionBackend = sessionStep.backend ?? backend;
+            this.pendingContinuations.set(run.id, {
+                stepId: queuedContinuation.id,
+                sessionId: sessionBackend === backend ? sessionStep.sessionId : undefined,
+                backend,
+                prompt: RESTART_CONTINUATION_PROMPT,
+                images: [],
+            });
+            this.queue.push(run.id);
+            this.store.appendEvent(run.id, {
+                type: 'lifecycle',
+                message: `${reason} — interrupted continuation re-queued`,
+            });
+            return;
+        }
+        const workflow = await this.reviveWorkflow(run);
+        if (!workflow) {
+            this.store.updateRun(run.id, {
+                status: 'failed',
+                error: 'interrupted — workflow definition not recoverable after a restart',
+                finishedAt: new Date().toISOString(),
+            });
+            this.store.appendEvent(run.id, {
+                type: 'lifecycle',
+                message: `${reason} — workflow definition not recoverable, task failed`,
+            });
+            return;
+        }
+        // Re-apply the inbox ceiling (#471). `execute()` gates again at spawn time, so the agent is
+        // safe either way — but a run queued while the inbox was on and recovered after it was
+        // switched off would otherwise keep echoing `generateFollowups: true` on a run that
+        // demonstrably produced none. Normalize the record, the way startRun does.
+        const generateFollowups = followupsEnabled() ? run.generateFollowups : false;
+        if (generateFollowups !== run.generateFollowups) {
+            this.store.updateRun(run.id, { generateFollowups });
+        }
+        this.pendingJobs.set(run.id, {
+            workflow,
+            // Folded through the same helper `pump()` uses (#472) so a restart carries the stack.
+            // Idempotent: hydration always composes from `run.task` + the stack, never from an
+            // already-folded `input.task`, so re-hydrating at dequeue yields the same string.
+            input: this.hydrateQueuedInput(run.id, {
+                task: run.task,
+                model: run.model,
+                runner: run.runner,
+                generateFollowups,
+                // Re-thread autonomy (#489): the rebuilt input feeds `execute`, whose mid-run auto-nudge
+                // reads `input.autonomous`. Without this a recovered autonomous run would run
+                // non-autonomously and later wrongly park at `review`.
+                autonomous: run.autonomous,
+                // Preserve an explicit worktree opt-out across a queued restart.
+                worktree: run.worktree,
+            }),
+        });
+        this.queue.push(run.id);
+        this.store.appendEvent(run.id, { type: 'lifecycle', message: `${reason} — task re-queued` });
+    }
+    /**
+     * Startup recovery (#367) — re-adopt runs that were live when the previous
+     * cezar process exited (requires the store opened with `keepLive`):
+     *  - `queued`  → back into the queue (FIFO by createdAt), from the persisted
+     *    workflowDef (or the catalog by name for older records);
+     *  - `waiting` → the turn was over and the ball was in the user's court —
+     *    settle exactly like a closed session (review/done, Continue still works);
+     *  - `running` → mark interrupted, then immediately resume the last agent
+     *    session via the Continue path, pointing the agent at its handoff file.
+     * Call once, before the server starts taking requests.
+     */
+    async recover() {
+        const live = this.store
+            .listRuns()
+            .filter((r) => ['queued', 'waiting', 'running'].includes(r.status))
+            .sort((a, b) => a.createdAt.localeCompare(b.createdAt));
+        // A crash never reaches `dropActive`, so its temp directory (#785) outlived the run.
+        // Startup is the one moment we know which runs are still live, so sweep every other
+        // per-run directory here — bounded to `<dataDir>/tmp`, never a sibling.
+        sweepAgentTmpDirs(this.dataDir, live.map((r) => r.id));
+        for (const run of live) {
+            if (run.status === 'queued') {
+                await this.reviveQueuedRun(run, 'cezar restarted');
+                continue;
+            }
+            if (run.status === 'waiting') {
+                for (const step of run.steps) {
+                    if (step.status === 'waiting' || step.status === 'running') {
+                        this.store.updateStep(run.id, step.id, { status: 'done', finishedAt: new Date().toISOString() });
+                    }
+                }
+                this.store.appendEvent(run.id, {
+                    type: 'lifecycle',
+                    message: 'cezar restarted — the open session was settled',
+                });
+                await this.settleSuccess(run.id);
+                continue;
+            }
+            // `running`: the process died mid-turn. Mark it interrupted (the state
+            // continueRun expects), then pick the work back up from the last session.
+            const finishedAt = new Date().toISOString();
+            for (const step of run.steps) {
+                if (step.status === 'running' || step.status === 'waiting') {
+                    this.store.updateStep(run.id, step.id, { status: 'failed', finishedAt });
+                }
+            }
+            this.store.updateRun(run.id, {
+                status: 'failed',
+                error: 'interrupted — cezar process exited during the run',
+                finishedAt,
+                currentStepId: undefined,
+            });
+            const resumed = this.continueRun(run.id, {
+                text: RESTART_CONTINUATION_PROMPT,
+            }, true);
+            this.store.appendEvent(run.id, {
+                type: 'lifecycle',
+                message: resumed.ok
+                    ? 'cezar restarted — resuming the interrupted task from its last session'
+                    : `cezar restarted — could not resume the interrupted task (${resumed.error ?? 'unknown'})`,
+            });
+        }
+        // Re-arm usage-limit resumes (spec 2026-08-03-auto-resume-after-usage-limit): the wait is
+        // routinely longer than a cezar session, so the deadline is durable and the timer is rebuilt
+        // from it. `pump()` reconciles again on every sweep, so this is the fast path, not the only
+        // one — see `reconcileAutoResumes`.
+        this.reconcileAutoResumes();
+        void this.pump();
+    }
+    /** The persisted definition when it looks sane, else the catalog by name. */
+    async reviveWorkflow(run) {
+        // "Looks sane" is the STORE's job now: it parses `workflowDef` against the definition schema
+        // and `.catch`es a def that no longer fits to `undefined`, so anything present here already
+        // has the `steps` array the old inline `Array.isArray` check was asking for.
+        const def = run.workflowDef;
+        if (def)
+            return def;
+        const { workflows } = await loadWorkflows(this.repoRoot);
+        return workflows.find((w) => w.name === run.workflow) ?? null;
+    }
+    /** Remove a run from the live registries — keeps `waiting ⊆ active`. */
+    dropActive(runId) {
+        const state = this.active.get(runId);
+        state?.releaseRepoRoot?.();
+        if (state)
+            state.releaseRepoRoot = undefined;
+        this.waiting.delete(runId);
+        this.monitoring.delete(runId);
+        if (state)
+            this.clearMonitoringWakeTimer(state, runId);
+        this.active.delete(runId);
+        this.memoryPausing.delete(runId);
+        this.lastNamerKey.delete(runId);
+        this.forceStarted.delete(runId);
+        // The run's slot is gone from busySlots() as of the deletes above — hand it
+        // to the workspace's oldest queued run, in ANY project. Every terminal path
+        // funnels through here, so this one call covers them all.
+        // Same reasoning as retention below — every terminal path funnels through here, so the
+        // usage-limit question ("did this run stop because the account is out of window, and when
+        // does that window reopen?") is asked once, in one place, off the record the failing path
+        // has already written. Nothing to do for any other outcome.
+        //
+        // BEFORE releasing the slot, and that order is the whole point: `releaseSlot` pumps every
+        // manager, and a pump reads the hold off the records. Publishing the schedule afterwards
+        // left a window — measured as exactly one extra task — where the queue saw a free slot and
+        // an account that looked healthy, and started work that was already doomed.
+        this.scheduleAutoResumeIfLimited(runId);
+        this.releaseSlot();
+        // A run leaving the active registry is a terminal transition (done/review/
+        // failed/cancelled) — the one moment the finished-worktree count can grow.
+        // Enforce count-based retention (#483) here so a single hook covers every
+        // terminal path. Fire-and-forget: retention must never delay or throw into
+        // the lifecycle.
+        void this.enforceRetention();
+        // The run's temp directory (#785) goes on the same terminal transition, and
+        // unconditionally — it is scratch, not an artifact, so unlike a worktree
+        // there is no keep-count to respect and nothing left to recover from it. A
+        // Continue (or an auto-resume) re-creates it through `agentEnv`.
+        removeAgentTmpDir(this.dataDir, runId);
+    }
+    // ---- usage-limit auto-resume (spec 2026-08-03-auto-resume-after-usage-limit) --------------
+    /**
+     * A run just failed: if the provider said "usage limit, back at T", promise to resume it at
+     * `T + AUTO_RESUME_GRACE_MS` instead of leaving the task dead until someone notices.
+     *
+     * Every refusal below is silent-but-honest — the run stays `failed` with its Continue button,
+     * which is exactly the pre-feature behavior — except the safety cap, which says so on the
+     * transcript, because a run that stops resuming itself needs to explain why.
+     */
+    scheduleAutoResumeIfLimited(runId) {
+        if (this.autoResumeTimers.has(runId))
+            return; // already promised
+        const run = this.store.getRun(runId);
+        if (!run || run.status !== 'failed')
+            return;
+        // Archiving IS resigning from a task. Reviving one because a window happened to reopen would
+        // be the feature working against the clearest signal the user can give it.
+        if (run.archived)
+            return;
+        const limit = parseUsageLimit(run.error);
+        if (!limit)
+            return;
+        if (!this.semaphore.autoResumeOnUsageLimit())
+            return;
+        // No session to resume = nothing this feature can do; `continueRun` would refuse anyway.
+        if (!run.steps.some((step) => step.sessionId))
+            return;
+        const attempts = run.autoResumeAttempts ?? 0;
+        if (attempts >= MAX_AUTO_RESUMES) {
+            this.store.appendEvent(runId, {
+                type: 'note',
+                message: `automatic resume cap reached (${MAX_AUTO_RESUMES}) — continue this task manually`,
+            });
+            return;
+        }
+        const wakeAt = new Date(limit.resetAt.getTime() + AUTO_RESUME_GRACE_MS);
+        this.armAutoResume(runId, wakeAt.getTime());
+        this.store.appendEvent(runId, {
+            type: 'lifecycle',
+            message: `usage limit reached — resuming automatically at ${formatWakeInstant(wakeAt)}`,
+        });
+    }
+    /** Publish the deadline on the record (the cockpit's only source) and arm the timer for it. */
+    armAutoResume(runId, deadline) {
+        this.store.updateRun(runId, { autoResumeAt: new Date(deadline).toISOString() });
+        const timer = setTimeout(() => this.fireAutoResume(runId), Math.max(0, deadline - Date.now()));
+        timer.unref?.();
+        this.autoResumeTimers.set(runId, timer);
+    }
+    /**
+     * The window has reopened. Re-check the record synchronously — hours may have passed, and the
+     * user may have continued, deleted or cancelled the run in them — then hand the resume to the
+     * ordinary queued-continuation path so it obeys both concurrency caps like any other work.
+     */
+    fireAutoResume(runId) {
+        this.autoResumeTimers.delete(runId);
+        const run = this.store.getRun(runId);
+        if (!run || run.status !== 'failed' || !run.autoResumeAt)
+            return;
+        // Belt and braces against the one gap `reconcileAutoResumes` cannot close: the setting going
+        // off in the window between the last pump and this tick.
+        if (!this.semaphore.autoResumeOnUsageLimit()) {
+            this.clearAutoResume(runId);
+            return;
+        }
+        const attempts = (run.autoResumeAttempts ?? 0) + 1;
+        // `continueRun` retires the pending resume (timer + record fields) on the way in — this is a
+        // resume, not a user turn, so the counter is put back straight after.
+        const resumed = this.continueRun(runId, { text: AUTO_RESUME_PROMPT }, true);
+        if (!resumed.ok) {
+            // Refusals happen before `continueRun` retires anything, so the deadline is still on the
+            // record — and a deadline in the past is a promise the cockpit keeps displaying and the
+            // engine will never keep. Retire it here instead, and say why.
+            this.clearAutoResume(runId);
+            this.store.appendEvent(runId, {
+                type: 'note',
+                message: `automatic resume could not start — ${resumed.error ?? 'unknown'}`,
+            });
+            return;
+        }
+        this.store.updateRun(runId, { autoResumeAttempts: attempts });
+        this.store.appendEvent(runId, {
+            type: 'lifecycle',
+            message: `usage limit reset — resuming automatically (${attempts}/${MAX_AUTO_RESUMES})`,
+        });
+        // A deferred continuation only ENQUEUES itself; the queue moves when something pumps it, and
+        // `recover()` — the other deferring caller — pumps once after its whole bulk sweep. A timer
+        // firing on its own has no such follow-up, so without this the resumed run sits at `queued`
+        // until some unrelated run happens to finish. This is the pump for it.
+        void this.pump();
+    }
+    /**
+     * Make the armed timers agree with the records and the current setting. Runs on every `pump()`
+     * — which is where a settings change lands (a config PUT refreshes the shared semaphore, which
+     * pumps every manager) — and once from `recover()`.
+     *
+     * It is a RECONCILE rather than a one-shot restore because the deadline is durable state and
+     * the timer is not: a restart, a rebuilt project context, a manager disposed mid-wait, or a
+     * refusal all leave a record promising a resume that no timer is holding. Rebuilding from the
+     * record covers every one of those at once — the alternative is a hint counting down to a time
+     * that has already passed, which is exactly the failure this method exists to make impossible.
+     *
+     * Cheap: an in-memory scan, and arming is skipped for every run already held.
+     */
+    reconcileAutoResumes() {
+        if (!this.semaphore.autoResumeOnUsageLimit()) {
+            // Sweep the RECORDS, not the timer map. A record promising a resume that no timer is
+            // holding is the exact population this method exists for, and it is also the one the
+            // setting can be switched off in front of: cezar restarted while it was off, the config
+            // was hand-edited, or the project context was disposed mid-wait. Retiring only the armed
+            // timers leaves such a record with a live `autoResumeAt`, which `accountHolds()` reads as
+            // a deadline hold — so nothing new starts on that account, `rescueStalledQueue` treats the
+            // phantom appointment as a legitimate reason to sit still, and the cockpit shows a
+            // `scheduled` row for a resume that will never come. `clearAutoResume` covers the armed
+            // ones too, so this one loop is the whole cancellation.
+            const pending = new Set([
+                ...this.autoResumeTimers.keys(),
+                ...this.store.listRuns().filter((run) => run.autoResumeAt !== undefined).map((run) => run.id),
+            ]);
+            for (const runId of pending) {
+                this.clearAutoResume(runId);
+                this.store.appendEvent(runId, {
+                    type: 'note',
+                    message: 'automatic resume cancelled — auto-resume is switched off',
+                });
+            }
+            return;
+        }
+        for (const run of this.store.listRuns()) {
+            if (run.status !== 'failed' || !run.autoResumeAt)
+                continue;
+            if (this.autoResumeTimers.has(run.id))
+                continue;
+            const deadline = Date.parse(run.autoResumeAt);
+            // A deadline that is unreadable, belongs to a run that has spent its cap, or belongs to a
+            // task the user has archived is retired rather than re-armed: it can only mislead. One
+            // that has just passed arms at zero — the window is open, which is the point.
+            if (run.archived
+                || !Number.isFinite(deadline)
+                || (run.autoResumeAttempts ?? 0) >= MAX_AUTO_RESUMES) {
+                this.store.updateRun(run.id, { autoResumeAt: undefined });
+                continue;
+            }
+            // …and one missed by more than a day is retired loudly: reviving a task from another era
+            // is a surprise, not a service, and this is what keeps a sweep from resurrecting every
+            // limit-stopped task a user has long since walked away from.
+            if (Date.now() - deadline > AUTO_RESUME_MISSED_WINDOW_MS) {
+                this.store.updateRun(run.id, { autoResumeAt: undefined });
+                this.store.appendEvent(run.id, {
+                    type: 'note',
+                    message: 'automatic resume expired — its window reopened over a day ago; continue this task manually',
+                });
+                continue;
+            }
+            this.armAutoResume(run.id, deadline);
+        }
+    }
+    /**
+     * Hand a run that has not spawned anything back to the queue, when the account it would run on
+     * went into a usage-limit hold (spec 2026-08-03-auto-resume-after-usage-limit).
+     *
+     * The dequeue-time gate in `pump()` cannot be the only one: a run can sit between dequeue and
+     * spawn for a long time — an in-place run waiting for the exclusive repo-root lease is the
+     * measured case — and the account can close in that gap. This is the last honest moment to
+     * refuse, because everything after it costs a real agent turn.
+     *
+     * "Untouched" is the contract: the run has created no session and no worktree, so it goes back
+     * as plain `queued` with its `startedAt` cleared, and `pump()` will pick it up when the window
+     * reopens. Returns true when the caller must abandon the run.
+     */
+    requeueWhileHeld(runId, workflow, input, runner, state) {
+        const run = this.store.getRun(runId);
+        if (!run || run.status === 'cancelled' || state?.cancelled)
+            return false;
+        // The watchdog sent this one through. Checked, never consumed: the spawn path asks this
+        // question TWICE — here at the top of `execute`, and again after the exclusive repo-root
+        // lease is granted — so a one-shot flag would clear at the first gate and let the second one
+        // hand an in-place run straight back, re-wedging the queue the rescue had just freed.
+        // `dropActive` retires the entry on every terminal path, so the set still cleans itself up.
+        if (this.forceStarted.has(runId))
+            return false;
+        if (!accountHeldFor({ ...run, runner }, this.semaphore.accountHolds(), runner))
+            return false;
+        state?.releaseRepoRoot?.();
+        if (state)
+            state.releaseRepoRoot = undefined;
+        this.pendingJobs.set(runId, { workflow, input });
+        this.queue.push(runId);
+        this.store.updateRun(runId, { status: 'queued', startedAt: undefined, currentStepId: undefined });
+        this.store.appendEvent(runId, {
+            type: 'note',
+            message: 'held in the queue — this agent account is waiting out a usage limit',
+        });
+        this.dropActive(runId);
+        return true;
+    }
+    /**
+     * The failsafe: a queue must never be able to wedge.
+     *
+     * Everything else in this file makes an idle queue CORRECT under some condition — a slot cap, a
+     * repo-root lease, and now a usage-limit hold. That is also what makes a wedged queue look
+     * correct, and the hold has already produced one in the field: two resumes fired together, each
+     * holding the account the other was waiting on, and the whole workspace stopped with every task
+     * `queued`. That specific bug is fixed and tested, but "the queue stopped and nothing will ever
+     * restart it" is too expensive a failure mode to leave resting on any single fix being right.
+     *
+     * The test is deliberately about JUSTIFICATION rather than about any particular bug: idling is
+     * legitimate while work is running (here or in another project), or while a real appointment is
+     * still ahead — a scheduled resume that will fire and pump on its own. Anything else is a
+     * queue with work in it, nothing running anywhere, and no event coming to wake it. That gets one
+     * forced sweep, which starts work under the ordinary caps and lets the account's real state
+     * re-assert itself: if the window truly is shut, that task meets the limit and re-establishes an
+     * honest hold, with a real deadline behind it this time.
+     *
+     * Public so a test can drive the wedge directly instead of waiting out the interval.
+     */
+    async rescueStalledQueue(now = Date.now()) {
+        // First, the worst shape: a record that says `queued` while the engine holds no job, no
+        // continuation and no queue entry for it. `pump()` cannot see such a run — it iterates the
+        // queue, and this one is not in it — so nothing will ever start it. Re-adopt it through the
+        // same path boot recovery uses.
+        for (const run of this.store.listRuns()) {
+            if (run.status !== 'queued')
+                continue;
+            if (this.active.has(run.id) || this.starting.has(run.id))
+                continue;
+            if (this.pendingJobs.has(run.id) || this.pendingContinuations.has(run.id))
+                continue;
+            if (this.queue.includes(run.id))
+                continue;
+            console.warn(`[cez] queue watchdog: re-adopting queued run ${run.id} the engine had lost`);
+            await this.reviveQueuedRun(run, 'queue watchdog');
+        }
+        if (this.queue.length === 0)
+            return;
+        if (this.busySlots() > 0 || this.starting.size > 0)
+            return;
+        if (this.semaphore.busy() > 0)
+            return;
+        // A future deadline is a real reason to sit still: that timer will fire and pump.
+        for (const run of this.store.listRuns()) {
+            if (run.status !== 'failed' || !run.autoResumeAt)
+                continue;
+            const deadline = Date.parse(run.autoResumeAt);
+            if (Number.isFinite(deadline) && deadline > now)
+                return;
+        }
+        if (this.semaphore.accountHolds().inFlight.size === 0) {
+            // Not the hold, then — some other wakeup went missing. An ordinary pump is the whole fix,
+            // and it is idempotent, so this stays quiet.
+            void this.pump();
+            return;
+        }
+        console.warn('[cez] queue watchdog: work is queued, nothing is running, and the usage-limit hold has no'
+            + ' deadline behind it — starting the next task anyway');
+        this.forceNextPump = true;
+        void this.pump();
+    }
+    /**
+     * The accounts this project is currently holding: one key per run parked on a usage-limit
+     * resume that has not come due yet (spec 2026-08-03-auto-resume-after-usage-limit).
+     *
+     * Published to the shared semaphore so the hold spans PROJECTS — one Claude account can be
+     * driving tasks in three repos, and a limit closes it for all of them. Derived from the
+     * records on every ask rather than tracked as state: a deadline that passes, a resume that
+     * fires, a cancel, an archive and a delete all lift the hold with no bookkeeping.
+     *
+     * Deliberately excludes a deadline that has already passed — that run is about to resume, and
+     * holding the queue for it would only stall the very work the window reopened for.
+     */
+    accountHolds(now = Date.now()) {
+        const deadline = new Set();
+        const inFlight = new Set();
+        for (const run of this.store.listRuns()) {
+            // A holding run always carries the runner it actually ran on, so the fallback is unused
+            // here — it is spelled out rather than `!` so a future record shape degrades, not throws.
+            const key = () => runAccountKey(run, run.runner ?? 'claude');
+            if (run.status === 'failed' && run.autoResumeAt) {
+                const at = Date.parse(run.autoResumeAt);
+                if (Number.isFinite(at) && at > now)
+                    deadline.add(key());
+            }
+            else if (resumeInFlight(run)) {
+                inFlight.add(key());
+            }
+        }
+        return { deadline, inFlight };
+    }
+    /**
+     * The PER-TASK off switch (`DELETE /api/v1/runs/:id/auto-resume`, and the archive route):
+     * stop resuming THIS task, without touching the workspace setting or any other task.
+     *
+     * Idempotent — a run with nothing pending answers the same way, because "this task will not
+     * resume itself" is equally true either way. Returns false only when the run does not exist,
+     * which is the route's 404.
+     */
+    cancelAutoResume(runId) {
+        const run = this.store.getRun(runId);
+        if (!run)
+            return false;
+        const pending = run.autoResumeAt !== undefined || this.autoResumeTimers.has(runId);
+        this.clearAutoResume(runId);
+        if (pending) {
+            this.store.appendEvent(runId, {
+                type: 'note',
+                message: 'automatic resume cancelled for this task',
+            });
+            // This run may have been the last thing holding its account's queue — nothing else will
+            // notice, since the hold is derived and its release is not an event.
+            void this.pump();
+        }
+        return true;
+    }
+    /** Retire a pending resume — timer, deadline and counter. The counter goes too because every
+     *  caller is a fresh epoch: a human Continue, or a resume that re-stamps its own count. */
+    clearAutoResume(runId) {
+        const timer = this.autoResumeTimers.get(runId);
+        if (timer)
+            clearTimeout(timer);
+        this.autoResumeTimers.delete(runId);
+        const run = this.store.getRun(runId);
+        if (!run)
+            return;
+        if (run.autoResumeAt !== undefined || run.autoResumeAttempts !== undefined) {
+            this.store.updateRun(runId, { autoResumeAt: undefined, autoResumeAttempts: undefined });
+        }
+    }
+    /** Reclaim finished worktrees beyond the keep-limit (#483) — directory only,
+     *  `cez/<id8>` branch kept. Best-effort; a failure never affects run
+     *  lifecycle. `review`/live runs are excluded by the selector. */
+    async enforceRetention() {
+        try {
+            const keep = await resolveWorktreeRetention(this.repoRoot);
+            await reclaimWorktrees(this.repoRoot, this.store, keep);
+        }
+        catch {
+            // retention is best-effort; swallow so terminal transitions never break.
+        }
+    }
+    /** Last live-refresh namer inputs per run — unchanged inputs skip the call. */
+    lastNamerKey = new Map();
+    /**
+     * Acquire the one-at-a-time lease for runs executing in `repoRoot`.
+     *
+     * A lease waiter is idle, so it parks in `waiting` and gives its
+     * `maxParallel` slot back (the #347 rule): isolated worktrees keep using
+     * every configured slot while root runs line up. The store status stays
+     * `running` — only the queue's busy count changes, so the GUI never shows a
+     * lease-blocked run as awaiting user input.
+     *
+     * The lease is held for the run's whole lifetime, including the idle
+     * `waiting` parks between agent turns. A parked session is still live and
+     * writes to the working tree the moment it resumes, so handing the tree to
+     * another run there would reintroduce the concurrent-edit bug (#438) this
+     * lease exists to prevent.
+     *
+     * Returns false when the run was cancelled while waiting: the lease was
+     * never granted and the caller must not touch the working tree.
+     */
+    async acquireRepoRoot(runId, state) {
+        // `cancel()` can land between the run going `running` and reaching here,
+        // while `interrupt` is still the default no-op — never enter the chain.
+        if (state.cancelled)
+            return false;
+        const previous = this.repoRootTail;
+        let release = () => undefined;
+        this.repoRootTail = new Promise((resolve) => {
+            release = resolve;
+        });
+        // Until `previous` resolves this run does not own the tree yet, so a drop
+        // during the wait must not hand the tree to the next waiter — chain our
+        // release behind `previous` instead of resolving the tail early.
+        state.releaseRepoRoot = () => {
+            void previous.then(release);
+        };
+        let abort = () => undefined;
+        const cancelled = new Promise((resolve) => {
+            abort = resolve;
+        });
+        const parked = state.interrupt;
+        state.interrupt = () => {
+            parked();
+            abort();
+        };
+        this.waiting.add(runId);
+        this.releaseSlot();
+        try {
+            await Promise.race([previous, cancelled]);
+        }
+        finally {
+            state.interrupt = parked;
+            this.waiting.delete(runId);
+        }
+        if (state.cancelled)
+            return false;
+        state.releaseRepoRoot = release;
+        return true;
+    }
+    cancel(runId) {
+        // Still waiting in the queue: just drop it there.
+        const queuedAt = this.queue.indexOf(runId);
+        if (queuedAt >= 0) {
+            this.queue.splice(queuedAt, 1);
+            this.pendingJobs.delete(runId);
+            this.pendingContinuations.delete(runId);
+            this.store.updateRun(runId, { status: 'cancelled', finishedAt: new Date().toISOString() });
+            this.store.appendEvent(runId, { type: 'lifecycle', message: 'cancelled while queued' });
+            return true;
+        }
+        const state = this.active.get(runId);
+        if (!state)
+            return false;
+        state.cancelled = true;
+        this.clearIdleTimer(state);
+        state.interrupt();
+        return true;
+    }
+    isActive(runId) {
+        return this.active.has(runId) || this.starting.has(runId) || this.queue.includes(runId);
+    }
+    /**
+     * Fold a queued run's persisted prompt — `run.task` plus everything stacked
+     * onto it (#472) — into the job input that is about to execute.
+     *
+     * Called from `pump()` immediately before `execute()`, which makes the RECORD
+     * the single source of truth for a queued run's prompt. Before this, the
+     * executing copy lived in `pendingJobs` (memory) while the record held a
+     * second one, so an edit that PATCHed the record silently did nothing until a
+     * restart. `recover()` rebuilds through the same helper, so both paths agree.
+     *
+     * **Read-only, and that is load-bearing.** It composes into the in-memory
+     * `input` and never writes the folded string back to `RunRecord.task`; the
+     * task and its stack stay separate on disk for the life of the run. Writing
+     * back would re-append the whole stack on every recovery and compound without
+     * bound — asserted directly by a test.
+     */
+    hydrateQueuedInput(runId, input) {
+        const run = this.store.getRun(runId);
+        if (!run)
+            return input;
+        const stack = run.queuedMessages ?? [];
+        const task = [run.task, ...stack.map((m) => m.text)]
+            .map((part) => part.trim())
+            .filter((part) => part.length > 0)
+            .join('\n\n');
+        // Keep the original in-memory blocks for a live process (including the
+        // best-effort case where persistence failed). Recovery has no such copy,
+        // so rebuild it from the durable task-image URLs.
+        const images = input.images?.length
+            ? input.images
+            : this.readPersistedImages(runId, run.taskImages ?? [], 'task').blocks;
+        const stackedImages = this.readPersistedImages(runId, stack.flatMap((m) => m.images ?? []), 'queued').blocks;
+        return {
+            ...input,
+            task,
+            ...(images.length ? { images } : { images: undefined }),
+            ...(stackedImages.length ? { stackedImages } : { stackedImages: undefined }),
+        };
+    }
+    /** Apply edits and messages made while a restart continuation waits for
+     * capacity. The durable record remains the source of truth, just as it is for
+     * an ordinary queued workflow (#472), so a second restart reconstructs and
+     * hydrates the same amendments instead of dropping them. */
+    hydrateQueuedContinuation(runId, continuation) {
+        const run = this.store.getRun(runId);
+        if (!run) {
+            return { ...continuation, persistedImages: [], persistedAttachments: [] };
+        }
+        const stack = run.queuedMessages ?? [];
+        const amendedTask = [run.task, ...stack.map((message) => message.text)]
+            .map((part) => part.trim())
+            .filter((part) => part.length > 0)
+            .join('\n\n');
+        const prompt = amendedTask
+            ? `${continuation.prompt}\n\nCurrent task and queued updates:\n\n${amendedTask}`
+            : continuation.prompt;
+        const persisted = this.readPersistedImages(runId, stack.flatMap((message) => message.images ?? []), 'queued');
+        return {
+            ...continuation,
+            prompt,
+            persistedImages: persisted.blocks,
+            persistedAttachments: persisted.attachments,
+        };
+    }
+    readPersistedImages(runId, urls, kind) {
+        const blocks = [];
+        const attachments = [];
+        for (const url of urls) {
+            const name = url.split('/').pop();
+            if (!name || name.includes('..') || name.includes('/') || name.includes('\\'))
+                continue;
+            const path = join(this.dataDir, 'runs', `${runId}-images`, name);
+            try {
+                const data = readFileSync(path);
+                blocks.push({
+                    type: 'image',
+                    source: { type: 'base64', media_type: mediaTypeFor(name), data: data.toString('base64') },
+                });
+                attachments.push({ name, url, path });
+            }
+            catch {
+                // Degrade, never fail the boot (AGENTS.md): the user deleted `.ai/cezar/`
+                // or the file is unreadable — start with the text and say which image went.
+                this.store.appendEvent(runId, {
+                    type: 'note',
+                    message: `${kind} attachment ${name} could not be read — starting without it`,
+                });
+            }
+        }
+        return { blocks, attachments };
+    }
+    /**
+     * Still waiting for a slot? Checked against the engine's own queue rather than
+     * the record's `status` (#472): the record is written by `execute()` a tick
+     * after `pump()` dequeues, so a status read can see `queued` for a run that has
+     * already started. The pending maps are deleted synchronously at dequeue, so
+     * they are the authoritative answer for "can this prompt still be amended".
+     */
+    isQueued(runId) {
+        return this.pendingJobs.has(runId) || this.pendingContinuations.has(runId);
+    }
+    /** Split `ContentBlock[]` into the persisted shape a stacked message holds. */
+    toQueuedMessage(runId, content) {
+        const text = content
+            .filter((b) => b.type === 'text')
+            .map((b) => b.text)
+            .join('\n');
+        const images = content
+            .filter((b) => b.type === 'image')
+            .map((b) => this.persistImage(runId, b.source.media_type, b.source.data, 'pasted'))
+            .filter((saved) => saved !== null)
+            .map((saved) => saved.url);
+        return {
+            id: randomUUID(),
+            text,
+            ...(images.length ? { images } : {}),
+            createdAt: new Date().toISOString(),
+        };
+    }
+    /**
+     * Append a prompt message onto a still-queued run (#472). Returns the stored
+     * entry, or null when the run has already started — the caller then falls
+     * through to `deferMessage`.
+     */
+    enqueueMessage(runId, content) {
+        if (!this.isQueued(runId))
+            return null;
+        const run = this.store.getRun(runId);
+        if (!run)
+            return null;
+        const message = this.toQueuedMessage(runId, content);
+        this.store.updateRun(runId, { queuedMessages: [...(run.queuedMessages ?? []), message] });
+        return message;
+    }
+    /** Edit a stacked message in place. Omitted fields retain their current value. */
+    editQueuedMessage(runId, msgId, edit) {
+        if (!this.isQueued(runId))
+            return null;
+        const run = this.store.getRun(runId);
+        const stack = run?.queuedMessages;
+        if (!stack)
+            return null;
+        const at = stack.findIndex((m) => m.id === msgId);
+        if (at < 0)
+            return null;
+        const current = stack[at];
+        const replacementImages = edit.images === undefined
+            ? current.images
+            : this.toQueuedMessage(runId, edit.images).images;
+        const replacement = {
+            id: msgId,
+            text: edit.text ?? current.text,
+            ...(replacementImages?.length ? { images: replacementImages } : {}),
+            createdAt: current.createdAt,
+        };
+        const next = [...stack];
+        next[at] = replacement;
+        this.store.updateRun(runId, { queuedMessages: next });
+        // Images the edit dropped are now orphans.
+        this.dropOrphanImages(runId, stack[at].images ?? [], next);
+        return replacement;
+    }
+    /** Remove a stacked message and its now-orphaned attachments. */
+    removeQueuedMessage(runId, msgId) {
+        if (!this.isQueued(runId))
+            return false;
+        const run = this.store.getRun(runId);
+        const stack = run?.queuedMessages;
+        if (!stack)
+            return false;
+        const target = stack.find((m) => m.id === msgId);
+        if (!target)
+            return false;
+        const next = stack.filter((m) => m.id !== msgId);
+        this.store.updateRun(runId, { queuedMessages: next });
+        this.dropOrphanImages(runId, target.images ?? [], next);
+        return true;
+    }
+    /**
+     * Delete image files no longer referenced by anything (#472). Best effort — a
+     * leftover file is harmless and goes with the run. Never touches a URL still
+     * referenced by another stacked entry or by the initial prompt's `taskImages`.
+     */
+    dropOrphanImages(runId, candidates, stack) {
+        if (!candidates.length)
+            return;
+        const run = this.store.getRun(runId);
+        const referenced = new Set([
+            ...(run?.taskImages ?? []),
+            ...stack.flatMap((m) => m.images ?? []),
+        ]);
+        for (const url of candidates) {
+            if (referenced.has(url))
+                continue;
+            const name = url.split('/').pop();
+            // Defend the join against a crafted URL: only a bare file name may be deleted.
+            if (!name || name.includes('..') || name.includes('/') || name.includes('\\'))
+                continue;
+            try {
+                rmSync(join(this.dataDir, 'runs', `${runId}-images`, name), { force: true });
+            }
+            catch {
+                /* best effort */
+            }
+        }
+    }
+    /**
+     * Edit the initial prompt of a still-queued run (#472). Re-derives the
+     * heuristic title and the PR/issue chips, but never re-runs the LLM namer —
+     * it already fired at creation and a second model call per edit is unjustified.
+     */
+    editTask(runId, task) {
+        if (!this.isQueued(runId))
+            return false;
+        const run = this.store.getRun(runId);
+        if (!run)
+            return false;
+        const workflow = this.pendingJobs.get(runId)?.workflow;
+        const skillHint = workflow?.steps.find((s) => stepKind(s) === 'agent' && s.skill)?.skill?.trim();
+        const refs = refineTaskRefs(extractTaskRefs(task), skillHint);
+        // Hand-edited titles always win (#389): `user` beats the heuristic, and a
+        // `marker` title the agent declared beats it too.
+        const keepTitle = run.titleOrigin === 'user' || run.titleOrigin === 'marker';
+        this.store.updateRun(runId, {
+            task,
+            ...(keepTitle || !workflow ? {} : { title: makeRunTitle(task, workflow) }),
+            ...(refs.prNumber !== undefined ? { prNumber: refs.prNumber } : {}),
+            ...(refs.issueNumber !== undefined ? { issueNumber: refs.issueNumber } : {}),
+        });
+        return true;
+    }
+    /**
+     * Buffer a message that arrived in the gap between dequeue and session-open
+     * (#472). `pump()` has already folded the stack and `execute()` is spawning the
+     * backend, so there is nothing left to amend and no session to deliver into —
+     * without this rung the message would 409, a genuinely dropped message in the
+     * feature built to stop dropping them. Flushed as an ordinary follow-up turn
+     * the instant the session opens; dropped if the run never starts, which the
+     * existing error path already surfaces.
+     *
+     * The buffer lives on the manager rather than the `ActiveRun` because the
+     * `ActiveRun` does not exist yet for part of this window.
+     */
+    deferMessage(runId, content) {
+        // The window spans two sub-states: `starting` (no `ActiveRun` yet) and the
+        // longer stretch where the `ActiveRun` exists but the backend is still being
+        // spawned. `execute()` deletes the run from `starting` as soon as it builds
+        // the state — seconds before the session opens — so checking `starting`
+        // alone would reopen exactly the drop this rung exists to close.
+        const state = this.active.get(runId);
+        const startingUp = this.starting.has(runId) || (state !== undefined && !state.sessionEverOpened && !state.cancelled);
+        if (!startingUp)
+            return false;
+        const pending = this.deferredMessages.get(runId) ?? [];
+        pending.push(content);
+        this.deferredMessages.set(runId, pending);
+        return true;
+    }
+    /** Deliver anything `deferMessage` buffered, once the session is live. */
+    flushDeferred(runId) {
+        const pending = this.deferredMessages.get(runId);
+        if (!pending?.length)
+            return;
+        // Re-buffer whatever the session refused rather than dropping it. `sendMessage`
+        // answers false when the session is not open yet — and silently losing a message
+        // here would be precisely the failure `deferMessage` exists to prevent. Anything
+        // left over is retried by the next session that opens on this run.
+        const unsent = pending.filter((content) => !this.sendMessage(runId, content));
+        if (unsent.length)
+            this.deferredMessages.set(runId, unsent);
+        else
+            this.deferredMessages.delete(runId);
+    }
+    /**
+     * Deliver a user message into the run's live claude session (mid-turn or
+     * while `waiting`). Returns false when there is no open session — the GUI
+     * then offers "Continue" instead.
+     */
+    sendMessage(runId, content) {
+        const delivered = this.deliverMessage(runId, content, true);
+        if (delivered) {
+            const state = this.active.get(runId);
+            if (state)
+                state.monitoringWakeups = 0;
+            this.store.updateRun(runId, { monitoringWakeCapReached: undefined });
+        }
+        return delivered;
+    }
+    /** Shared live-session delivery. Synthetic scheduler prompts reuse lifecycle
+     * bookkeeping without masquerading as user-authored transcript messages. */
+    deliverMessage(runId, content, userAuthored) {
+        const state = this.active.get(runId);
+        if (!state?.session?.open || state.cancelled)
+            return false;
+        const text = content
+            .filter((b) => b.type === 'text')
+            .map((b) => b.text)
+            .join('\n');
+        // Persist the attached images so the thread can render them (not just count them) — the same
+        // on-disk store + `/images/` route the agent's own screenshots use. `pasted` prefix marks
+        // these as user attachments (vs. agent tool screenshots) on disk (#357).
+        const persisted = userAuthored ? content
+            .filter((b) => b.type === 'image')
+            .map((b) => this.persistImage(runId, b.source.media_type, b.source.data, 'pasted'))
+            .filter((saved) => saved !== null) : [];
+        const images = persisted.map((saved) => saved.url);
+        if (userAuthored) {
+            this.store.appendEvent(runId, {
+                type: 'user-message',
+                stepId: state.currentStepId,
+                text,
+                imageCount: content.filter((b) => b.type === 'image').length,
+                images,
+            });
+        }
+        // Tell the agent where the pasted files live on disk (#357): the base64 blocks below still
+        // ride along so the model can *view* them, but a real path is what lets it *operate* on them
+        // (save, `cp`, attach to a GitHub issue/PR) — and it's the only usable reference on backends
+        // (codex, opencode) that drop image blocks entirely before reaching the model.
+        const expanded = userAuthored ? expandRegistrySlashSkill(content, state.skills ?? []) : content;
+        const deliverable = persisted.length ? [...expanded, pastedAttachmentsNote(persisted)] : expanded;
+        const delivered = state.session.sendMessage(deliverable);
+        if (delivered) {
+            this.clearIdleTimer(state);
+            this.clearMonitoringWakeTimer(state, runId);
+            this.waiting.delete(runId); // resumed — the run counts against slots again
+            this.monitoring.delete(runId);
+            // Clear any `monitoring` activity — the agent is actively working again
+            // (spec 2026-07-18-subagent-monitoring-status, #490).
+            this.store.updateRun(runId, { status: 'running', activity: undefined });
+            if (state.currentStepId) {
+                this.store.updateStep(runId, state.currentStepId, { status: 'running' });
+            }
+        }
+        return delivered;
+    }
+    /** Close the open session gracefully — the run then completes as `done`
+     *  (or rests at `review` when the worktree holds changes, spec 009).
+     *  On a run already resting at `review` (no session — the engine loop is
+     *  over), "Finish" is the third review exit: accept the changes without a
+     *  PR and flip straight to `done`. */
+    finish(runId) {
+        const state = this.active.get(runId);
+        if (state?.session?.open) {
+            this.clearIdleTimer(state);
+            this.store.appendEvent(runId, { type: 'lifecycle', message: 'session closed by user' });
+            state.session.end();
+            return true;
+        }
+        const run = this.store.getRun(runId);
+        if (run?.status === 'review' && !this.isActive(runId)) {
+            this.store.updateRun(runId, { status: 'done' });
+            this.store.appendEvent(runId, { type: 'lifecycle', message: 'review accepted — finished without a PR' });
+            return true;
+        }
+        return false;
+    }
+    /**
+     * "Continue" (spec 003): reopen a finished run's claude session in-process
+     * (`claude --resume <sessionId>`) as a new synthetic step. The session then
+     * behaves exactly like an interactive step: `waiting` after each turn,
+     * messages via sendMessage, closed by finish/idle/cancel.
+     */
+    continueRun(runId, opts = {}, 
+    /** Restart recovery may discover several interrupted tasks at once. Those
+     *  continuations are queued; an explicit user Continue remains immediate. */
+    deferForCapacity = false) {
+        if (agentModelsLocked(this.repoRoot) && opts.model?.trim()) {
+            return { ok: false, error: AGENT_MODELS_LOCKED_ERROR };
+        }
+        if (this.active.has(runId))
+            return { ok: false, error: 'run is still active' };
+        const run = this.store.getRun(runId);
+        if (!run)
+            return { ok: false, error: 'not found' };
+        // `review` is continuable too — that's the "Send back" path (spec 009).
+        if (!['done', 'failed', 'cancelled', 'review'].includes(run.status)) {
+            return { ok: false, error: `cannot continue a ${run.status} run` };
+        }
+        const sessionStep = [...run.steps].reverse().find((s) => s.sessionId);
+        if (!sessionStep?.sessionId)
+            return { ok: false, error: 'no agent session to resume' };
+        const targetRunner = opts.runner ?? run.runner ?? 'claude';
+        // Session ids are provider-owned opaque values. New records carry explicit
+        // affinity; for legacy records, the run's current runner is the conservative
+        // owner until a continuation emits a new, attributed session id (#562).
+        const sessionBackend = sessionStep.backend ?? run.runner ?? 'claude';
+        // A session id only resolves inside the config dir that created it (spec
+        // 2026-07-29-agent-profiles), so switching ACCOUNT ends the session exactly like switching
+        // backend does: `claude --resume <id>` under another login finds nothing and would silently
+        // open a fresh conversation while the thread claimed it had resumed. A step that recorded no
+        // account predates the feature and therefore ran under the discovered one.
+        const sessionAccount = sessionStep.profileId ?? DEFAULT_AGENT_ACCOUNT_ID;
+        const accountSwitched = opts.agentProfile !== undefined && opts.agentProfile !== sessionAccount;
+        const resume = sessionBackend === targetRunner && !accountSwitched;
+        // Follow-up runner/model/account override (#401, spec 2026-07-29-agent-profiles): the composer
+        // lets the user pick which backend, model and login handle this continuation — the same flat
+        // pill the /new composer offers. Omitted → the run's current backend/model/account is kept
+        // (backward compat). A provided choice is persisted BEFORE scheduling, so it becomes the
+        // run's current backend — `runContinuation` reads it off the record, later continuations
+        // default to it, and the header reflects the active engine. An empty model ('') clears the
+        // pin, letting the runner pick the model (auto).
+        if (opts.runner !== undefined || opts.model !== undefined || opts.agentProfile !== undefined) {
+            // Guard the pairing before persisting anything: the model override applies to the runner
+            // this continuation will actually use (`opts.runner ?? record.runner ?? 'claude'` — the
+            // same resolution `runContinuation` reads off the record). A model that is recognizably
+            // another runner's preset would corrupt the run; free-form/custom ids pass untouched.
+            if (opts.model && modelConflictsWithRunner(opts.model, targetRunner)) {
+                return { ok: false, error: `model '${opts.model}' is not a ${targetRunner} model` };
+            }
+            // A runner switch that carries NO explicit model must not leave the previous backend's pin
+            // on the record: the guard above only sees `opts.model`, so without this an inherited
+            // `opus` would survive a switch to codex and `runContinuation` would hand it to the codex
+            // runner. Clearing (not rejecting) is right — the pin belonged to the old backend and is
+            // meaningless for the new one, which is exactly what the composer already displays (auto).
+            // Only a recognizably foreign preset is cleared; a free-form/custom id is left alone.
+            const inheritedPinIsForeign = opts.model === undefined &&
+                run.model !== undefined &&
+                modelConflictsWithRunner(run.model, targetRunner);
+            // An account belongs to ONE agent, so a runner switch that names no account must not leave
+            // the previous backend's login on the record. It is inert immediately (resolution applies
+            // the run's account only to steps on the run's own runner) and wrong later, when a further
+            // continuation switches back and inherits a login the user picked for a different task.
+            const inheritedAccountIsForeign = opts.agentProfile === undefined &&
+                run.agentProfile !== undefined &&
+                targetRunner !== (run.runner ?? 'claude');
+            this.store.updateRun(runId, {
+                ...(opts.runner !== undefined ? { runner: opts.runner } : {}),
+                ...(opts.model !== undefined
+                    ? { model: opts.model === '' ? undefined : opts.model }
+                    : inheritedPinIsForeign
+                        ? { model: undefined }
+                        : {}),
+                // Persisted BEFORE scheduling, like the runner/model pair: `runContinuation` resolves the
+                // account off the record, and every later continuation then defaults to it.
+                ...(opts.agentProfile !== undefined
+                    ? { agentProfile: opts.agentProfile }
+                    : inheritedAccountIsForeign
+                        ? { agentProfile: undefined }
+                        : {}),
+            });
+        }
+        // Everything that could refuse this continuation has now passed, so a pending usage-limit
+        // resume is superseded either way: this IS that resume (it re-stamps its own counter), or a
+        // human got there first — and then the counter starts over, because the cap only exists to
+        // bound UNATTENDED resumes.
+        this.clearAutoResume(runId);
+        const continuations = run.steps.filter((s) => s.id.startsWith('continue-')).length;
+        const stepId = `continue-${continuations + 1}`;
+        this.store.addStep(runId, { id: stepId, name: 'Continue', kind: 'agent' });
+        const prompt = opts.text?.trim() || 'Continue.';
+        const images = opts.images ?? [];
+        if (deferForCapacity) {
+            this.pendingContinuations.set(runId, {
+                stepId,
+                sessionId: resume ? sessionStep.sessionId : undefined,
+                backend: targetRunner,
+                prompt,
+                images,
+            });
+            this.queue.push(runId);
+            this.store.updateRun(runId, {
+                status: 'queued',
+                error: undefined,
+                finishedAt: undefined,
+                currentStepId: undefined,
+            });
+            return { ok: true };
+        }
+        void this.runContinuation(runId, stepId, resume ? sessionStep.sessionId : undefined, targetRunner, prompt, images).catch((err) => {
+            const message = err instanceof Error ? err.message : String(err);
+            this.store.updateRun(runId, {
+                status: 'failed',
+                error: `continue crashed: ${message}`,
+                finishedAt: new Date().toISOString(),
+            });
+            this.dropActive(runId);
+        });
+        return { ok: true };
+    }
+    async runContinuation(runId, stepId, sessionId, backend, prompt, 
+    /** Screenshots pasted into the follow-up composer — delivered with the
+     *  reopened session's opening message, exactly like a live-session
+     *  message's attachments. */
+    images = [], 
+    /** Queued-message screenshots were persisted when they were enqueued and
+     *  reconstructed at dequeue. Keep them separate from fresh `images` so
+     *  opening a recovered continuation does not persist duplicate files. */
+    persistedImages = [], persistedAttachments = []) {
+        // Continuation runs in the task's worktree when it still exists (spec
+        // 006) — the resumed session sees exactly what the original run left.
+        // Retention (#483) may have reclaimed this run's worktree directory while
+        // keeping its branch and worktreePath. Re-materialize it on resume and clear
+        // the stamp so the session regains its isolated tree and the run is eligible
+        // for retention again — otherwise it keeps a dir on disk while staying
+        // invisible to the enforcer forever. Best-effort; falls back to repoRoot.
+        await rematerializeReclaimedWorktree(this.repoRoot, this.store, runId);
+        const record = this.store.getRun(runId);
+        // The env is a live ceiling: a run created while the inbox was on must not keep writing
+        // follow-ups after it is switched off.
+        const generateFollowups = followupsEnabled() && record?.generateFollowups !== false;
+        const cwd = record?.worktreePath && existsSync(record.worktreePath)
+            ? record.worktreePath
+            : this.repoRoot;
+        const state = { cancelled: false, interrupt: () => undefined, cwd };
+        this.active.set(runId, state);
+        this.starting.delete(runId);
+        if (state.cwd === this.repoRoot) {
+            if (repositoryRootLockDisabled()) {
+                this.store.appendEvent(runId, {
+                    type: 'note',
+                    message: REPOSITORY_ROOT_LOCK_DISABLED_NOTE,
+                });
+            }
+            else {
+                this.store.appendEvent(runId, {
+                    type: 'note',
+                    message: 'waiting for exclusive access to the repository working tree',
+                });
+                if (!(await this.acquireRepoRoot(runId, state))) {
+                    this.store.updateRun(runId, {
+                        status: 'cancelled',
+                        finishedAt: new Date().toISOString(),
+                        currentStepId: undefined,
+                    });
+                    this.store.appendEvent(runId, { type: 'lifecycle', message: 'run cancelled' });
+                    this.dropActive(runId);
+                    return;
+                }
+            }
+        }
+        this.armAutosave(state);
+        if (record)
+            seedHandoffFile(this.dataDir, record); // idempotent — normally already there
+        // Registry snapshot for `/skill` expansion. `execute` loads this for the workflow's own
+        // sessions; a continuation builds its OWN ActiveRun, and without this the resumed session
+        // expanded against an empty registry and leaked `/om-...` verbatim to the backend, which
+        // answered "Unknown skill" (#811). Best-effort — discovery must never break Continue.
+        state.skills = await discoverSkills(this.repoRoot).catch(() => []);
+        this.store.updateRun(runId, {
+            status: 'running',
+            error: undefined,
+            finishedAt: undefined,
+            currentStepId: stepId,
+            activity: undefined, // resuming a monitoring run — it's actively working again (#490)
+        });
+        this.store.updateStep(runId, stepId, {
+            status: 'running',
+            iterations: 1,
+            startedAt: new Date().toISOString(),
+            sessionId,
+            backend,
+        });
+        this.store.appendEvent(runId, { type: 'step-start', stepId, name: 'Continue', kind: 'agent', iteration: 1 });
+        // Attachments pasted into the follow-up composer, on the same terms as a live-session
+        // message (#357): persisted to the run's own image store so the thread renders the bubble's
+        // images rather than a bare count, and handed to the agent BOTH as base64 blocks (so it can
+        // view them) and as absolute paths appended to the prompt (so it can operate on them — and
+        // because codex/opencode drop image blocks before they reach the model).
+        const freshAttachments = images
+            .filter((b) => b.type === 'image')
+            .map((b) => this.persistImage(runId, b.source.media_type, b.source.data, 'pasted'))
+            .filter((saved) => saved !== null);
+        const openingImages = [...images, ...persistedImages];
+        const attachments = [...freshAttachments, ...persistedAttachments];
+        this.store.appendEvent(runId, {
+            type: 'user-message',
+            stepId,
+            text: prompt,
+            imageCount: openingImages.filter((b) => b.type === 'image').length,
+            ...(attachments.length ? { images: attachments.map((saved) => saved.url) } : {}),
+        });
+        let stepCost = 0;
+        let turnText = '';
+        let sessionError;
+        const sink = this.makeUiSink(runId, stepId);
+        const onEvent = (event) => {
+            if (event.type === 'image') {
+                const saved = this.persistImage(runId, event.mediaType, event.data);
+                if (saved)
+                    this.store.appendEvent(runId, { type: 'image', stepId, ...saved });
+                return;
+            }
+            if (event.type === 'text') {
+                turnText = appendTurnText(turnText, event.text);
+                const text = stripAskMarker(stripTaskMarkers(stripMonitoringMarker(stripDoneMarker(event.text))));
+                if (text)
+                    this.store.appendEvent(runId, { type: 'text', text, stepId });
+                return;
+            }
+            this.store.appendEvent(runId, { ...event, stepId });
+            if (event.type === 'error') {
+                sessionError ??= event.message;
+                state.session?.interrupt();
+                return;
+            }
+            if (sessionError)
+                return;
+            if (event.type === 'session') {
+                this.store.updateStep(runId, stepId, { sessionId: event.sessionId, backend });
+            }
+            if (event.type === 'token-usage') {
+                this.store.updateStep(runId, stepId, { tokensUsed: event.tokensUsed });
+            }
+            if (event.type === 'cost') {
+                stepCost += event.usd;
+                this.store.updateStep(runId, stepId, { costUsd: stepCost });
+            }
+            if (event.type === 'turn-end') {
+                // Belt-and-braces: v2 `turn.completed` already flushed the delta
+                // coalescers; the v1 turn boundary flushes again (idempotent) so no
+                // buffered delta can outlive its turn.
+                sink.flushAll();
+                void this.recordTurnEnd(runId, turnText); // titleSummary + diffStat (#389)
+                const sessionOpen = !state.cancelled && state.session?.open;
+                const done = sessionOpen && DONE_MARKER_RE.test(turnText.trimEnd());
+                // `CEZ:ASK` → the user is genuinely blocked; wins over `CEZ:MONITORING`
+                // (a pending question is always attention), loses to `CEZ:DONE` (#473).
+                const askResult = sessionOpen && !done ? parseAskMarkerResult(turnText) : undefined;
+                const ask = askResult?.kind === 'valid' ? askResult.request : null;
+                const askRejection = askResult ? askMarkerRejection(askResult) : undefined;
+                const monitoring = sessionOpen && !done && !ask && MONITORING_MARKER_RE.test(turnText.trimEnd());
+                turnText = '';
+                if (askRejection)
+                    this.store.appendEvent(runId, { type: 'note', message: askRejection, stepId });
+                if (done) {
+                    // Goal achieved (agent contract, #347) — same as in runAgentStep.
+                    this.store.appendEvent(runId, { type: 'lifecycle', message: 'goal achieved — session closed' });
+                    appendHandoffHeartbeat(this.dataDir, runId, 'turn complete — goal achieved, session closed');
+                    state.session?.end();
+                    return;
+                }
+                if (sessionOpen) {
+                    // Autonomous (#autonomous): never hand the ball back to the user. Nudge the agent to
+                    // keep going (bounded by MAX_AUTO_CONTINUES) instead of parking at `waiting`.
+                    const autoContinued = state.autonomous &&
+                        (state.autoContinues ?? 0) < MAX_AUTO_CONTINUES &&
+                        !state.cancelled &&
+                        (() => {
+                            const sent = state.session?.sendMessage([{ type: 'text', text: AUTONOMOUS_NUDGE }]);
+                            if (!sent)
+                                return false;
+                            state.autoContinues = (state.autoContinues ?? 0) + 1;
+                            this.store.appendEvent(runId, {
+                                type: 'note',
+                                message: `autonomous — continuing without pausing (${state.autoContinues}/${MAX_AUTO_CONTINUES})`,
+                            });
+                            return true;
+                        })();
+                    if (!autoContinued) {
+                        // `CEZ:ASK` → park `waiting` (attention) AND surface the structured
+                        // question as an ask card (#473). `CEZ:MONITORING` → non-attention
+                        // `running`/`activity:'monitoring'` (#490). Both share the waiting
+                        // lifecycle (free the slot, keep the idle timer); the autonomous
+                        // nudge above still wins over either.
+                        if (ask)
+                            emitAskRequested(sink, ask);
+                        if (monitoring) {
+                            this.store.updateRun(runId, { status: 'running', activity: 'monitoring' });
+                            this.store.updateStep(runId, stepId, { status: 'running' });
+                            this.monitoring.add(runId);
+                            this.clearIdleTimer(state);
+                            this.armMonitoringWakeTimer(runId, state);
+                        }
+                        else {
+                            this.store.updateRun(runId, { status: 'waiting', activity: undefined });
+                            this.store.updateStep(runId, stepId, { status: 'waiting' });
+                            this.monitoring.delete(runId);
+                            this.clearMonitoringWakeTimer(state, runId);
+                        }
+                        this.waiting.add(runId);
+                        if (!monitoring)
+                            this.armIdleTimer(runId, state);
+                        this.releaseSlot();
+                    }
+                }
+                // A turn that completed is the ONLY evidence the provider's window actually reopened, so
+                // it is what retires the consecutive-resume counter — which in turn releases the account
+                // hold for every other task queued behind it (spec
+                // 2026-08-03-auto-resume-after-usage-limit). `settleSuccess` does the same for a run that
+                // finishes outright; this covers the far more common "parked for the user" ending.
+                if (this.store.getRun(runId)?.autoResumeAttempts !== undefined) {
+                    this.store.updateRun(runId, { autoResumeAttempts: undefined });
+                }
+                appendHandoffHeartbeat(this.dataDir, runId, `turn complete — status=${monitoring ? 'monitoring' : sessionOpen ? 'waiting' : 'running'}`);
+            }
+        };
+        // Backend + model come off the record: the run's current backend by default, or the
+        // follow-up override that `continueRun` persisted before scheduling (#401).
+        const continueBackend = backend;
+        /** Settle this turn as a failure before anything is spawned — the shape both
+         *  pre-spawn gates below need (model identity, #405; temp directory, #785). */
+        const failBeforeSpawn = (message) => {
+            const failedAt = new Date().toISOString();
+            sink.sessionEnded('error', message);
+            this.store.updateStep(runId, stepId, {
+                status: 'failed',
+                error: message,
+                finishedAt: failedAt,
+            });
+            this.store.updateRun(runId, {
+                status: 'failed',
+                error: `continue failed: ${message}`,
+                finishedAt: failedAt,
+                currentStepId: undefined,
+            });
+            this.store.appendEvent(runId, {
+                type: 'lifecycle',
+                message: `continue failed — ${message}`,
+            });
+            this.dropActive(runId);
+        };
+        // Apply the SAME canonical-identity gate the first spawn applies (#405, review M1).
+        // A follow-up may switch both runner and model (#401), so without this the record keeps
+        // asserting the identity the run STARTED with while a different model serves the turn —
+        // the exact defect that PR existed to remove — and the raw record string reaches the CLI
+        // in the un-normalised wire form the first step already converted away (`anthropic/opus`
+        // instead of `opus`). Fail loud here too rather than let the backend pick a default.
+        let continueModel;
+        try {
+            const normalized = normalizeModelForBackend(continueBackend, agentModelsLocked(this.repoRoot) ? undefined : record?.model, { configuredProvider: await configuredModelProvider(continueBackend, state.cwd) });
+            continueModel = normalized?.backendModel;
+            this.store.updateRun(runId, {
+                modelIdentity: normalized ? formatModelIdentity(normalized.identity) : undefined,
+            });
+        }
+        catch (err) {
+            if (!(err instanceof ModelIdentityError))
+                throw err;
+            failBeforeSpawn(err.message);
+            return;
+        }
+        // Resuming reattaches to a session that lives inside ONE account's config dir, so the
+        // continuation must run under the account that created it — not whatever the project has
+        // been switched to since. The owning step is the one carrying this session id.
+        const owningStep = sessionId === undefined
+            ? undefined
+            : record?.steps.find((s) => s.sessionId === sessionId);
+        const resumedProfileId = owningStep?.profileId;
+        // The owning step also names the session's tools: resolve `allowedTools`/`bashAllowlist`
+        // from the persisted `workflowDef` exactly as the first spawn did (`runAgentStep`).
+        // Rebuilding with the bare DEFAULT_ALLOWED_TOOLS silently revoked every per-step grant
+        // (MCP servers, subagents) on Continue, restart recovery and the usage-limit auto-resume
+        // — and dropping `bashAllowlist` WIDENED Bash from an allowlist to unrestricted
+        // (`AgentRunSpec.allowedTools`, #430). Record steps share ids with `workflowDef.steps`;
+        // a synthetic `continue-N` owner and a fresh-session continuation (backend switch — no
+        // owning session) both extend the run's tail, so they resolve from the definition's last
+        // agent step. A legacy record without `workflowDef` (#367), or a session no step owns,
+        // keeps today's defaults.
+        const defSteps = record?.workflowDef?.steps;
+        const toolsStep = defSteps === undefined || (sessionId !== undefined && owningStep === undefined)
+            ? undefined
+            : defSteps.find((s) => s.id === owningStep?.id)
+                ?? [...defSteps].reverse().find((s) => stepKind(s) === 'agent');
+        // The temp-directory preflight (#785) rides along with the account resolution: a resumed
+        // turn hits the same broken `/tmp` a fresh one would, and an agent whose shell silently
+        // returns nothing is worse than a turn that refuses to start and says why.
+        let continueProfile;
+        try {
+            continueProfile = await this.agentEnvForStep(runId, continueBackend, {
+                generateFollowups,
+                recordedProfileId: resumedProfileId,
+            });
+        }
+        catch (err) {
+            if (!(err instanceof AgentTempDirError))
+                throw err;
+            failBeforeSpawn(err.message);
+            return;
+        }
+        this.store.updateStep(runId, stepId, { profileId: continueProfile.profileId });
+        const runner = createRunner(continueBackend);
+        state.currentStepId = stepId;
+        this.beginUsageInvocation(runId, state, stepId);
+        // A continuation's opening message becomes the session's `userPrompt` and never passes
+        // through `deliverMessage`, so it needs the SAME delivery-only `/skill` rewrite the
+        // live path applies (#811). Delivery-only: the `user-message` event above already
+        // persisted the user's original text, and the transcript must keep showing that.
+        const openingPrompt = expandRegistrySlashSkillText(prompt, state.skills ?? []);
+        const session = runner.startSession({
+            // The Continue step is a fresh agent session on the same run — the
+            // run's extra system prompt (already resolved at execute time and
+            // echoed on the record) rides along with the handoff contract.
+            systemPrompt: composeSystemPrompt(record?.systemPrompt, generateFollowups ? HANDOFF_INSTRUCTIONS : HANDOFF_ONLY_INSTRUCTIONS),
+            userPrompt: attachments.length
+                ? `${openingPrompt}\n\n${pastedAttachmentsText(attachments)}`
+                : openingPrompt,
+            ...(openingImages.length ? { images: openingImages } : {}),
+            cwd: state.cwd,
+            allowedTools: allowedToolsForStep(toolsStep, continueBackend),
+            bashAllowlist: toolsStep?.bashAllowlist,
+            additionalDirectories: agentDirectories(join(this.dataDir, 'runs'), continueProfile.env),
+            env: continueProfile.env,
+            model: continueModel,
+            sessionId,
+            resume: sessionId !== undefined,
+            timeoutMs: 0,
+        }, onEvent, { onUiEvent: (event) => this.handleRunnerUiEvent(runId, state, sink, event) });
+        state.session = session;
+        state.sessionEverOpened = true;
+        this.flushDeferred(runId);
+        state.interrupt = () => session.interrupt();
+        if (session.pid !== undefined)
+            registerRunProcess(runId, session.pid);
+        const finishedAt = () => new Date().toISOString();
+        try {
+            await session.result;
+            if (sessionError)
+                throw new Error(sessionError);
+            sink.sessionEnded(state.cancelled ? 'cancelled' : 'end_turn');
+            if (state.cancelled) {
+                this.store.updateStep(runId, stepId, { status: 'cancelled', finishedAt: finishedAt() });
+                this.store.updateRun(runId, { status: 'cancelled', finishedAt: finishedAt(), currentStepId: undefined });
+                this.store.appendEvent(runId, { type: 'lifecycle', message: 'run cancelled' });
+                appendHandoffHeartbeat(this.dataDir, runId, `step "${stepId}" complete — status=cancelled`);
+            }
+            else {
+                this.store.updateStep(runId, stepId, { status: 'done', finishedAt: finishedAt() });
+                this.store.appendEvent(runId, { type: 'step-end', stepId, status: 'done' });
+                await this.settleSuccess(runId);
+                appendHandoffHeartbeat(this.dataDir, runId, `step "${stepId}" complete — status=done`);
+            }
+        }
+        catch (err) {
+            const message = err instanceof Error ? err.message : String(err);
+            sink.sessionEnded('error', message);
+            this.store.updateStep(runId, stepId, { status: 'failed', error: message, finishedAt: finishedAt() });
+            appendHandoffHeartbeat(this.dataDir, runId, `step "${stepId}" complete — status=failed`);
+            this.store.updateRun(runId, {
+                status: 'failed',
+                error: `continue failed: ${message}`,
+                finishedAt: finishedAt(),
+                currentStepId: undefined,
+            });
+            this.store.appendEvent(runId, { type: 'lifecycle', message: `continue failed — ${message}` });
+        }
+        finally {
+            this.recordUsagePeaks(runId);
+            this.clearIdleTimer(state);
+            this.clearAutosaveTimer(state);
+            if (state.cwd !== this.repoRoot)
+                await autosaveCommit(state.cwd, 'turn end');
+            this.dropActive(runId);
+        }
+    }
+    // ---- execution -----------------------------------------------------------
+    async execute(runId, workflow, input) {
+        const state = {
+            cancelled: false,
+            interrupt: () => undefined,
+            cwd: this.repoRoot,
+            autonomous: input.autonomous === true,
+            autoContinues: 0,
+        };
+        this.active.set(runId, state);
+        this.starting.delete(runId);
+        const emit = (event) => this.store.appendEvent(runId, event);
+        // Resolve the agent backend for this run: the task choice (GUI) wins over
+        // the config default. Per-step `runner` can still override it below.
+        const config = await loadConfig(this.repoRoot);
+        const taskBackend = input.runner ?? config.defaultRunner;
+        // The account may have gone into a usage-limit hold since this run was dequeued — the queue
+        // gate cannot be the only one, because dequeue is not the moment of no return. Nothing has
+        // happened yet here, so the run goes back to the queue untouched (spec
+        // 2026-08-03-auto-resume-after-usage-limit).
+        if (this.requeueWhileHeld(runId, workflow, input, taskBackend))
+            return;
+        // Extra system prompt (R2 2.3): POST override > config default; echoed on
+        // the record so the UI/API can show what the run actually used.
+        const extraSystemPrompt = resolveExtraSystemPrompt(input.systemPrompt, config.systemPrompt);
+        // Canonical provider/model identity (#405) — the normalised `provider/model`
+        // the task ran with, persisted for cost attribution / reproducible replay
+        // beside the free-text `model`. Best-effort here (a per-step `runner`/`model`
+        // can still override below); the authoritative fail-loud gate is at spawn.
+        let modelIdentity;
+        try {
+            const normalized = normalizeModelForBackend(taskBackend, agentModelsLocked(this.repoRoot) ? undefined : input.model, { configuredProvider: await configuredModelProvider(taskBackend, this.repoRoot) });
+            modelIdentity = normalized ? formatModelIdentity(normalized.identity) : undefined;
+        }
+        catch {
+            // An unresolvable task-level model surfaces loudly at the step below; the
+            // metadata echo stays absent rather than guessing.
+        }
+        this.store.updateRun(runId, {
+            status: 'running',
+            startedAt: new Date().toISOString(),
+            runner: taskBackend,
+            systemPrompt: extraSystemPrompt,
+            modelIdentity,
+        });
+        emit({ type: 'lifecycle', message: `run started — workflow "${workflow.name}" (runner: ${taskBackend})` });
+        // Worktree per task (spec 006): the agent works on its own branch in
+        // `.ai/cezar/worktrees/<id>`, never in the user's working tree. A Git task
+        // that requests isolation fails closed if the worktree cannot be
+        // established; only explicit opt-out and non-Git modes run in place.
+        const repo = await getRepoInfo(this.repoRoot);
+        if (repo && input.worktree === false) {
+            // Composer opt-out: run in the repo working tree, no branch/worktree. The
+            // repository-root lease serializes these runs by default; the explicit
+            // CEZ_DISABLE_REPO_LOCK=1 escape hatch allows unsafe overlap.
+            // Pin the starting commit: the session's Changes and Commits views use it
+            // as their stable lower bound while reading the current working copy.
+            const startingCommit = await getHeadCommit(repo.root);
+            if (startingCommit)
+                this.store.updateRun(runId, { baseBranch: startingCommit });
+            emit({ type: 'note', message: 'worktree off — running in the repo working tree' });
+        }
+        else if (repo) {
+            emit({
+                type: 'note',
+                message: `worktree on — using an isolated task worktree (${input.worktree === true ? 'explicit request' : 'default'})`,
+            });
+            // Fork from the configured base branch (config.json `baseBranch`, e.g.
+            // `develop`) — also the target of the eventual draft PR. Unresolvable
+            // (typo, not fetched) → note + the currently checked-out branch.
+            //
+            // A task that already recorded a fork point keeps it: its worktree is
+            // reused as-is, and re-resolving against a since-changed config would
+            // silently re-anchor the `merge-base` every diff/shortstat is measured
+            // from, shifting "what did this task change" under an existing task.
+            const recorded = this.store.getRun(runId)?.baseBranch;
+            let base = recorded ?? repo.branch;
+            const configured = recorded ? undefined : config.baseBranch;
+            if (configured) {
+                const resolved = await resolveBaseRef(this.repoRoot, configured);
+                if (resolved) {
+                    base = resolved;
+                }
+                else {
+                    emit({
+                        type: 'note',
+                        message: `configured base branch "${configured}" not found (locally or on origin) — using "${repo.branch}"`,
+                    });
+                }
+            }
+            try {
+                const wt = await createWorktree(this.repoRoot, runId, base);
+                state.cwd = wt.path;
+                this.store.updateRun(runId, {
+                    worktreePath: wt.path,
+                    branch: wt.branch,
+                    baseBranch: wt.baseBranch,
+                });
+                emit({ type: 'note', message: `worktree ready — branch ${wt.branch} (base ${wt.baseBranch})` });
+                // Seed from this manager's project root: each multi-project context has
+                // its own manager/repoRoot and must never copy another project's layer.
+                const seededConfig = await seedAgentConfigLocalLayer(this.repoRoot, state.cwd).catch(() => []);
+                if (seededConfig.length > 0) {
+                    emit({ type: 'note', message: `seeded personal agent config: ${seededConfig.join(', ')}` });
+                }
+                this.armAutosave(state);
+            }
+            catch (err) {
+                const message = err instanceof Error ? err.message : String(err);
+                const error = `worktree creation failed: ${message}`;
+                emit({ type: 'note', message: `${error} — task stopped before workflow execution` });
+                this.store.updateRun(runId, {
+                    status: 'failed',
+                    error,
+                    finishedAt: new Date().toISOString(),
+                    currentStepId: undefined,
+                });
+                emit({ type: 'lifecycle', message: `run failed — ${error}` });
+                this.dropActive(runId);
+                return;
+            }
+        }
+        else {
+            emit({ type: 'note', message: 'not a git repository — running in place, one task at a time' });
+        }
+        if (state.cwd === this.repoRoot) {
+            if (repositoryRootLockDisabled()) {
+                emit({
+                    type: 'note',
+                    message: REPOSITORY_ROOT_LOCK_DISABLED_NOTE,
+                });
+            }
+            else {
+                emit({
+                    type: 'note',
+                    message: 'waiting for exclusive access to the repository working tree',
+                });
+                // A cancel during the wait leaves the lease ungranted; the step loop
+                // below breaks on `cancelled` before touching the tree and settles the
+                // run through the usual path.
+                await this.acquireRepoRoot(runId, state);
+            }
+            // THE window that matters for an in-place run. Waiting for the exclusive tree can take
+            // minutes, and a run parked on that lease holds no slot (#347) — so the queue keeps
+            // advancing behind it and the dequeue-time gate is long past. Measured with five in-place
+            // tasks and `maxParallel: 2`: four of them started. Re-ask here, where the very next thing
+            // is a spawn, and hand the run back to the queue if the account closed meanwhile. This
+            // check also covers the explicit lock-bypass path, where the account may close while the
+            // run is preparing its first step.
+            if (this.requeueWhileHeld(runId, workflow, input, taskBackend, state))
+                return;
+        }
+        // Handoff journal (spec 007) — seeded after the worktree exists so the
+        // header can name the branch. Idempotent: an existing file stays as-is.
+        const seeded = this.store.getRun(runId);
+        if (seeded)
+            seedHandoffFile(this.dataDir, seeded);
+        const skills = await discoverSkills(this.repoRoot);
+        // Every ActiveRun construction site must carry the registry — `runContinuation` builds
+        // its own, and the one that skipped this leaked raw `/skill` text to the backend (#811).
+        state.skills = skills;
+        const retriesUsed = new Map();
+        let checkFailure = null;
+        let runError = null;
+        // `startRun` already persisted task images so a queued bubble can render them
+        // (#612). Reuse those files for the agent-facing path note instead of minting
+        // duplicate pasted files when execution finally begins.
+        let startAttachments = (this.store.getRun(runId)?.taskImages ?? [])
+            .map((url) => {
+            const name = url.split('/').pop();
+            if (!name || name.includes('..') || name.includes('/') || name.includes('\\'))
+                return null;
+            const path = join(this.dataDir, 'runs', `${runId}-images`, name);
+            return existsSync(path) ? { name, url, path } : null;
+        })
+            .filter((saved) => saved !== null);
+        // Task screenshots go with the FIRST agent step's opening message only —
+        // later steps and retry loops run in fresh sessions without them. Stacked
+        // attachments (#472) ride along too, but are NOT re-persisted above: they
+        // already live on disk, and adding them to `taskImages` would both duplicate
+        // the files and make the task bubble claim the stack's images as its own.
+        let startImages = input.stackedImages?.length ? [...(input.images ?? []), ...input.stackedImages] : input.images;
+        const lastAgentIdx = findLastAgentStepIndex(workflow);
+        let i = 0;
+        while (i < workflow.steps.length) {
+            if (state.cancelled)
+                break;
+            const step = workflow.steps[i];
+            const kind = stepKind(step);
+            const record = this.store.getRun(runId)?.steps.find((s) => s.id === step.id);
+            const iteration = (record?.iterations ?? 0) + 1;
+            this.store.updateRun(runId, { currentStepId: step.id });
+            this.store.updateStep(runId, step.id, {
+                status: 'running',
+                iterations: iteration,
+                startedAt: new Date().toISOString(),
+                error: undefined,
+            });
+            emit({ type: 'step-start', stepId: step.id, name: step.name ?? step.id, kind, iteration });
+            if (kind === 'agent') {
+                // The last agent step of the workflow is interactive: after its turn
+                // the session stays open for follow-ups until finish/idle/cancel.
+                const interactive = i === lastAgentIdx && i === workflow.steps.length - 1;
+                const failure = await this.runAgentStep(runId, state, step, input, skills, checkFailure, interactive, emit, startImages, taskBackend, extraSystemPrompt, chainStepNote(workflow.steps, i), startAttachments);
+                startImages = undefined;
+                startAttachments = [];
+                checkFailure = null;
+                if (state.cancelled)
+                    break;
+                if (failure) {
+                    this.finishStep(runId, step.id, 'failed', failure, emit);
+                    runError = `step "${step.id}" failed: ${failure}`;
+                    break;
+                }
+                this.finishStep(runId, step.id, 'done', undefined, emit);
+                i++;
+                continue;
+            }
+            const { ok, output } = await this.runCheckStep(state, step, emit);
+            if (state.cancelled)
+                break;
+            if (ok) {
+                this.finishStep(runId, step.id, 'done', undefined, emit);
+                i++;
+                continue;
+            }
+            const used = retriesUsed.get(step.id) ?? 0;
+            if (step.onFail && used < step.onFail.max) {
+                retriesUsed.set(step.id, used + 1);
+                checkFailure = output;
+                this.finishStep(runId, step.id, 'failed', 'check failed — looping back', emit);
+                const retryIdx = workflow.steps.findIndex((s) => s.id === step.onFail?.retry);
+                emit({
+                    type: 'note',
+                    stepId: step.id,
+                    message: `check failed — retrying from "${step.onFail.retry}" (attempt ${used + 1}/${step.onFail.max})`,
+                });
+                // Steps we're about to re-run go back to pending so the GUI rail
+                // reads top-to-bottom truthfully.
+                for (const s of workflow.steps.slice(retryIdx, i + 1)) {
+                    this.store.updateStep(runId, s.id, { status: 'pending' });
+                }
+                i = retryIdx;
+                continue;
+            }
+            this.finishStep(runId, step.id, 'failed', `\`${step.command}\` exited non-zero`, emit);
+            runError = `check "${step.id}" failed${step.onFail ? ` after ${used + 1} attempts` : ''}`;
+            break;
+        }
+        // Final autosave: the branch always ends holding the finished state.
+        this.clearAutosaveTimer(state);
+        if (state.cwd !== this.repoRoot)
+            await autosaveCommit(state.cwd, 'run finalize');
+        const finishedAt = new Date().toISOString();
+        if (state.cancelled) {
+            const run = this.store.getRun(runId);
+            for (const s of run?.steps ?? []) {
+                if (s.status === 'running' || s.status === 'waiting') {
+                    this.store.updateStep(runId, s.id, { status: 'cancelled' });
+                }
+            }
+            this.store.updateRun(runId, { status: 'cancelled', finishedAt, currentStepId: undefined });
+            emit({ type: 'lifecycle', message: 'run cancelled' });
+        }
+        else if (runError) {
+            this.store.updateRun(runId, { status: 'failed', error: runError, finishedAt, currentStepId: undefined });
+            emit({ type: 'lifecycle', message: `run failed — ${runError}` });
+        }
+        else {
+            await this.settleSuccess(runId);
+        }
+        this.clearIdleTimer(state);
+        this.dropActive(runId);
+    }
+    /** Returns an error message, or null on success. */
+    async runAgentStep(runId, state, step, input, skills, checkFailure, interactive, emit, images, taskBackend, extraSystemPrompt, 
+    /** The chain-boundary note for this step (#410), or undefined when the
+     *  workflow has a single agent step and there is no boundary to explain. */
+    chainNote, 
+    /** Pasted attachments already materialized to disk (#357) — their absolute
+     *  paths are appended to `userPrompt` so the agent can operate on the
+     *  real files, not just view the inline image blocks. */
+    attachments = []) {
+        let systemPrompt;
+        if (step.skill) {
+            const skill = skills.find((s) => s.name === step.skill);
+            if (skill) {
+                // The body alone often does not identify the selected skill. Keep its
+                // name and catalog description in the normalized runner payload so a
+                // numeric task such as "432" still gives the model enough context to
+                // describe the work — and therefore derive a useful title (#432).
+                systemPrompt = skillSystemPrompt(skill);
+                // Directory team skills (SKILL.md + references/) get materialized
+                // into <cwd>/.claude/skills/<name>/ — the run's worktree when there
+                // is one — so claude sees the companion files on disk; the shared
+                // info/exclude keeps them out of git (and out of autosave commits).
+                if (skill.source === 'team' && skill.team?.dir) {
+                    const seeded = await materializeSkillDir(state.cwd, skill).catch(() => false);
+                    if (seeded) {
+                        emit({
+                            type: 'note',
+                            stepId: step.id,
+                            message: `team skill "${skill.name}" materialized to .claude/skills/${skill.name}/`,
+                        });
+                    }
+                }
+            }
+            else {
+                emit({
+                    type: 'note',
+                    stepId: step.id,
+                    message: `skill "${step.skill}" not found in .ai/cezar/skills, .ai/skills or the team skills repo — running with the plain prompt`,
+                });
+            }
+        }
+        let userPrompt = applyTemplate(step.prompt ?? '{{task}}', input.task);
+        if (chainNote)
+            userPrompt = `${chainNote}\n\n---\n\n${userPrompt}`;
+        if (checkFailure) {
+            userPrompt += `\n\nA verification command failed after the previous attempt. Fix the cause. Failing output:\n\n${checkFailure}`;
+        }
+        if (images?.length) {
+            emit({
+                type: 'note',
+                stepId: step.id,
+                message: `${images.length} screenshot${images.length > 1 ? 's' : ''} attached to the task`,
+            });
+            // Point the agent at the on-disk files for the pasted subset (#357) — the
+            // base64 blocks above still let it *view* the images; this is what lets it
+            // *use* them as files (save, attach to an issue/PR, copy into the repo).
+            if (attachments.length)
+                userPrompt += `\n\n${pastedAttachmentsText(attachments)}`;
+        }
+        const sessionId = randomUUID();
+        const backend = step.runner ?? taskBackend;
+        this.store.updateStep(runId, step.id, { sessionId, backend });
+        const stepRecord = this.store.getRun(runId)?.steps.find((s) => s.id === step.id);
+        const startTokens = stepRecord?.tokensUsed ?? 0;
+        let stepCost = stepRecord?.costUsd ?? 0;
+        let turnText = '';
+        let sessionError;
+        const sink = this.makeUiSink(runId, step.id);
+        const onEvent = (event) => {
+            if (event.type === 'image') {
+                const saved = this.persistImage(runId, event.mediaType, event.data);
+                if (saved)
+                    emit({ type: 'image', stepId: step.id, ...saved });
+                return;
+            }
+            if (event.type === 'text') {
+                turnText = appendTurnText(turnText, event.text);
+                const text = stripAskMarker(stripTaskMarkers(stripMonitoringMarker(stripDoneMarker(event.text))));
+                if (text)
+                    emit({ type: 'text', text, stepId: step.id });
+                return;
+            }
+            emit({ ...event, stepId: step.id });
+            if (event.type === 'error') {
+                sessionError ??= event.message;
+                state.session?.interrupt();
+                return;
+            }
+            if (sessionError)
+                return;
+            if (event.type === 'session') {
+                // Codex/OpenCode mint their own session id — persist it so resume works.
+                this.store.updateStep(runId, step.id, { sessionId: event.sessionId, backend });
+            }
+            if (event.type === 'token-usage') {
+                this.store.updateStep(runId, step.id, { tokensUsed: startTokens + event.tokensUsed });
+            }
+            if (event.type === 'cost') {
+                stepCost += event.usd;
+                this.store.updateStep(runId, step.id, { costUsd: stepCost });
+            }
+            if (event.type === 'turn-end') {
+                // v2 `turn.completed` already flushed the coalescers; the v1 turn
+                // boundary flushes again (idempotent) as a backstop.
+                sink.flushAll();
+                void this.recordTurnEnd(runId, turnText); // titleSummary + diffStat (#389)
+                const sessionOpen = !state.cancelled && state.session?.open;
+                const done = interactive && sessionOpen && DONE_MARKER_RE.test(turnText.trimEnd());
+                // `CEZ:ASK` → the user is blocked; wins over `CEZ:MONITORING`, loses to
+                // `CEZ:DONE` (#473).
+                const askResult = interactive && sessionOpen && !done ? parseAskMarkerResult(turnText) : undefined;
+                const ask = askResult?.kind === 'valid' ? askResult.request : null;
+                const askRejection = askResult ? askMarkerRejection(askResult) : undefined;
+                const monitoring = interactive &&
+                    sessionOpen &&
+                    !done &&
+                    !ask &&
+                    MONITORING_MARKER_RE.test(turnText.trimEnd());
+                turnText = '';
+                if (askRejection)
+                    emit({ type: 'note', stepId: step.id, message: askRejection });
+                if (done) {
+                    // Goal achieved (agent contract, #347): close the session instead
+                    // of parking at `waiting` — the run completes and frees its slot.
+                    emit({ type: 'lifecycle', message: 'goal achieved — session closed' });
+                    appendHandoffHeartbeat(this.dataDir, runId, 'turn complete — goal achieved, session closed');
+                    state.session?.end();
+                    return;
+                }
+                const waiting = interactive && sessionOpen;
+                if (waiting) {
+                    // Turn over, session open. Either the ball is in the user's court
+                    // (`waiting`) — optionally with a structured `CEZ:ASK` question the
+                    // cockpit renders as an ask card (#473) — or the agent declared it is
+                    // still working on its own downstream work with `CEZ:MONITORING`, which
+                    // parks as `running`/`activity:'monitoring'`, a non-attention state,
+                    // instead of raising "needs you" (#490). Lifecycle is identical: the
+                    // run frees its slot and keeps the idle timer.
+                    if (ask)
+                        emitAskRequested(sink, ask);
+                    if (monitoring) {
+                        this.store.updateRun(runId, { status: 'running', activity: 'monitoring' });
+                        this.store.updateStep(runId, step.id, { status: 'running' });
+                        this.monitoring.add(runId);
+                        this.clearIdleTimer(state);
+                        this.armMonitoringWakeTimer(runId, state);
+                    }
+                    else {
+                        this.store.updateRun(runId, { status: 'waiting', activity: undefined });
+                        this.store.updateStep(runId, step.id, { status: 'waiting' });
+                        this.monitoring.delete(runId);
+                        this.clearMonitoringWakeTimer(state, runId);
+                    }
+                    this.waiting.add(runId);
+                    if (!monitoring)
+                        this.armIdleTimer(runId, state);
+                    this.releaseSlot(); // the freed slot can start a queued run right away — in any project
+                }
+                // The window is proven open — see the twin in `runContinuation`.
+                if (this.store.getRun(runId)?.autoResumeAttempts !== undefined) {
+                    this.store.updateRun(runId, { autoResumeAttempts: undefined });
+                }
+                // Cez's own heartbeat — the handoff stays current even when the
+                // agent forgets to write (spec 007).
+                appendHandoffHeartbeat(this.dataDir, runId, `turn complete — status=${monitoring ? 'monitoring' : waiting ? 'waiting' : 'running'}`);
+            }
+        };
+        const stepBackend = step.runner ?? taskBackend;
+        // Normalise the selected model to canonical `provider/model` and back to the
+        // backend's own wire form via the ONE shared mapper (#405). Fail-loud: an
+        // unresolvable model (e.g. a bare id on opencode) returns the step error
+        // instead of letting the backend silently substitute its default.
+        let backendModel;
+        try {
+            const normalized = normalizeModelForBackend(stepBackend, agentModelsLocked(this.repoRoot) ? undefined : step.model ?? input.model, { configuredProvider: await configuredModelProvider(stepBackend, state.cwd) });
+            backendModel = normalized?.backendModel;
+            // Persist the identity of what ACTUALLY runs (#405, review M1). The run-start echo
+            // (line ~993) is best-effort from `taskBackend`/`input.model`; a per-step `runner`/`model`
+            // override makes it assert a model that never ran. Re-write it here, from the resolved
+            // step identity, so the record — the product of this PR — is always one that ran.
+            this.store.updateRun(runId, {
+                modelIdentity: normalized ? formatModelIdentity(normalized.identity) : undefined,
+            });
+        }
+        catch (err) {
+            if (err instanceof ModelIdentityError)
+                return err.message;
+            throw err;
+        }
+        // Which agent account this step spawns under, and — recorded on the step before the spawn —
+        // which one its session belongs to. `sessionId` and `profileId` are a pair: a resume that
+        // reads the wrong account's config dir finds no session and silently starts a fresh one.
+        // Resolved together with the temp-directory preflight (#785): the step fails with a named,
+        // actionable error instead of spawning a backend whose shell would return empty output.
+        let stepProfile;
+        try {
+            stepProfile = await this.agentEnvForStep(runId, stepBackend, {
+                generateFollowups: followupsEnabled() && input.generateFollowups !== false,
+            });
+        }
+        catch (err) {
+            if (err instanceof AgentTempDirError)
+                return err.message;
+            throw err;
+        }
+        this.store.updateStep(runId, step.id, { profileId: stepProfile.profileId });
+        const runner = createRunner(stepBackend);
+        let session;
+        state.currentStepId = step.id;
+        this.beginUsageInvocation(runId, state, step.id);
+        try {
+            session = runner.startSession({
+                // Skill body, then the run's extra prompt (POST override or config
+                // default), then the handoff/todos contract — every agent step.
+                systemPrompt: composeSystemPrompt(systemPrompt, extraSystemPrompt, followupsEnabled() && input.generateFollowups !== false
+                    ? HANDOFF_INSTRUCTIONS
+                    : HANDOFF_ONLY_INSTRUCTIONS),
+                userPrompt,
+                images,
+                cwd: state.cwd,
+                allowedTools: allowedToolsForStep(step, stepBackend),
+                bashAllowlist: step.bashAllowlist,
+                // The handoff file lives outside the worktree — grant access.
+                additionalDirectories: agentDirectories(join(this.dataDir, 'runs'), stepProfile.env),
+                env: stepProfile.env,
+                model: backendModel,
+                sessionId,
+                // Interactive sessions have no wall clock — the idle timer rules.
+                timeoutMs: interactive ? 0 : undefined,
+            }, onEvent, {
+                autoEndAfterFirstTurn: !interactive,
+                onUiEvent: (event) => this.handleRunnerUiEvent(runId, state, sink, event),
+            });
+        }
+        catch (err) {
+            state.currentStepId = undefined;
+            return err instanceof Error ? err.message : String(err);
+        }
+        state.session = session;
+        state.sessionEverOpened = true;
+        this.flushDeferred(runId);
+        state.currentStepId = step.id;
+        state.interrupt = () => session.interrupt();
+        if (session.pid !== undefined)
+            registerRunProcess(runId, session.pid);
+        try {
+            const result = await session.result;
+            if (sessionError) {
+                sink.sessionEnded('error', sessionError);
+                return sessionError;
+            }
+            // v2 counterpart of v1's `done` (spec: the mappers leave session-close
+            // events to the RunManager — only it knows how the session settled).
+            sink.sessionEnded(state.cancelled ? 'cancelled' : 'end_turn');
+            this.store.updateStep(runId, step.id, { tokensUsed: startTokens + result.tokensUsed });
+            return null;
+        }
+        catch (err) {
+            const message = err instanceof Error ? err.message : String(err);
+            sink.sessionEnded('error', message); // alongside v1's fatal `error`
+            return message;
+        }
+        finally {
+            this.recordUsagePeaks(runId);
+            this.clearIdleTimer(state);
+            this.monitoring.delete(runId);
+            this.waiting.delete(runId);
+            this.clearMonitoringWakeTimer(state, runId);
+            state.session = undefined;
+            state.currentStepId = undefined;
+            state.interrupt = () => undefined;
+        }
+    }
+    /**
+     * Protocol-v2 sink for one agent session (R2 step 2.1): the runner's
+     * `onUiEvent` stream flows through here. Persisted snapshots ride the same
+     * NDJSON file as v1 (the store stamps `seq`/`ts`, `appendEvent` fans them
+     * out live too); coalesced `item.delta` flushes go out live-only via
+     * `emitEphemeral` — raw deltas never hit disk (spec §performance
+     * guardrails). One sink per session: cumulative usage dedup and the
+     * item-shape cache are session-scoped, like the mapper state feeding them.
+     */
+    makeUiSink(runId, stepId) {
+        return new UiEventSink({
+            persist: (event) => this.store.appendEvent(runId, { ...event, stepId }),
+            emitLive: (event) => this.store.emitEphemeral(runId, { ...event, stepId }),
+        });
+    }
+    /** Native backend asks arrive before turn-end. Persist and park immediately
+     * so the cockpit shows attention and the run releases its workspace slot. */
+    handleRunnerUiEvent(runId, state, sink, event) {
+        this.recordUsageUiEvent(runId, state, event);
+        sink.handle(event);
+        if (event.type !== 'ask.requested' || state.cancelled)
+            return;
+        this.clearIdleTimer(state);
+        this.monitoring.delete(runId);
+        this.clearMonitoringWakeTimer(state, runId);
+        this.waiting.add(runId);
+        this.store.updateRun(runId, { status: 'waiting', activity: undefined });
+        if (state.currentStepId)
+            this.store.updateStep(runId, state.currentStepId, { status: 'waiting' });
+        this.releaseSlot();
+    }
+    /** Persist the invocation checkpoint before launching a runner. A throw or
+     * process exit before `turn.started` therefore leaves a durable mismatch. */
+    beginUsageInvocation(runId, state, stepId) {
+        const step = this.store.getRun(runId)?.steps.find((candidate) => candidate.id === stepId);
+        if (!step)
+            return;
+        const epoch = (step.usageInvocationEpoch ?? 0) + 1;
+        this.persistUsageCheckpoint(runId, stepId, {
+            usageInvocationEpoch: epoch,
+            usageInvocationsStarted: (step.usageInvocationsStarted ?? 0) + 1,
+        });
+        state.usageInvocation = {
+            stepId,
+            epoch,
+            observed: false,
+            startedTurns: new Set(),
+            recordedTurns: new Set(),
+        };
+    }
+    /** Fold backend-neutral completed-turn usage into the current step exactly
+     * once. Invocation/turn counters are written before the event reaches the
+     * NDJSON sink so crashes cannot preserve a falsely complete subtotal. */
+    recordUsageUiEvent(runId, state, event) {
+        const invocation = state.usageInvocation;
+        if (!invocation)
+            return;
+        const step = this.store.getRun(runId)?.steps.find((candidate) => candidate.id === invocation.stepId);
+        if (!step)
+            return;
+        if (event.type === 'turn.started') {
+            if (invocation.startedTurns.has(event.turnId))
+                return;
+            invocation.startedTurns.add(event.turnId);
+            const firstObservedTurn = !invocation.observed;
+            invocation.observed = true;
+            this.persistUsageCheckpoint(runId, invocation.stepId, {
+                usageTurnsStarted: (step.usageTurnsStarted ?? 0) + 1,
+                ...(firstObservedTurn
+                    ? { usageInvocationsObserved: (step.usageInvocationsObserved ?? 0) + 1 }
+                    : {}),
+            });
+            return;
+        }
+        if (event.type !== 'turn.completed')
+            return;
+        if (!invocation.startedTurns.has(event.turnId) || invocation.recordedTurns.has(event.turnId))
+            return;
+        const input = event.usage?.input;
+        const output = event.usage?.output;
+        if (typeof input !== 'number' ||
+            !Number.isFinite(input) ||
+            input < 0 ||
+            typeof output !== 'number' ||
+            !Number.isFinite(output) ||
+            output < 0) {
+            return;
+        }
+        invocation.recordedTurns.add(event.turnId);
+        this.persistUsageCheckpoint(runId, invocation.stepId, {
+            inputTokens: (step.inputTokens ?? 0) + input,
+            outputTokens: (step.outputTokens ?? 0) + output,
+            usageTurnsRecorded: (step.usageTurnsRecorded ?? 0) + 1,
+        });
+    }
+    /** Usage completeness is a crash boundary, unlike high-frequency token
+     * snapshots: the checkpoint must reach `runs.json` before the runner starts
+     * or the matching UI event is persisted and forwarded. */
+    persistUsageCheckpoint(runId, stepId, patch) {
+        this.store.updateStep(runId, stepId, patch);
+        this.store.flush();
+    }
+    /**
+     * Turn-end bookkeeping (#389), shared by `runAgentStep` and
+     * `runContinuation` — called (fire-and-forget) from every `turn-end` event:
+     *
+     *  - `titleSummary`: derived from the turn's text, set ONCE — only while the
+     *    record has none. A user's inline edit also lands in `titleSummary`
+     *    (see `PATCH /api/runs/:id`), so an edit is never overwritten either.
+     *  - `diffStat`: cheap `git diff --shortstat` vs the base, refreshed every
+     *    turn. Async and best-effort — a git failure becomes at most a `note`
+     *    event, NEVER a run failure. `updateRun` fans the record out over SSE,
+     *    so the list views pick both up with no extra wiring.
+     *
+     * Not `private` so the integration tests can drive a turn-end directly —
+     * a real agent session is the only other way to reach this path.
+     */
+    /**
+     * The namer's apply path (task auto-naming spec). Fire-and-forget: called
+     * without await from `startRun` (creation) and `recordTurnEnd` (live
+     * refresh). A user-owned title (`titleOrigin: 'user'`) is never overwritten;
+     * namer-owned titles may be replaced by fresher namer results.
+     */
+    async autoNameRun(runId, skillName, task, live) {
+        // CEZ_AUTONAME=0 kills all LLM naming; dry-run skips it too unless
+        // CEZ_AUTONAME=1 forces the mock path — see autoNamingActive.
+        if (!autoNamingActive())
+            return;
+        try {
+            let skillDescription;
+            if (skillName) {
+                const skills = await discoverSkills(this.repoRoot).catch(() => []);
+                skillDescription = skills.find((s) => s.name === skillName)?.description;
+            }
+            const result = await generateRunName(this.repoRoot, { task, skillName, skillDescription, ...live });
+            if (!result)
+                return;
+            const run = this.store.getRun(runId);
+            // Marker-owned state outranks the namer (spec 2026-07-18-task-ref-markers):
+            // a declared title blocks the whole apply (this call raced the marker),
+            // and a declared pr/issue kind blocks that kind field-by-field.
+            if (!run || run.titleOrigin === 'user' || run.titleOrigin === 'marker')
+                return;
+            this.store.updateRun(runId, {
+                titleSummary: result.titleSummary,
+                titleOrigin: 'auto',
+                ...(result.prNumber !== undefined && run.markerRefs?.pr === undefined
+                    ? { prNumber: result.prNumber }
+                    : {}),
+                ...(result.issueNumber !== undefined && run.markerRefs?.issue === undefined
+                    ? { issueNumber: result.issueNumber }
+                    : {}),
+            });
+        }
+        catch {
+            // Naming is best-effort — nothing here may disturb the run.
+        }
+    }
+    async recordTurnEnd(runId, turnText) {
+        try {
+            const run = this.store.getRun(runId);
+            if (!run)
+                return;
+            this.applyTurnMarkers(runId, run, turnText);
+            // Titles are the namer's job (task auto-naming spec) — turn text is
+            // deliberately NEVER a title source; see maybeRefreshTitle below. The
+            // one exception is an explicit CEZ:TITLE declaration (applied above).
+            if (run.worktreePath && existsSync(run.worktreePath)) {
+                // `taskBranch` + `runStartedAt` are what keep this number *this task's* (#751): a
+                // review/QA run repoints the worktree onto the branch under review, and without the
+                // branch to compare HEAD against and the moment it was checked out, the stat would
+                // claim that whole branch's diff.
+                const stat = await worktreeShortstat(run.worktreePath, run.baseBranch ?? 'HEAD', {
+                    taskBranch: run.branch,
+                    runStartedAt: run.startedAt,
+                });
+                if (stat)
+                    this.store.updateRun(runId, { diffStat: stat });
+                else
+                    this.store.appendEvent(runId, { type: 'note', message: 'diff stat unavailable — git diff --shortstat failed in the worktree' });
+            }
+            await this.maybeRefreshTitle(runId, turnText);
+        }
+        catch {
+            // Bookkeeping only — nothing here may disturb the run.
+        }
+    }
+    /**
+     * In-band declarations from the finished turn (spec
+     * 2026-07-18-task-ref-markers): the main thread's own `CEZ:PR=` /
+     * `CEZ:ISSUE=` / `CEZ:TITLE=` lines, parsed from the accumulated turn text
+     * like `CEZ:DONE` — never from tool output. Declared numbers overwrite the
+     * regex/namer display tier (the store re-resolves the referenced-PR chip);
+     * a declared title takes `titleOrigin: 'marker'`, which beats the namer but
+     * never a user rename, and silences the live refresh below.
+     */
+    applyTurnMarkers(runId, run, turnText) {
+        const markers = parseTaskMarkers(turnText);
+        if (markers.pr !== undefined || markers.issue !== undefined) {
+            this.store.applyMarkerRefs(runId, { pr: markers.pr, issue: markers.issue });
+        }
+        if (markers.title && run.titleOrigin !== 'user') {
+            const current = this.store.getRun(runId);
+            const refNumber = current?.prNumber ?? current?.issueNumber;
+            const validated = postValidateTitle(markers.title, refNumber);
+            // Same junk guard as composeNameResult: a declaration that validates to
+            // nothing (or to a bare number prefix) must not blank the title.
+            if (validated && validated !== `${refNumber}:`) {
+                this.store.updateRun(runId, { titleSummary: validated, titleOrigin: 'marker' });
+            }
+        }
+    }
+    /**
+     * Live title refresh (task auto-naming spec, step 3): re-run the namer with
+     * the turn's context. Skips: toggle off (`liveTitleUpdates` config over
+     * `CEZ_TITLE_UPDATES` env, default ON), user-owned title, marker-owned title
+     * (the agent declares via `CEZ:TITLE` — the token-saving fast path), dry-run
+     * mocks (canned answers add nothing), empty turn text, unchanged namer inputs.
+     */
+    async maybeRefreshTitle(runId, turnText) {
+        if (!autoNamingActive())
+            return;
+        if (!turnText.trim())
+            return;
+        const config = await loadConfig(this.repoRoot);
+        if (!liveTitleUpdatesEnabled(config))
+            return;
+        const run = this.store.getRun(runId);
+        if (!run || run.titleOrigin === 'user' || run.titleOrigin === 'marker')
+            return;
+        const statText = run.diffStat ? `${run.diffStat.files} files, +${run.diffStat.adds} -${run.diffStat.dels}` : undefined;
+        const key = `${turnText.slice(0, 200)}|${statText ?? ''}`;
+        if (this.lastNamerKey.get(runId) === key)
+            return;
+        this.lastNamerKey.set(runId, key);
+        const workflow = await this.reviveWorkflow(run);
+        const skillName = workflow?.steps.find((s) => stepKind(s) === 'agent' && s.skill)?.skill?.trim();
+        void this.autoNameRun(runId, skillName, run.task, { turnText, diffStat: statText });
+    }
+    /**
+     * End-of-session telemetry (#348): stop sampling the run's process tree and
+     * fold the session's peaks into the run record. `max` with existing values —
+     * a run can hold several sessions (multiple agent steps, Continue) and the
+     * record keeps the highest water mark across all of them.
+     */
+    recordUsagePeaks(runId) {
+        const peaks = unregisterRunProcess(runId);
+        if (!peaks)
+            return;
+        const run = this.store.getRun(runId);
+        this.store.updateRun(runId, {
+            peakRssBytes: Math.max(run?.peakRssBytes ?? 0, peaks.peakRssBytes),
+            peakProcCount: Math.max(run?.peakProcCount ?? 0, peaks.peakProcCount),
+        });
+    }
+    /**
+     * Diff-first review gate (spec 009), shared by `execute` and
+     * `runContinuation`: a *successful* run whose worktree holds changes rests
+     * at `review` instead of `done` — the user inspects the diff first, then
+     * sends feedback back, opens a draft PR, or just finishes. Failed/cancelled
+     * runs never enter review; no worktree or an empty diff means plain `done`.
+     *
+     * The gate is opt-in (#489): the review park happens only when it is enabled
+     * (`reviewGateEnabled` — config toggle over the `CEZ_REVIEW_GATE` env, default
+     * OFF) AND the run is not autonomous. Autonomous runs — and runs with the gate
+     * off — settle straight to `done`, leaving the diff in the worktree untouched.
+     */
+    async settleSuccess(runId) {
+        const run = this.store.getRun(runId);
+        let review = false;
+        if (run?.worktreePath && existsSync(run.worktreePath)) {
+            const diff = await worktreeDiff(run.worktreePath, run.baseBranch ?? 'HEAD');
+            const hasDiff = diff.trim().length > 0 && !diff.startsWith('(diff failed');
+            const config = await loadConfig(this.repoRoot);
+            review = hasDiff && reviewGateEnabled(config) && run.autonomous !== true;
+        }
+        this.store.updateRun(runId, {
+            status: review ? 'review' : 'done',
+            finishedAt: new Date().toISOString(),
+            currentStepId: undefined,
+            // A run that got all the way to a settled turn is not in a limit loop, so the resume
+            // counter starts over — otherwise a task that legitimately met the limit once a week would
+            // creep toward the cap forever and stop resuming for no reason anyone could see.
+            autoResumeAttempts: undefined,
+        });
+        this.store.appendEvent(runId, {
+            type: 'lifecycle',
+            message: review
+                ? 'changes ready for review — send feedback, open a draft PR, or finish'
+                : 'run finished',
+        });
+    }
+    /**
+     * Agent screenshot (an image block inside a tool result) or a user-pasted
+     * attachment: the base64 data never enters the NDJSON event log — it lands
+     * as a file under `.ai/cezar/runs/<id>-images/` and the transcript event
+     * carries only the name + serving URL. `namePrefix` distinguishes the two
+     * origins on disk (`screenshot-<n>.<ext>` for agent tool screenshots,
+     * `pasted-<n>.<ext>` for user-pasted attachments, #357) and the absolute
+     * `path` lets the agent operate on the file directly (save/attach/upload).
+     * Best effort: on failure the attachment is dropped, the transcript still
+     * shows the tool result's `[screenshot]` placeholder (or the image count).
+     */
+    persistImage(runId, mediaType, data, namePrefix = 'screenshot') {
+        try {
+            const ext = /png/.test(mediaType) ? 'png'
+                : /jpe?g/.test(mediaType) ? 'jpg'
+                    : /webp/.test(mediaType) ? 'webp'
+                        : /gif/.test(mediaType) ? 'gif'
+                            : 'img';
+            const dir = join(this.dataDir, 'runs', `${runId}-images`);
+            mkdirSync(dir, { recursive: true });
+            // Seed from the highest numeric suffix already on disk, NOT the file count:
+            // `screenshot-*` and `pasted-*` share one numbering space, so counting would
+            // re-issue a live number after any deletion. Only matters on the first write
+            // of a process (restart case) — afterwards the map is authoritative.
+            let seq = this.queuedImageSeq.get(runId);
+            if (seq === undefined)
+                seq = highestImageSeq(dir);
+            // `persistImage` is fully synchronous, so two pastes cannot interleave between
+            // the read of the counter and the write. The exclusive-create flag is the
+            // belt-and-braces guard for a stale seed: it degrades to a renamed file rather
+            // than a silent overwrite.
+            for (let attempt = 0; attempt < 100; attempt += 1) {
+                seq += 1;
+                const name = `${namePrefix}-${seq}.${ext}`;
+                const path = join(dir, name);
+                try {
+                    writeFileSync(path, Buffer.from(data, 'base64'), { flag: 'wx' });
+                }
+                catch (err) {
+                    if (err.code === 'EEXIST')
+                        continue;
+                    throw err;
+                }
+                this.queuedImageSeq.set(runId, seq);
+                // Versioned, because that is the only surface served now. The cockpit still upgrades
+                // the unversioned URLs sitting in OLD transcripts when it renders them
+                // (`resolveApiUrl`), but a URL minted today must be fetchable as written.
+                return { name, url: `/api/v1/runs/${runId}/images/${name}`, path };
+            }
+            return null;
+        }
+        catch {
+            return null;
+        }
+    }
+    armIdleTimer(runId, state) {
+        this.clearIdleTimer(state);
+        state.idleTimer = setTimeout(() => {
+            if (state.session?.open && !state.cancelled) {
+                this.store.appendEvent(runId, {
+                    type: 'lifecycle',
+                    message: `session closed after ${Math.round(IDLE_TIMEOUT_MS / 60_000)}m of inactivity`,
+                });
+                state.session.end();
+            }
+        }, IDLE_TIMEOUT_MS);
+        state.idleTimer.unref?.();
+    }
+    clearIdleTimer(state) {
+        if (state.idleTimer) {
+            clearTimeout(state.idleTimer);
+            state.idleTimer = undefined;
+        }
+    }
+    reconcileMonitoringWakeTimers() {
+        for (const runId of this.monitoring) {
+            const state = this.active.get(runId);
+            if (state)
+                this.armMonitoringWakeTimer(runId, state);
+        }
+    }
+    armMonitoringWakeTimer(runId, state) {
+        const minutes = this.semaphore.monitoringWakeIntervalMinutes();
+        if (minutes === null) {
+            this.clearMonitoringWakeTimer(state, runId);
+            return;
+        }
+        if ((state.monitoringWakeups ?? 0) >= MAX_AUTO_CONTINUES) {
+            this.clearMonitoringWakeTimer(state, runId);
+            if (!this.store.getRun(runId)?.monitoringWakeCapReached) {
+                this.store.updateRun(runId, { monitoringWakeCapReached: true });
+                this.store.appendEvent(runId, {
+                    type: 'note',
+                    message: `automatic monitoring wake-up cap reached (${MAX_AUTO_CONTINUES}); session remains parked`,
+                });
+            }
+            return;
+        }
+        if (state.monitoringWakeTimer && state.monitoringWakeIntervalMinutes === minutes)
+            return;
+        this.clearMonitoringWakeTimer(state, runId);
+        state.monitoringWakeIntervalMinutes = minutes;
+        this.store.updateRun(runId, { monitoringWakeCapReached: undefined });
+        const deadline = Date.now() + minutes * 60_000;
+        this.store.updateRun(runId, { monitoringWakeAt: new Date(deadline).toISOString() });
+        state.monitoringWakeTimer = setTimeout(() => {
+            state.monitoringWakeTimer = undefined;
+            this.store.updateRun(runId, { monitoringWakeAt: undefined });
+            if (!this.monitoring.has(runId) || !state.session?.open || state.cancelled)
+                return;
+            const wakeups = state.monitoringWakeups ?? 0;
+            if (wakeups >= MAX_AUTO_CONTINUES) {
+                this.store.updateRun(runId, { monitoringWakeCapReached: true });
+                this.store.appendEvent(runId, {
+                    type: 'note',
+                    message: `automatic monitoring wake-up cap reached (${MAX_AUTO_CONTINUES}); session remains parked`,
+                });
+                return;
+            }
+            state.monitoringWakeups = wakeups + 1;
+            this.store.appendEvent(runId, {
+                type: 'note',
+                message: `automatic monitoring wake-up (${state.monitoringWakeups}/${MAX_AUTO_CONTINUES})`,
+            });
+            this.deliverMessage(runId, [{ type: 'text', text: MONITORING_WAKE_NUDGE }], false);
+        }, Math.max(0, deadline - Date.now()));
+        state.monitoringWakeTimer.unref?.();
+    }
+    clearMonitoringWakeTimer(state, runId) {
+        if (state.monitoringWakeTimer)
+            clearTimeout(state.monitoringWakeTimer);
+        state.monitoringWakeTimer = undefined;
+        state.monitoringWakeIntervalMinutes = undefined;
+        if (runId)
+            this.store.updateRun(runId, { monitoringWakeAt: undefined });
+    }
+    /** Autosave-commit the worktree every 90 s while the run lives (spec 006).
+     *  Opt-in via CEZ_AUTOSAVE=1 (#471) — see periodicAutosaveEnabled. */
+    armAutosave(state) {
+        if (!periodicAutosaveEnabled())
+            return;
+        if (state.cwd === this.repoRoot || state.autosaveTimer)
+            return;
+        state.autosaveTimer = setInterval(() => {
+            void autosaveCommit(state.cwd, 'periodic');
+        }, AUTOSAVE_INTERVAL_MS);
+        state.autosaveTimer.unref?.();
+    }
+    clearAutosaveTimer(state) {
+        if (state.autosaveTimer) {
+            clearInterval(state.autosaveTimer);
+            state.autosaveTimer = undefined;
+        }
+    }
+    runCheckStep(state, step, emit) {
+        const command = step.command;
+        emit({ type: 'note', stepId: step.id, message: `$ ${command}` });
+        return new Promise((resolve) => {
+            // Check steps run in the same cwd as the agent steps — the worktree.
+            const child = spawn('bash', ['-lc', command], { cwd: state.cwd, env: process.env });
+            state.interrupt = () => child.kill('SIGTERM');
+            let output = '';
+            const collect = (chunk) => {
+                if (output.length < CHECK_OUTPUT_CAP) {
+                    output += chunk.toString('utf8');
+                    if (output.length >= CHECK_OUTPUT_CAP)
+                        output += '\n… (output truncated)';
+                }
+            };
+            child.stdout.on('data', collect);
+            child.stderr.on('data', collect);
+            child.on('error', (err) => {
+                state.interrupt = () => undefined;
+                const message = `failed to spawn: ${err.message}`;
+                emit({ type: 'check-output', stepId: step.id, command, text: message, exitCode: -1 });
+                resolve({ ok: false, output: message });
+            });
+            child.on('close', (code) => {
+                state.interrupt = () => undefined;
+                const trimmed = output.trim() || '(no output)';
+                emit({ type: 'check-output', stepId: step.id, command, text: trimmed, exitCode: code ?? -1 });
+                resolve({ ok: code === 0, output: trimmed });
+            });
+        });
+    }
+    finishStep(runId, stepId, status, error, emit) {
+        this.store.updateStep(runId, stepId, {
+            status,
+            error,
+            finishedAt: new Date().toISOString(),
+        });
+        emit({ type: 'step-end', stepId, status, ...(error ? { error } : {}) });
+        appendHandoffHeartbeat(this.dataDir, runId, `step "${stepId}" complete — status=${status}`);
+    }
+}
+function findLastAgentStepIndex(workflow) {
+    for (let i = workflow.steps.length - 1; i >= 0; i--) {
+        const step = workflow.steps[i];
+        if (step && stepKind(step) === 'agent')
+            return i;
+    }
+    return -1;
+}
+function applyTemplate(template, task) {
+    return template.replaceAll('{{task}}', task);
+}
+/**
+ * Immediate title shown while a run is queued. The namer's `titleSummary`
+ * replaces it once the model answers; this is the honest, permanent fallback
+ * when no model is available (#432, spec 2026-07-17-task-auto-naming). When
+ * the task references a PR/issue, the number leads: `469: /om-auto-review-pr`.
+ */
+export function makeRunTitle(task, workflow) {
+    const firstLine = task.trim().split('\n')[0] ?? '';
+    const skill = workflow.steps.find((step) => stepKind(step) === 'agent' && step.skill)?.skill?.trim();
+    const contextual = skill && !firstLine.startsWith(`/${skill}`)
+        ? `/${skill}${firstLine ? ` ${firstLine}` : ''}`
+        : firstLine;
+    const refNumber = titleRefNumber(refineTaskRefs(extractTaskRefs(task), skill));
+    // `469` or `/om-auto-review-pr 469` reads as `469: /om-auto-review-pr` — the
+    // number leads so it survives the tasks table's narrow truncation.
+    const skillArg = skill && contextual.startsWith(`/${skill}`) ? contextual.slice(skill.length + 1).trim() : null;
+    const body = refNumber !== undefined && skill && (skillArg === '' || /^#?\d+$/.test(skillArg ?? ''))
+        ? `/${skill}`
+        : contextual;
+    const prefixed = refNumber !== undefined && !body.trimStart().replace(/^#/, '').startsWith(String(refNumber))
+        ? `${refNumber}: ${body}`
+        : body;
+    const chars = [...(prefixed || '(untitled task)')];
+    return chars.length > 80 ? `${chars.slice(0, 79).join('').trimEnd()}…` : chars.join('');
+}
+/**
+ * Skill identity is context, while the Markdown body remains instructions.
+ *
+ * For an on-disk skill we also hand the agent the ABSOLUTE directory of the
+ * installed copy. A run executes in an isolated worktree that has no local
+ * `.agents/skills` (gitignored, absent in a fresh checkout), so without this
+ * the agent cannot read the skill's companion files (`references/*.md`) — or,
+ * worse, reads a stale copy materialized from the team-repo cache. The path
+ * resolves against the MAIN project root (`discoverSkills(repoRoot)`), i.e. the
+ * current `npx skills`-installed copy, so a worktree agent and the main
+ * checkout read the exact same, up-to-date files. Team skills are omitted here:
+ * they are materialized into the worktree separately (see the call site).
+ */
+export function skillSystemPrompt(skill) {
+    const lines = [
+        `Selected skill: /${skill.name}`,
+        ...(skill.description ? [`Description: ${skill.description}`] : []),
+    ];
+    if (skill.source && skill.source !== 'team' && skill.path) {
+        const dir = dirname(skill.path);
+        lines.push('', `Skill files are installed on disk at: ${dir}`, `Read any file this skill references (for example references/*.md) from that absolute directory. ` +
+            `It is the current installed copy — use it even though your working directory is a separate worktree that does not contain the skill.`);
+    }
+    lines.push('', 'Skill instructions:', skill.body.trim());
+    return lines.join('\n');
+}
+/**
+ * Expand a registry-backed slash skill in one prompt string before it reaches a
+ * backend. Claude otherwise intercepts an unknown leading slash command, and
+ * Codex/OpenCode have no native slash-skill lookup at all (#676).
+ *
+ * Only a match at character zero counts, and unknown commands pass through
+ * byte-for-byte — a backend's OWN slash commands must keep working. The caller
+ * persists the original user text before applying this delivery-only rewrite.
+ *
+ * Both delivery seams route through here: live-session messages via
+ * `expandRegistrySlashSkill`, and a continuation's opening prompt, which becomes
+ * the session's `userPrompt` and never passes through `deliverMessage` at all
+ * (#811).
+ */
+export function expandRegistrySlashSkillText(text, skills) {
+    const match = /^\/([A-Za-z0-9][A-Za-z0-9._-]*)(?=\s|$)/.exec(text);
+    if (!match)
+        return text;
+    const skill = skills.find((candidate) => candidate.name === match[1]);
+    if (!skill)
+        return text;
+    const request = text.slice(match[0].length).trim();
+    return request ? `${skillSystemPrompt(skill)}\n\nUser request:\n${request}` : skillSystemPrompt(skill);
+}
+/**
+ * `expandRegistrySlashSkillText` over a live chat message: only the first text
+ * block is eligible, and an unchanged block returns the caller's array
+ * identity untouched.
+ */
+export function expandRegistrySlashSkill(content, skills) {
+    const textIndex = content.findIndex((block) => block.type === 'text');
+    if (textIndex < 0)
+        return content;
+    const block = content[textIndex];
+    if (!block || block.type !== 'text')
+        return content;
+    const text = expandRegistrySlashSkillText(block.text, skills);
+    if (text === block.text)
+        return content;
+    const expanded = [...content];
+    expanded[textIndex] = { type: 'text', text };
+    return expanded;
+}
+//# sourceMappingURL=run.js.map