@nredd/pi-subagents 0.0.0-stage → 0.22.0-nredd.6

package/src/agent-manager.ts

@@ -0,0 +1,1874 @@
+/**
+ * agent-manager.ts — Tracks agents, background execution, resume support.
+ *
+ * There are two independent concurrency pools, never one:
+ *
+ * - Background (`maxConcurrent`, default 10) bounds detached agents.
+ * - Foreground (`maxConcurrentForeground`, default 0 = unlimited) bounds
+ *   agents a caller is blocking on inline — `spawnAndWait`.
+ *
+ * Independent by design: a foreground agent blocks the parent anyway, so
+ * charging it to the background pool would let a saturated pool starve the main
+ * session of work it could have done itself. Excess agents in either pool are
+ * queued and auto-started as slots free up. Nested children take no slot in
+ * either — see `occupiesPoolSlot` / `occupiesForegroundSlot`.
+ */
+
+import { randomUUID } from "node:crypto";
+import { statSync } from "node:fs";
+import { isAbsolute } from "node:path";
+import type { Model } from "@earendil-works/pi-ai";
+import type { AgentSession, ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
+import { type AssistantMessageInfo, resumeAgent, runAgent, type ToolActivity } from "./agent-runner.js";
+import { getAgentConfig } from "./agent-types.js";
+import { type Clock, realClock, type TimerHandle } from "./clock.js";
+import { assignHandle, handleBase } from "./mention.js";
+import type { ModelCallEvent } from "./model-call.js";
+import { describeModel, describePhysicalModel, describeVirtualModel, isVirtualModel } from "./model-resolver.js";
+import type { AgentInvocation, AgentRecord, AgentTombstone, IsolationMode, MentionResolution, Pool, SubagentType, ThinkingLevel } from "./types.js";
+import { addUsage, type LifetimeUsage } from "./usage.js";
+import type { CompiledSchema } from "./workflow/json-schema.js";
+import { cleanupWorktree, createWorktree, isWorktreeIsolationEnabled, pruneWorktrees, type WorktreeInfo } from "./worktree.js";
+
+/**
+ * Relabel a record's model from the response that just finished. A virtual
+ * selection (`router/default`) is only a name until a request is routed; the
+ * assistant message is the one place that records the physical model, so the
+ * label follows it (and follows a router fallback to the next model).
+ */
+function relabelFromResponse(record: AgentRecord, message: AssistantMessageInfo | undefined): void {
+  const registry = (record.session as { modelRegistry?: Parameters<typeof describePhysicalModel>[1] } | undefined)
+    ?.modelRegistry;
+  const physical = describePhysicalModel(message, registry);
+  if (!physical) return;
+  record.invocation ??= {};
+  Object.assign(record.invocation, physical);
+}
+
+export type OnAgentComplete = (record: AgentRecord) => void;
+export type OnAgentStart = (record: AgentRecord) => void;
+/**
+ * Fired by `steer()` once a message is delivered or queued. Every steer path
+ * goes through it except the steer tools (`steer_subagent` in index.ts, and its
+ * nested twin in nested-tools.ts), which await the session directly.
+ */
+export type OnAgentSteered = (record: AgentRecord, message: string) => void;
+export type OnAgentCompact = (record: AgentRecord, info: CompactionInfo) => void;
+/**
+ * Fired once per assistant `message_end`, for EVERY agent this manager owns —
+ * top-level and nested alike, spawns and resumes. The one place where each
+ * message is seen exactly once: `AgentRecord.lifetimeUsage` is deliberately
+ * double-booked into ancestors (see `nested-tools.ts`) so a hidden child's spend
+ * shows up on the record a human can see, which makes those records useless as
+ * a basis for anything that must not count a message twice — parent-session
+ * accounting above all.
+ */
+export type OnAgentUsage = (record: AgentRecord, usage: LifetimeUsage) => void;
+export type CompactionInfo = { reason: "manual" | "threshold" | "overflow"; tokensBefore: number };
+
+/**
+ * Default max concurrent background agents.
+ *
+ * Raised from 4 when top-level spawns started defaulting to background
+ * (`backgroundByDefault`): foreground agents bypass this pool entirely, so
+ * while foreground was the default a fan-out of six ran six. With background
+ * as the default every top-level agent takes a slot, and a limit of 4 would
+ * have silently queued the tail of exactly the parallel fan-outs the `Agent`
+ * tool description tells the model to send.
+ */
+const DEFAULT_MAX_CONCURRENT = 10;
+
+/**
+ * Default max concurrent foreground (blocking) agents — `0` = unlimited, the
+ * extension's existing convention for "no ceiling" (`defaultMaxTurns`).
+ *
+ * Off by default because nothing here ever bounded foreground work, and pi
+ * dispatches a message's tool calls through `Promise.all`, so an unqualified
+ * fan-out of blocking `Agent` calls has always run all at once. Users who want
+ * it bounded — chiefly local models, where parallel agents thrash the prompt
+ * cache (#253) — opt in; everyone else keeps today's behaviour exactly.
+ */
+const DEFAULT_MAX_CONCURRENT_FOREGROUND = 0;
+
+/**
+ * How long a finished record stays in memory once its result was consumed
+ * (fetched, or delivered inline to a blocking caller). Nothing is waiting on it
+ * any more; the window only covers a follow-up `get_subagent_result` or steer.
+ */
+const CONSUMED_RECORD_TTL_MS = 10 * 60_000;
+
+/**
+ * How long a finished record nobody has read stays in memory, measured from
+ * completion. Far longer than the consumed window: an orchestrator that was busy
+ * when the notice arrived still gets to fetch the full result an hour later.
+ */
+const UNCONSUMED_RECORD_TTL_MS = 60 * 60_000;
+
+/**
+ * Read a retention override in minutes (fractions and 0 allowed) from `name`,
+ * else `fallbackMs`. An env var rather than a setting: the windows are not a
+ * user knob, but eviction is otherwise unobservable in an end-to-end test,
+ * which cannot wait an hour. Invalid values warn and keep the default.
+ */
+function retentionMs(name: string, fallbackMs: number): number {
+  const raw = process.env[name]?.trim();
+  if (!raw) return fallbackMs;
+  const minutes = Number(raw);
+  if (!Number.isFinite(minutes) || minutes < 0) {
+    console.warn(`[pi-subagents] ignoring ${name}="${raw}": expected minutes >= 0`);
+    return fallbackMs;
+  }
+  return minutes * 60_000;
+}
+
+/** Sweep cadence: every minute, or as often as the shortest window needs, never under a second. */
+const MAX_SWEEP_INTERVAL_MS = 60_000;
+const MIN_SWEEP_INTERVAL_MS = 1_000;
+
+/** How many evicted ids keep an answer to "what happened to that agent". */
+const MAX_EVICTED_NOTES = 500;
+
+/** Mark a record's result as read, which shortens its retention and suppresses its notice. */
+export function markConsumed(record: Pick<AgentRecord, "resultConsumed" | "consumedAt">, now = Date.now()): void {
+  record.resultConsumed = true;
+  record.consumedAt = now;
+}
+
+/**
+ * What survives a record's eviction, so a later `get_subagent_result` or steer
+ * can say what became of the agent instead of "not found".
+ */
+export interface EvictedNote {
+  id: string;
+  type: SubagentType;
+  description: string;
+  status: AgentRecord["status"];
+  /** Whole minutes between completion and eviction. */
+  evictedAfterMin: number;
+  /** The `.output` transcript when one was written, else the session file. */
+  transcript?: string;
+}
+
+/** The text a tool returns for an id whose record was evicted. */
+export function describeEvicted(note: EvictedNote): string {
+  const where = note.transcript ? `; transcript: ${note.transcript}` : "";
+  return `Agent "${note.id}" (${note.description}) ${note.status}, evicted after ${note.evictedAfterMin} min${where}`;
+}
+
+/**
+ * How many evicted agents stay addressable by name. Only a bound on memory —
+ * a session that spawns hundreds of agents shouldn't retain every one — and
+ * far above the handful anyone keeps in their head.
+ */
+const MAX_TOMBSTONES = 100;
+
+/**
+ * Validate a caller-supplied SpawnOptions.cwd. `undefined`/`null` mean "unset"
+ * (parent cwd). Anything else must be an absolute path to an existing
+ * directory — curated errors instead of TypeErrors from path/fs internals
+ * (RPC callers send arbitrary JSON: null, numbers, file paths).
+ */
+function assertValidSpawnCwd(cwd: unknown): asserts cwd is string | undefined | null {
+  if (cwd == null) return;
+  if (typeof cwd !== "string" || !isAbsolute(cwd)) {
+    throw new Error(`SpawnOptions.cwd must be an absolute path: "${String(cwd)}"`);
+  }
+  let isDirectory = false;
+  try {
+    isDirectory = statSync(cwd).isDirectory();
+  } catch {
+    throw new Error(`SpawnOptions.cwd does not exist: "${cwd}"`);
+  }
+  if (!isDirectory) {
+    throw new Error(`SpawnOptions.cwd is not a directory: "${cwd}"`);
+  }
+}
+
+/**
+ * Whether a record occupies one of the `maxConcurrent` background slots.
+ * Nested children don't: their parent already holds a slot, so counting (and
+ * therefore queueing) them would deadlock a parent that waits on its own child.
+ *
+ * Note this bounds nothing horizontally — the depth cap limits how DEEP nesting
+ * goes, not how WIDE. A parent's only limit on concurrent children is that each
+ * spawn costs it a turn, which is unbounded when max turns is unlimited.
+ */
+function occupiesPoolSlot(
+  record: Pick<AgentRecord, "isBackground" | "parentAgentId" | "workflowId">,
+): boolean {
+  return !!record.isBackground && isTopLevelAgent(record);
+}
+
+/**
+ * Whether a record is one of the session's own agents, rather than something
+ * another agent or a workflow owns.
+ *
+ * The single definition behind every user-facing surface — the fleet list, the
+ * widget, the `/agents` menus, `@handle` resolution, and the completion events
+ * and session entries. An owned child reports through its owner, so surfacing
+ * it separately would double-count the same work in the places a person reads.
+ */
+export function isTopLevelAgent(
+  record: Pick<AgentRecord, "parentAgentId" | "workflowId">,
+): boolean {
+  return record.parentAgentId === undefined && record.workflowId === undefined;
+}
+
+/**
+ * Whether a record occupies one of the `maxConcurrentForeground` slots.
+ *
+ * Keyed on `blocking` — a caller awaiting this record inline — rather than on
+ * `isBackground === false`, because `spawn()` is also the funnel for DETACHED
+ * starts (cross-extension RPC, `@handle` mentions, the registry) that may pass
+ * `isBackground: false` and are documented to run immediately regardless. Those
+ * block nobody, so bounding them buys nothing and would park a record with no
+ * one waiting to release it.
+ *
+ * Nested children are excluded for the same reason as `occupiesPoolSlot`, and
+ * more sharply: their parent is blocked *awaiting them*, so queueing a child
+ * behind its own parent is a guaranteed deadlock rather than a possible one.
+ * Enforced here rather than at the call site so no caller can reintroduce it.
+ *
+ * A workflow's children go out through `spawnAndWait` and so are `blocking`
+ * too, and are excluded on the same `isTopLevelAgent` test as the background
+ * pool: the run already caps how many of its agents run at once, and charging
+ * them here as well would let one fan-out queue behind a limit meant for the
+ * session's own work.
+ *
+ * Like the background pool this bounds width at the top level only — a parent's
+ * own fan-out is limited by nothing but its turn budget.
+ */
+function occupiesForegroundSlot(
+  record: Pick<AgentRecord, "blocking" | "parentAgentId" | "workflowId">,
+): boolean {
+  return !!record.blocking && isTopLevelAgent(record);
+}
+
+interface SpawnArgs {
+  pi: ExtensionAPI;
+  ctx: ExtensionContext;
+  type: SubagentType;
+  prompt: string;
+  options: SpawnOptions;
+}
+
+interface SpawnOptions {
+  description: string;
+  /**
+   * Optional memorable name for this instance, becoming a second handle
+   * (`@auth-audit`) alongside the type-derived one. Slugged, not validated —
+   * anything unusable degrades via `handleBase` rather than failing the spawn.
+   */
+  name?: string;
+  /**
+   * Reopen this pi session file instead of starting a fresh conversation, so a
+   * mention of an evicted agent continues where it left off. The agent's
+   * definition is still resolved from its type, so the continuation runs under
+   * the type's CURRENT config.
+   */
+  resumeSessionFile?: string;
+  /**
+   * Take an evicted agent's names back verbatim instead of allocating fresh
+   * ones, so a resumed conversation keeps the handle the user just typed —
+   * `handleBase(type)` cannot reproduce a numbered `explore-2`. Safe without an
+   * `assignHandle` pass because tombstoned names are excluded from allocation
+   * (`takenHandles`), so nothing live can be holding them.
+   *
+   * Internal capability, like `resumeSessionFile`: a forged handle would
+   * duplicate a live agent's name and make `resolveMention` ambiguous, so
+   * `spawnTopLevel` strips it from anything a caller sends.
+   */
+  reclaim?: { handle: string; alias?: string };
+  model?: Model<any>;
+  maxTurns?: number;
+  isolated?: boolean;
+  inheritContext?: boolean;
+  thinkingLevel?: ThinkingLevel;
+  isBackground?: boolean;
+  /**
+   * Skip whichever pool's queue check applies to this spawn — start immediately
+   * even if the configured concurrency limit would otherwise queue it. The slot
+   * is still COUNTED once the run starts, so a bypassing spawn transiently
+   * exceeds the limit rather than being invisible to it.
+   *
+   * Used by the scheduler, so a fired job can't be deferred past its trigger
+   * window, and by the `/agents` agent-file generator, which has no way to
+   * cancel a wait (see its call site).
+   */
+  bypassQueue?: boolean;
+  /**
+   * A caller is awaiting this record inline (`spawnAndWait`) — what
+   * `maxConcurrentForeground` bounds. Set only by `spawnAndWait`; stripped from
+   * caller-supplied options by `spawnTopLevel`, since a forged `blocking` would
+   * defer a detached start behind a queue its caller cannot see or release.
+   */
+  blocking?: boolean;
+  /**
+   * The workflow run this child belongs to, when a workflow spawned it.
+   *
+   * Ownership, not decoration. A workflow's children are the workflow's — they
+   * report through its card, its notification and its dialog, so they are
+   * filtered out of every top-level surface exactly as nested children are, and
+   * they take no `maxConcurrent` slot: the run has its own concurrency cap, and
+   * counting them twice would let one workflow starve the whole session.
+   */
+  workflowId?: string;
+  /**
+   * Make the child report through a `StructuredOutput` tool built from this
+   * compiled schema. Set only by the workflow host, for `agent({ schema })`.
+   */
+  structuredOutput?: CompiledSchema;
+  /** Isolation mode — "worktree" creates a temp git worktree for the agent. */
+  isolation?: IsolationMode;
+  /**
+   * Working directory for the agent (absolute path). Default: parent session
+   * cwd. The agent's tools operate here, but .pi config (extensions, skills,
+   * settings, memory) still loads from the parent session's project — the
+   * target directory's `.pi` extensions never execute. With isolation:
+   * "worktree", the worktree is created FROM this directory and the result
+   * branch lands in that repo.
+   */
+  cwd?: string;
+  /**
+   * Last chance to look at an isolated agent's worktree, awaited immediately
+   * before it is committed to a branch and removed.
+   *
+   * Exists because that removal happens inside the settle path, before
+   * `spawnAndWait` resolves: by the time a caller has the finished record, the
+   * directory the child actually wrote in is gone. Anything that must inspect
+   * or verify that tree — a workflow `gate` is the motivating case — has to run
+   * here or it silently inspects the main tree instead.
+   *
+   * Fires only on the normal settle path, and only when a worktree was created.
+   * Not on the error path and not on the stop-during-copy guard: those are
+   * already failing, and delaying cleanup there would leak a copy for no gain.
+   * A rejection is swallowed — the hook can never keep the worktree alive.
+   */
+  onBeforeWorktreeCleanup?: (worktreePath: string) => Promise<void>;
+  /** Resolved invocation snapshot captured for UI display. */
+  invocation?: AgentInvocation;
+  /** Parent abort signal — when aborted, the subagent is also stopped. */
+  signal?: AbortSignal;
+  /**
+   * Called synchronously once the record is in the map and its promise is set,
+   * before `onSessionCreated` fires — where callers attach the output file.
+   *
+   * Carried on the options rather than parked on the manager for the duration
+   * of a spawn: with a foreground queue, `startAgent` can run at drain time,
+   * long after any such field would have been restored, and the callback would
+   * silently never fire (or fire into an unrelated caller's closure).
+   */
+  onSpawned?: (id: string) => void;
+  /**
+   * Called synchronously when the spawn is queued instead of started, with how
+   * many entries in its own pool are ahead of it. The foreground UI uses it to
+   * say so while it waits; nothing else needs it.
+   */
+  onQueued?: (id: string, ahead: number) => void;
+  /** Called on tool start/end with activity info (for streaming progress to UI). */
+  onToolActivity?: (activity: ToolActivity) => void;
+  /** Called on streaming text deltas from the assistant response. */
+  onTextDelta?: (delta: string, fullText: string) => void;
+  /** Called for every streaming event of each model call (for the in-flight activity line). */
+  onModelCall?: (event: ModelCallEvent) => void;
+  /** Called when the agent session is created (for accessing session stats). */
+  onSessionCreated?: (session: AgentSession) => void;
+  /** Called at the end of each agentic turn with the cumulative count. */
+  onTurnEnd?: (turnCount: number) => void;
+  /** Called once per assistant message_end with that message's usage delta. */
+  onAssistantUsage?: (usage: { input: number; output: number; cacheWrite: number }) => void;
+  /** Called when the session successfully compacts. */
+  onCompaction?: (info: CompactionInfo) => void;
+  /** Nesting depth: top-level subagent = 1. */
+  depth?: number;
+  /** Parent agent ID for ownership-scoped nested controls. */
+  parentAgentId?: string;
+  /** Effective inherited nesting cap for this branch. */
+  maxSubagentDepth?: number;
+  /** Config-discovery root inherited by nested launches when it differs from the working directory. */
+  configCwd?: string;
+  /** Project-trust decision for `configCwd`; only meaningful paired with it, set by nested launches. */
+  configProjectTrusted?: boolean;
+  /** Root session id, inherited by nested launches so transcripts stay grouped. */
+  rootSessionId?: string;
+}
+
+interface ResumeOptions {
+  /**
+   * Run the resumed turn detached in the background: return immediately with
+   * the record still "running" (or "queued" at the concurrency limit) and
+   * notify on completion via onComplete, exactly like a background spawn.
+   * Default (false/undefined) runs the resume inline and returns the settled
+   * record — the historical behavior.
+   */
+  isBackground?: boolean;
+  /**
+   * Explicit turn limit for the resumed run: this, else the limit the agent
+   * was spawned with, else its type's own `max_turns`, else the default —
+   * counted afresh from the resume prompt (#368).
+   */
+  maxTurns?: number;
+  /** Called on streaming text and model-call changes during this resume. */
+  onTextDelta?: (delta: string, fullText: string) => void;
+  onModelCall?: (event: ModelCallEvent) => void;
+  /** Called at the end of each resumed turn with the count since the resume. */
+  onTurnEnd?: (turnCount: number) => void;
+  /** Called on tool start/end with activity info (for streaming progress to UI). */
+  onToolActivity?: (activity: ToolActivity) => void;
+  /** Called once per assistant message_end with that message's usage delta. */
+  onAssistantUsage?: (usage: { input: number; output: number; cacheWrite: number }) => void;
+  /** Called when the session successfully compacts. */
+  onCompaction?: (info: CompactionInfo) => void;
+  /**
+   * Background resume only: called synchronously when the run actually starts —
+   * immediately, or later from drainQueue. Callers wire per-run side effects
+   * (output-file streaming) here rather than at the call site, so a resume that
+   * is stopped while still queued never leaves a subscription behind: `abort()`
+   * drops a queued record without reaching `settle()`, which is what would have
+   * torn that subscription down.
+   */
+  onStarted?: () => void;
+}
+
+/** Best-effort ceiling on one child's shutdown handlers, so teardown can't strand a quit. */
+const CHILD_SHUTDOWN_TIMEOUT_MS = 3_000;
+
+/**
+ * Close the extension lifecycle `runAgent` opened with `bindExtensions`, then dispose.
+ *
+ * `AgentSession.dispose()` only calls `ExtensionRunner.invalidate()` — pi emits the event
+ * itself in `AgentSessionRuntime.dispose()` beforehand, and this is the one place that binds
+ * extensions onto a session without going through that path. Without the emit, everything an
+ * extension armed in `session_start` leaks once per spawn, and its next tick throws
+ * `assertActive()` from a bare timer callback — an uncaughtException that kills pi (#242).
+ */
+async function shutdownChildSession(session: AgentSession | undefined): Promise<void> {
+  try {
+    const runner = session?.extensionRunner;
+    // Optional all the way down: on a pi without the getter, or a stubbed session from a
+    // partial `onSessionCreated`, skip the emit — the same degrade as before this fix.
+    if (runner?.hasHandlers?.("session_shutdown")) {
+      // Raced, not awaited outright. `emit` runs every handler serially with no timeout of
+      // its own, and dispose() is reached from pi's own `session_shutdown` with the TUI
+      // already torn down — one hung handler would leave a dead terminal.
+      await Promise.race([
+        runner.emit({ type: "session_shutdown", reason: "quit" }),
+        new Promise<void>(resolve => setTimeout(resolve, CHILD_SHUTDOWN_TIMEOUT_MS).unref()),
+      ]);
+    }
+  } catch { /* a partial session must degrade, not take the teardown down with it */ }
+  // Always, even on timeout: disposal is what this function ultimately exists to do.
+  try { session?.dispose?.(); } catch { /* ignore */ }
+}
+
+export class AgentManager {
+  private agents = new Map<string, AgentRecord>();
+  private cleanupInterval: TimerHandle;
+  /** `CONSUMED_RECORD_TTL_MS` unless `PI_SUBAGENTS_CONSUMED_TTL_MIN` overrides it. */
+  private readonly consumedTtlMs = retentionMs("PI_SUBAGENTS_CONSUMED_TTL_MIN", CONSUMED_RECORD_TTL_MS);
+  /** `UNCONSUMED_RECORD_TTL_MS` unless `PI_SUBAGENTS_UNREAD_TTL_MIN` overrides it. */
+  private readonly unreadTtlMs = retentionMs("PI_SUBAGENTS_UNREAD_TTL_MIN", UNCONSUMED_RECORD_TTL_MS);
+  private onComplete?: OnAgentComplete;
+  private onStart?: OnAgentStart;
+  private onCompact?: OnAgentCompact;
+  private onUsage?: OnAgentUsage;
+  private onSteered?: OnAgentSteered;
+  private maxConcurrent: number;
+  private maxConcurrentForeground = DEFAULT_MAX_CONCURRENT_FOREGROUND;
+  /** Base repos worktrees were created from — so dispose() can prune them all,
+   *  not just the parent repo (caller-supplied cwd can target other repos). */
+  private worktreeRepos = new Set<string>();
+
+  /**
+   * Startup phases, keyed by agent id. `spawn()` still returns synchronously,
+   * but an agent using worktree isolation is not running yet when it does —
+   * copying the repo is an awaited git call. This is what `awaitStartup` hands
+   * callers that must fail their tool call on a startup failure, and what
+   * `waitForAll` waits on while a record is "running" with no `promise` yet.
+   * Entries are dropped once the run is underway, and kept (rejected) after a
+   * startup failure so a late `awaitStartup` still sees it.
+   */
+  private startups = new Map<string, Promise<void>>();
+
+  /**
+   * Evicted agents that can still be reached by name, keyed by handle. Outlives
+   * the 10-minute record cleanup — that timer exists to bound memory, not to
+   * expire a conversation the user might still want — and is cleared alongside
+   * completed records on session start/switch.
+   */
+  private tombstones = new Map<string, AgentTombstone>();
+
+  /** Evicted records by id, oldest first (insertion order). */
+  private evictedNotes = new Map<string, EvictedNote>();
+
+  /**
+   * Agents waiting to start, tagged with the pool they wait on. One queue for
+   * both pools: `drainQueue` picks the earliest entry whose own pool has room,
+   * so neither can head-of-line-block the other, and every removal path
+   * (`abort`, `abortAll`, `dispose`) stays a single filter.
+   *
+   * `release` wakes a caller blocked in `spawnAndWait`, and is fired once the
+   * entry's `start` has SETTLED rather than at drain time: startup is async
+   * now, so releasing earlier would wake the caller before `record.promise`
+   * exists and it would read a still-starting agent as one that never ran.
+   * Removing an entry from this array MUST release it — a queued record has no
+   * promise to await, and pi has no tool-execution timeout to bail the caller
+   * out.
+   */
+  private queue: { id: string; pool: Pool; start: () => Promise<void>; release: () => void }[] = [];
+  /** Number of currently running background agents. */
+  private runningBackground = 0;
+  /** Number of currently running foreground (blocking) agents. */
+  private runningForeground = 0;
+
+  constructor(
+    onComplete?: OnAgentComplete,
+    maxConcurrent = DEFAULT_MAX_CONCURRENT,
+    onStart?: OnAgentStart,
+    onCompact?: OnAgentCompact,
+    onUsage?: OnAgentUsage,
+    onSteered?: OnAgentSteered,
+    private readonly clock: Clock = realClock,
+  ) {
+    this.onComplete = onComplete;
+    this.onStart = onStart;
+    this.onCompact = onCompact;
+    this.onUsage = onUsage;
+    this.onSteered = onSteered;
+    this.maxConcurrent = maxConcurrent;
+    // Sweep finished records past their retention window (see cleanup()).
+    const sweepMs = Math.min(
+      MAX_SWEEP_INTERVAL_MS,
+      Math.max(MIN_SWEEP_INTERVAL_MS, Math.min(this.consumedTtlMs, this.unreadTtlMs)),
+    );
+    this.cleanupInterval = this.clock.setInterval(() => this.cleanup(), sweepMs);
+    this.cleanupInterval.unref?.();
+  }
+
+  /** Update the max concurrent background agents limit. */
+  setMaxConcurrent(n: number) {
+    this.maxConcurrent = Math.max(1, n);
+    // Start queued agents if the new limit allows
+    this.drainQueue();
+  }
+
+  getMaxConcurrent(): number {
+    return this.maxConcurrent;
+  }
+
+  /** Update the max concurrent foreground (blocking) agents limit. 0 = unlimited. */
+  setMaxConcurrentForeground(n: number) {
+    // Floor 0, not 1: unlimited is a meaningful value here and the default.
+    this.maxConcurrentForeground = Math.max(0, n);
+    // Start queued agents if the new limit allows — including everything, when
+    // the limit is cleared back to unlimited mid-run.
+    this.drainQueue();
+  }
+
+  getMaxConcurrentForeground(): number {
+    return this.maxConcurrentForeground;
+  }
+
+  /**
+   * Which pool a spawn is charged to, or undefined for one that is charged to
+   * neither (nested children, detached non-background spawns).
+   *
+   * Nothing here queues when the limit is unset — `poolHasRoom` reports an
+   * unlimited pool as always having room, so that alone is what keeps the
+   * default path identical. The `> 0` guard is belt and braces on top: it also
+   * keeps the counter from churning and the settle path from calling a drain
+   * that would find nothing to do. Both are unobservable, which is why no test
+   * pins them; the observable half — that the default start stays synchronous —
+   * is pinned in `test/foreground-concurrency.test.ts`.
+   */
+  private poolFor(record: AgentRecord): Pool | undefined {
+    if (occupiesPoolSlot(record)) return "background";
+    if (this.maxConcurrentForeground > 0 && occupiesForegroundSlot(record)) return "foreground";
+    return undefined;
+  }
+
+  private poolHasRoom(pool: Pool): boolean {
+    return pool === "background"
+      ? this.runningBackground < this.maxConcurrent
+      : this.maxConcurrentForeground === 0 || this.runningForeground < this.maxConcurrentForeground;
+  }
+
+  /**
+   * Spawn an agent and return its ID immediately (for background use).
+   * If the concurrency limit is reached, the agent is queued.
+   *
+   * The id comes back synchronously, but with `isolation: "worktree"` the agent
+   * is not running yet when it does — the repo copy is an awaited git call.
+   * Callers that must fail a tool call on a startup failure await
+   * `awaitStartup(id)`; everyone else sees it on the record (status "error").
+   */
+  spawn(
+    pi: ExtensionAPI,
+    ctx: ExtensionContext,
+    type: SubagentType,
+    prompt: string,
+    options: SpawnOptions,
+  ): string {
+    // Validate before the queue branch — a queued spawn should fail at the
+    // call, not minutes later at drain. Throw (not warn): programmatic callers
+    // can fix and retry; the RPC layer converts throws into error envelopes.
+    assertValidSpawnCwd(options.cwd);
+
+    const id = randomUUID().slice(0, 17);
+    const abortController = new AbortController();
+    const record: AgentRecord = {
+      id,
+      type,
+      // Owned children — nested, or a workflow's — are filtered out of every
+      // top-level surface, so no handle: nothing can address them and they must
+      // not consume a name a top-level sibling could otherwise take.
+      handle: !isTopLevelAgent(options)
+        ? undefined
+        // A reclaimed handle is used as-is: it belongs to the conversation this
+        // spawn is reopening, and re-deriving it would lose the numbering.
+        : options.reclaim?.handle ?? assignHandle(handleBase(type), this.takenHandles()),
+      description: options.description,
+      // Reclaimed here, or filled in below from `name` — in which case it must
+      // see the handle this record just took, since both come out of the same
+      // namespace.
+      alias: isTopLevelAgent(options) ? options.reclaim?.alias : undefined,
+      // Overwritten below when the spawn is actually queued; a foreground spawn
+      // that queues flips to "queued" there rather than being guessed at here,
+      // since the pool decision needs the finished record.
+      status: options.isBackground ? "queued" : "running",
+      toolUses: 0,
+      startedAt: this.clock.now(),
+      abortController,
+      lifetimeUsage: { input: 0, output: 0, cacheWrite: 0, cost: 0 },
+      compactionCount: 0,
+      // Raw tri-state (not coerced to a boolean): true = background, false =
+      // foreground (has an inline tool-result surface), undefined = caller never
+      // declared it (e.g. a cross-extension RPC spawn). The widget's background-
+      // only filter excludes only explicit `false`, so undefined agents — which
+      // have no inline surface — stay visible instead of vanishing.
+      isBackground: options.isBackground,
+      // Whether anyone is awaiting this agent is a property of the agent, not
+      // of the call that made it — and both settle paths need it long after
+      // `options` has stopped being the interesting object.
+      blocking: options.blocking,
+      invocation: options.invocation,
+      depth: options.depth ?? 1,
+      parentAgentId: options.parentAgentId,
+      workflowId: options.workflowId,
+      maxSubagentDepth: options.maxSubagentDepth,
+      rootSessionId: options.rootSessionId,
+    };
+    this.agents.set(id, record);
+    // After the insert, so `takenHandles()` already counts this record's own
+    // handle — a spawn named after its own type gets `explore-2`, not a
+    // duplicate `explore` that would make resolution ambiguous.
+    if (record.handle !== undefined && record.alias === undefined && options.name !== undefined) {
+      record.alias = assignHandle(handleBase(options.name), this.takenHandles());
+    }
+
+    const args: SpawnArgs = { pi, ctx, type, prompt, options };
+
+    const pool = this.poolFor(record);
+    if (pool !== undefined && !options.bypassQueue && !this.poolHasRoom(pool)) {
+      // Queue it — started when a running agent in the same pool completes.
+      // Idempotent for background (already "queued"); the flip that matters is
+      // a blocking foreground spawn, optimistically marked "running" above.
+      record.status = "queued";
+      // A queued record never reaches startAgent's signal wiring, so arm the
+      // parent abort here or Esc could not release the position.
+      if (!this.armQueuedAbort(id, options.signal)) return id;
+      let release!: () => void;
+      record.startGate = new Promise<void>(resolve => { release = resolve; });
+      this.queue.push({
+        id,
+        pool,
+        start: () => this.launch(id, record, args, pool),
+        release: () => release(),
+      });
+      options.onQueued?.(id, this.queue.filter(e => e.pool === pool).length - 1);
+      return id;
+    }
+
+    this.launch(id, record, args, undefined);
+    return id;
+  }
+
+  /**
+   * Wire a parent abort signal for a record that is about to be QUEUED.
+   * `startAgent` does this for running agents, and a queued record never gets
+   * there, so without this Esc could not release a queue position.
+   *
+   * Returns false when the signal is ALREADY aborted, in which case the record
+   * is stopped here and must not be enqueued: `addEventListener` never fires on
+   * an aborted signal, so a `spawnAndWait` on it would wait forever — pi has no
+   * tool-execution timeout to bail it out.
+   *
+   * The listener is removable (`record.detachParentAbort`): `startAgent`
+   * replaces it with its own, and `sendToBackground` must be able to take it
+   * off, or the caller that armed it could stop a run it no longer owns.
+   */
+  private armQueuedAbort(id: string, signal?: AbortSignal): boolean {
+    if (signal === undefined) return true;
+    if (signal.aborted) {
+      const record = this.agents.get(id);
+      if (record) {
+        record.status = "stopped";
+        record.completedAt = this.clock.now();
+      }
+      return false;
+    }
+    const record = this.agents.get(id);
+    if (record) this.armParentAbort(record, signal);
+    return true;
+  }
+
+  /**
+   * Stop `record` when its caller's `signal` aborts, replacing any listener an
+   * earlier stage armed — at most one is live per record, so one call to
+   * `detachParentAbort` cuts the run loose from its caller.
+   */
+  private armParentAbort(record: AgentRecord, signal: AbortSignal | undefined): void {
+    this.detachParentAbort(record);
+    if (signal === undefined) return;
+    const onParentAbort = () => this.abort(record.id);
+    signal.addEventListener("abort", onParentAbort, { once: true });
+    record.detachParentAbort = () => signal.removeEventListener("abort", onParentAbort);
+  }
+
+  private detachParentAbort(record: AgentRecord): void {
+    record.detachParentAbort?.();
+    record.detachParentAbort = undefined;
+  }
+
+  /** Add `delta` to the running count of `pool`; no pool, no count. */
+  private chargePool(pool: Pool | undefined, delta: 1 | -1): void {
+    if (pool === "background") this.runningBackground += delta;
+    else if (pool === "foreground") this.runningForeground += delta;
+  }
+
+  /**
+   * Kick off an agent's startup and register it under `startups`. The returned
+   * promise never rejects — the failure is delivered through `awaitStartup`,
+   * and to the record.
+   *
+   * @param queuedPool - The pool this start was QUEUED on, or undefined for an
+   *   immediate start. A queue drain can be minutes after `spawn()` returned,
+   *   and nobody is awaiting `awaitStartup` by then, so a failure has to live
+   *   on the record as status "error" — what drainQueue did when the throw was
+   *   still synchronous. An immediate start instead drops the record, exactly
+   *   as the throw out of `spawn()` did: no orphan in `listAgents()`, and the
+   *   handle goes back.
+   */
+  private launch(id: string, record: AgentRecord, args: SpawnArgs, queuedPool: Pool | undefined): Promise<void> {
+    const startup = this.startAgent(id, record, args).then(
+      () => { this.startups.delete(id); },
+      (err) => {
+        this.startups.delete(id);
+        if (queuedPool !== undefined) {
+          // Mirrors settleRun: an inline caller gets this failure as a throw
+          // out of spawnAndWait, so an unconsumed record would ALSO nudge the
+          // session about it — the same failure reported twice.
+          if (queuedPool === "foreground") markConsumed(record, this.clock.now());
+          record.status = "error";
+          record.error = err instanceof Error ? err.message : String(err);
+          record.completedAt = this.clock.now();
+          this.onComplete?.(record);
+        } else {
+          this.agents.delete(id);
+        }
+        // The agent never kept its slot (startAgent gives it back on failure),
+        // so anything queued behind it can go now.
+        this.drainQueue();
+        throw err;
+      },
+    );
+    this.startups.set(id, startup);
+    // Nothing is obliged to await `startups` — swallow the rejection once here
+    // so an unawaited startup can't take the process down, and hand callers
+    // (drainQueue) that swallowed promise.
+    return startup.catch(() => {});
+  }
+
+  /**
+   * Resolves once the agent is actually running, and rejects with the startup
+   * failure (strict worktree isolation) that `spawn()` used to throw before the
+   * repo copy became async. Resolves immediately for an agent that is already
+   * running, still queued, or unknown — so callers can await it unconditionally.
+   *
+   * Call it in the same tick as the `spawn()` it belongs to: a failed startup
+   * takes its record (and this entry) with it, exactly as the throw did.
+   */
+  awaitStartup(id: string): Promise<void> {
+    return this.startups.get(id) ?? Promise.resolve();
+  }
+
+  /** Actually start an agent (called immediately or from queue drain). */
+  private async startAgent(
+    id: string,
+    record: AgentRecord,
+    { pi, ctx, type, prompt, options }: SpawnArgs,
+  ) {
+    // Re-validate a caller-supplied cwd: queued spawns can start minutes after
+    // spawn()'s check, and the directory may be gone by then (TOCTOU). Same
+    // curated errors; drainQueue parks a throw on the record as an error.
+    assertValidSpawnCwd(options.cwd);
+    // Single resolution point for the caller-supplied cwd — the worktree base
+    // repo and both cleanup calls below MUST agree on this value forever.
+    const customCwd = options.cwd ?? undefined; // null (RPC "unset") → undefined
+    const baseCwd = customCwd ?? ctx.cwd;
+
+    // Take the running state — and with it the concurrency slot — BEFORE the
+    // first await. Creating a worktree is an awaited git call, and drainQueue
+    // reads the pool counters synchronously in a loop: incrementing after the
+    // await would let it start every queued agent at once while the first is
+    // still copying its repo. Claiming "running" here also keeps abort() and
+    // abortAll() able to reach an agent whose worktree is still being created.
+    //
+    // The pool is resolved ONCE, here, and recorded for `settleRun`:
+    // `poolFor` reads `maxConcurrentForeground`, which the user can change from
+    // `/agents → Settings` mid-run, so recomputing it at settle time would
+    // decrement a pool this run never charged (counter underflow, limit
+    // silently lifted) or skip the decrement for one it did (leaked slot —
+    // every later blocking spawn queues forever). Only `sendToBackground`
+    // changes it afterwards, moving the counters with it. The two startup exits
+    // below never reach `settleRun`, so they hand the slot back themselves.
+    record.chargedPool = this.poolFor(record);
+    record.status = "running";
+    record.startedAt = this.clock.now();
+    record.startGate = undefined;
+    this.chargePool(record.chargedPool, 1);
+
+    // Worktree isolation: try to create a temporary git worktree. Strict —
+    // fail loud if not possible (no silent fallback to main tree). Done BEFORE
+    // the run is kicked off so a failure doesn't leave a half-running agent.
+    // The project switch is enforced here as well as at the tool boundary
+    // because cross-extension RPC forwards its options unvalidated — a schema
+    // that omits the field can't stop a caller that never saw the schema.
+    // The agent file's `isolation:` is resolved here too, not by each caller:
+    // mentions pass none, and RPC forwards only its caller's value. Workflow
+    // children are the exception, as in Claude Code, whose workflow agent()
+    // reads only the script's option and ignores the definition's.
+    const isolation = options.workflowId !== undefined
+      ? options.isolation
+      : getAgentConfig(type)?.isolation ?? options.isolation;
+    let worktreeCwd: string | undefined;
+    if (isolation === "worktree" && isWorktreeIsolationEnabled()) {
+      let wt: WorktreeInfo;
+      try {
+        wt = await createWorktree(pi, baseCwd, id);
+      } catch (err) {
+        this.chargePool(record.chargedPool, -1);
+        const cause = err instanceof Error ? err.message : String(err);
+        // Only a parent-cwd spawn can be fixed by pointing `cwd` elsewhere.
+        const hint = customCwd === undefined
+          ? " To isolate another repo, pass its path as the `Agent` tool's `cwd`; or omit `isolation`."
+          : " Pass a `cwd` inside a git repo with at least one commit, or omit `isolation`.";
+        throw new Error(`Cannot run with isolation: "worktree": ${cause}.${hint}`);
+      }
+      record.worktree = wt;
+      // workPath preserves subdirectory scoping for caller-supplied cwds: a
+      // cwd deep in a monorepo maps to the same subdir inside the copy, not
+      // the copied repo's root. Plain worktree spawns keep the historical
+      // behavior (agent at the copy's root) — moving them to workPath would
+      // also move .pi config discovery when the parent session sits in a repo
+      // subdirectory, silently dropping extensions/skills.
+      worktreeCwd = customCwd !== undefined ? wt.workPath : wt.path;
+      this.worktreeRepos.add(baseCwd);
+      // The widget's "worktree" tag reads this; only callers that resolve the
+      // agent file themselves seed it.
+      record.invocation ??= {};
+      record.invocation.isolation = "worktree";
+
+      // No longer "running" means a stop landed while the copy was being made
+      // (abort(), abortAll()) — a window that did not exist when creation was
+      // synchronous. The record is already terminal, so launching the run would
+      // burn tokens on work nobody is waiting for: discard the fresh (and by
+      // definition unchanged) worktree instead.
+      if (record.status !== "running") {
+        this.chargePool(record.chargedPool, -1);
+        record.worktreeResult = await cleanupWorktree(pi, baseCwd, wt, options.description);
+        this.drainQueue();
+        return;
+      }
+    }
+
+    this.onStart?.(record);
+
+    // Wire parent abort signal to stop the subagent when the parent is interrupted.
+    // A queued spawn can start minutes after the caller handed us its signal,
+    // by which time it may already be aborted — and `addEventListener` would
+    // never fire, leaving a child the parent can no longer reach.
+    if (options.signal?.aborted) this.abort(id);
+    else this.armParentAbort(record, options.signal);
+    const detach = () => this.detachParentAbort(record);
+
+    const promise = runAgent(ctx, type, prompt, {
+      pi,
+      agentId: id,
+      model: options.model,
+      maxTurns: options.maxTurns,
+      isolated: options.isolated,
+      inheritContext: options.inheritContext,
+      thinkingLevel: options.thinkingLevel,
+      structuredOutput: options.structuredOutput,
+      resumeSessionFile: options.resumeSessionFile,
+      nested: options.parentAgentId !== undefined,
+      workflow: options.workflowId !== undefined,
+      // Worktree wins for the working dir (the agent must run in the copy —
+      // which, with a custom cwd, was created from that target). Config stays
+      // with the parent project when a caller-supplied cwd is in play; it must
+      // stay undefined otherwise so plain worktree runs keep resolving config
+      // (incl. relative extension paths and memory) inside the worktree copy.
+      cwd: worktreeCwd ?? customCwd,
+      // Set iff a worktree was created (see above) — names the directory the
+      // copy came from, so the prompt can tell the agent not to work there.
+      worktreeBase: worktreeCwd ? baseCwd : undefined,
+      configCwd: options.configCwd ?? (customCwd !== undefined ? ctx.cwd : undefined),
+      configProjectTrusted: options.configCwd !== undefined ? options.configProjectTrusted : undefined,
+      signal: record.abortController!.signal,
+      onToolActivity: (activity) => {
+        if (activity.type === "end") record.toolUses++;
+        options.onToolActivity?.(activity);
+      },
+      onTurnEnd: options.onTurnEnd,
+      onTextDelta: options.onTextDelta,
+      onModelCall: options.onModelCall,
+      onAssistantUsage: (usage, message) => {
+        relabelFromResponse(record, message);
+        addUsage(record.lifetimeUsage, usage);
+        this.onUsage?.(record, usage);
+        options.onAssistantUsage?.(usage);
+      },
+      onCompaction: (info) => {
+        record.compactionCount++;
+        this.onCompact?.(record, info);
+        options.onCompaction?.(info);
+      },
+      nestedRuntime: {
+        manager: this,
+        parentAgentId: id,
+        depth: record.depth ?? 1,
+        maxSubagentDepth: record.maxSubagentDepth,
+      },
+      onSessionCreated: (session) => {
+        record.session = session;
+        // Capture now, while the session object exists: after eviction this
+        // path is the only thing that can reopen the conversation, and an
+        // in-memory session reports undefined, which correctly means
+        // "nothing to come back to".
+        // Optional chaining, not defensiveness for its own sake: this is the
+        // only field read off the session at creation, so an older pi or a
+        // stubbed session must degrade to "not resumable" rather than throw
+        // and take the whole spawn down with it.
+        record.sessionFile = session.sessionManager?.getSessionFile?.();
+        // Same reason, different field: the model and thinking level are only
+        // knowable once pi has resolved its defaults and clamped the level to
+        // what the model supports. Writing them back here makes the record
+        // authoritative, so every surface reads one place instead of each
+        // re-deriving "session, else the request" for itself.
+        if (session.model) {
+          record.invocation ??= {};
+          // Read the kept request first: a caller's level survives being clamped
+          // AND, one line later, being replaced by the effective one.
+          const requested = record.invocation.requestedThinking ?? record.invocation.thinking;
+          Object.assign(
+            record.invocation,
+            isVirtualModel(session.model) ? describeVirtualModel(session.model) : describeModel(session.model),
+          );
+          // Guarded for the reason above: a session that reports no level keeps
+          // the request rather than losing it. Overwriting unconditionally would
+          // turn an older or stubbed session into a blank `thinking:` tag, which
+          // is worse than the stale-but-true value it replaced.
+          if (session.thinkingLevel) {
+            record.invocation.thinking = session.thinkingLevel;
+            if (requested && requested !== session.thinkingLevel) {
+              record.invocation.requestedThinking = requested;
+            }
+          }
+        }
+        // Flush any steers that arrived before the session was ready
+        if (record.pendingSteers?.length) {
+          for (const msg of record.pendingSteers) {
+            session.steer(msg).catch(() => {});
+          }
+          record.pendingSteers = undefined;
+        }
+        options.onSessionCreated?.(session);
+      },
+    })
+      .then(async ({ responseText, session, aborted, steered, failure, structuredJson, structuredRetried }) => {
+        this.settleOutcome(record, { text: responseText, failure, aborted, steered });
+        // Kept beside `result`, never inside it: `result` is prose meant for a
+        // reader — it is previewed, transcribed, and appended to below — while
+        // this is a machine-readable payload one caller asked for by schema.
+        record.structuredJson = structuredJson;
+        record.structuredRetried = structuredRetried;
+        record.session = session;
+
+        detach();
+
+        // Final flush of streaming output file
+        if (record.outputCleanup) {
+          try { record.outputCleanup(); } catch { /* ignore */ }
+          record.outputCleanup = undefined;
+        }
+
+        // Clean up worktree if used
+        if (record.worktree) {
+          // The one moment the child's tree still exists and the child is done
+          // writing to it. try/catch, not decoration: a hook that throws must
+          // not leave the worktree behind.
+          if (options.onBeforeWorktreeCleanup) {
+            try {
+              await options.onBeforeWorktreeCleanup(record.worktree.path);
+            } catch { /* ignore — never block cleanup */ }
+          }
+          const wtResult = await cleanupWorktree(pi, baseCwd, record.worktree, options.description);
+          record.worktreeResult = wtResult;
+          if (wtResult.hasChanges && wtResult.branch) {
+            // With a caller-supplied cwd the branch lives in THAT repo, not the
+            // parent session's — say so, or the orchestrator merges in the wrong repo.
+            // Named by its toplevel: the cwd may be a subdirectory of it.
+            const repo = record.worktree.repoRoot;
+            const repoNote = customCwd !== undefined ? ` in \`${repo}\`` : "";
+            // Appended to the prose only. A structured child's caller parses
+            // `structuredJson`, which stays untouched — but `result` is also
+            // what a human reads, so the note still belongs on it.
+            record.result = (record.result ?? "") +
+              `\n\n---\nChanges saved to branch \`${wtResult.branch}\`${repoNote}. Merge with: \`git merge ${wtResult.branch}\`${customCwd !== undefined ? ` (run in \`${repo}\`)` : ""}`;
+          }
+        }
+
+        this.abortOwnedChildren(id);
+
+        this.settleRun(record, true);
+        return responseText;
+      })
+      .catch(async (err) => {
+        // Don't overwrite status if externally stopped via abort()
+        if (record.status !== "stopped") {
+          record.status = "error";
+        }
+        record.error = err instanceof Error ? err.message : String(err);
+        record.completedAt ??= this.clock.now();
+
+        detach();
+
+        // Final flush of streaming output file on error
+        if (record.outputCleanup) {
+          try { record.outputCleanup(); } catch { /* ignore */ }
+          record.outputCleanup = undefined;
+        }
+
+        // Best-effort worktree cleanup on error
+        if (record.worktree) {
+          try {
+            const wtResult = await cleanupWorktree(pi, baseCwd, record.worktree, options.description);
+            record.worktreeResult = wtResult;
+          } catch { /* ignore cleanup errors */ }
+        }
+
+        this.abortOwnedChildren(id);
+
+        this.settleRun(record, false);
+        return "";
+      });
+
+    record.promise = promise;
+
+    // Notify caller that spawn is complete (record is in the map, promise is set).
+    // Called synchronously — onSessionCreated fires asynchronously inside runAgent.
+    // Used by spawnAndWait to let the caller set up output files before streaming
+    // starts. Read off the options, so a spawn that started from a queue drain
+    // still reaches the caller that queued it.
+    options.onSpawned?.(id);
+  }
+
+  /**
+   * The shared tail of both settle paths: release whatever pool slot the run
+   * held, notify, and let the queue drain into the freed slot.
+   *
+   * The decrement lives HERE and nowhere else. `abort()` on a running record
+   * only fires its controller and leaves the run to settle normally, so
+   * decrementing there too would double-free — permanently lifting the limit.
+   *
+   * Foreground agents fire `onComplete` for lifecycle symmetry, with
+   * `resultConsumed` set so the callback skips notifications the inline result
+   * already delivered.
+   *
+   * @param guardCallback swallow a throwing `onComplete` (the success path does;
+   *   the error path historically did not, and keeps not doing so).
+   *
+   * Releases `record.chargedPool` — the pool the run was charged to at start
+   * (or moved to by `sendToBackground`), never a recomputed one, so a mid-run
+   * change to `maxConcurrentForeground` can't make the release disagree with
+   * the acquire.
+   */
+  private settleRun(record: AgentRecord, guardCallback: boolean): void {
+    if (!record.isBackground) markConsumed(record, this.clock.now());
+    // Nobody waits on a settled run; a later foreground resume must not look
+    // like a spawn `sendToBackground` could hand off.
+    record.blocking = false;
+    const pool = record.chargedPool;
+    this.chargePool(pool, -1);
+
+    if (guardCallback) {
+      try { this.onComplete?.(record); } catch { /* ignore completion side-effect errors */ }
+    } else {
+      this.onComplete?.(record);
+    }
+
+    // The isBackground half reproduces the pre-pool condition exactly — a
+    // background settle has always drained, even for a nested child that held
+    // no slot — so that path is unchanged whether or not the foreground pool is
+    // on. The `pool` half only adds the drain a freed FOREGROUND slot needs.
+    // A drain with nothing freed is a no-op anyway, but "no-op" is a claim
+    // about reachability, and matching the old condition needs no such claim.
+    if (record.isBackground || pool !== undefined) this.drainQueue();
+  }
+
+  /**
+   * Stop the nested children a settled parent owns. Nested records are hidden
+   * from the UI and only their owner can consume them, so a child outliving its
+   * parent would burn tokens unseen with no way to reach it. Grandchildren are
+   * covered transitively — each abort lands in that child's own settle path.
+   */
+  private abortOwnedChildren(parentId: string): void {
+    for (const [id, record] of this.agents) {
+      if (record.parentAgentId === parentId) this.abort(id);
+    }
+  }
+
+  /**
+   * Start queued agents up to each pool's concurrency limit.
+   *
+   * `findIndex` on the entry's OWN pool rather than `shift`: with one queue
+   * serving two independent limits, a saturated foreground pool at the head
+   * would otherwise stall every background agent behind it. Taking the earliest
+   * eligible entry keeps FIFO within each pool, which is what callers see.
+   */
+  private drainQueue() {
+    for (;;) {
+      const i = this.queue.findIndex(e => this.poolHasRoom(e.pool));
+      if (i === -1) return;
+      const [next] = this.queue.splice(i, 1);
+      const record = this.agents.get(next.id);
+      // Stale entries (aborted while queued) are not started — but are still
+      // released, since nothing else will.
+      if (!record || record.status !== "queued") { next.release(); continue; }
+      // Detached, and never rejects: a late failure (e.g. strict worktree
+      // isolation) lands on the record inside `launch`, exactly as the
+      // synchronous throw did here before, and draining continues either way.
+      //
+      // The release waits for that startup to SETTLE rather than firing here.
+      // Startup is async now, so a release at drain time would wake a blocked
+      // `spawnAndWait` while `record.promise` was still undefined, and it would
+      // read a perfectly healthy agent as one that never ran.
+      void next.start().then(() => next.release(), () => next.release());
+    }
+  }
+
+  /**
+   * Remove queued entries and wake anyone blocked on them. The single point
+   * that enforces "leaving the queue releases the waiter" — a missed release is
+   * an unbounded hang, not a failed call.
+   */
+  private dequeue(pred: (entry: { id: string; pool: Pool }) => boolean): void {
+    const kept: typeof this.queue = [];
+    for (const entry of this.queue) {
+      if (pred(entry)) entry.release();
+      else kept.push(entry);
+    }
+    this.queue = kept;
+  }
+
+  /**
+   * Spawn an agent and wait for completion (foreground use).
+   * Charged to the foreground pool (`maxConcurrentForeground`), which is
+   * unlimited by default; never to the background one.
+   * Returns { id, record } so callers can access the agent ID.
+   *
+   * @param onSpawned - Called synchronously once the run is kicked off, before
+   *   onSessionCreated fires. Use this to set record.outputFile so
+   *   streamToOutputFile can pick it up.
+   */
+  async spawnAndWait(
+    pi: ExtensionAPI,
+    ctx: ExtensionContext,
+    type: SubagentType,
+    prompt: string,
+    options: Omit<SpawnOptions, "isBackground">,
+    onSpawned?: (id: string) => void,
+  ): Promise<{ id: string; record: AgentRecord }> {
+    // `blocking` is what maxConcurrentForeground bounds, and this is its only
+    // source. onSpawned rides on the options rather than on a field of this
+    // manager: a queued spawn starts at drain time, long after any install/
+    // restore pair around this call would have put the field back — and it now
+    // fires after an await (worktree creation) even on the immediate path.
+    const id = this.spawn(pi, ctx, type, prompt, {
+      ...options,
+      isBackground: false,
+      blocking: true,
+      onSpawned,
+    });
+    const record = this.agents.get(id)!;
+
+    // Queued: nothing to await yet — the promise appears when the drain starts
+    // it. The gate resolves (never rejects) on every path out of the queue,
+    // start and abort alike, so a rejection can never escape into the caller's
+    // tool `execute` and take down pi's whole Promise.all tool batch.
+    if (record.status === "queued") await record.startGate;
+
+    // The run promise only exists once startup is past its awaited repo copy —
+    // without this the call would return before the agent had started at all.
+    // A startup failure (strict worktree isolation) rejects here, which is what
+    // the immediate path owes its caller: the Agent tool reports failure by
+    // letting this throw out of `execute`. A queued spawn's failure landed on
+    // the record instead (nobody was awaiting `startups` at drain time) and is
+    // rethrown below, so the contract is the same either way.
+    await this.awaitStartup(id);
+
+    // undefined when it was aborted while queued, or stopped mid-copy, and so
+    // never ran — the record is already terminal with a completedAt, which is
+    // what the caller renders.
+    if (record.promise) await record.promise;
+
+    // A record that ended "error" without ever getting a promise never ran: the
+    // same startup failure spawn() rethrows on the immediate path (#179). Keep
+    // one contract rather than letting queue pressure decide whether a strict
+    // worktree failure throws or returns as a result.
+    if (record.promise === undefined && record.status === "error") {
+      throw new Error(record.error ?? "Agent failed to start");
+    }
+    return { id, record };
+  }
+
+  /**
+   * Hand a live run to the background without restarting it. The session,
+   * promise, in-flight turn, transcript and worktree all continue under the
+   * same id; what changes is who answers for it. The caller's parent-abort
+   * listener comes off (its Esc no longer reaches the run — `abort()` and
+   * `/agents` still do), the slot moves to the background pool, and the run
+   * then settles through the ordinary background path, since `settleRun`
+   * reads the live `isBackground` and `chargedPool`.
+   *
+   * Releasing the caller is the caller's job — the `Agent` tool races its own
+   * wake-up against `spawnAndWait`. This only refuses what it can see: a run
+   * nobody is blocking on (a background or detached spawn, one already handed
+   * off, or a foreground RESUME — `blocking` is cleared when the spawned run
+   * settles, and a resume never charged the slot this would move), and one
+   * that reports through something other than the session (a nested child, a
+   * workflow's agent).
+   *
+   * The background slot is taken even when the pool is full: suspending a
+   * live run is not an option, so it runs over the limit and queued
+   * background work waits for the count to drop, as with `bypassQueue`.
+   */
+  sendToBackground(id: string): boolean {
+    const record = this.agents.get(id);
+    if (!record?.blocking || !isTopLevelAgent(record) || record.status !== "running" || !record.promise) return false;
+
+    this.detachParentAbort(record);
+    this.chargePool(record.chargedPool, -1);
+    record.chargedPool = "background";
+    this.chargePool(record.chargedPool, 1);
+    record.blocking = false;
+    record.isBackground = true;
+    record.resultConsumed = false;
+    // A freed foreground slot may have unblocked a queued blocking spawn.
+    this.drainQueue();
+    return true;
+  }
+
+  /**
+   * Resume an existing agent session with a new prompt.
+   */
+  async resume(
+    id: string,
+    prompt: string,
+    signal?: AbortSignal,
+    options?: ResumeOptions,
+  ): Promise<AgentRecord | undefined> {
+    const record = this.agents.get(id);
+    if (!record?.session) return undefined;
+
+    // Background resume: settle asynchronously and notify on completion exactly
+    // like a background spawn, returning immediately with the record still
+    // "running" — or "queued" when at the concurrency limit. Previously
+    // run_in_background was ignored on resume (the Agent tool's resume branch
+    // returned before its background branch, and resume() only ever awaited
+    // inline), so a resumed agent always blocked the caller until it finished.
+    if (options?.isBackground) {
+      // Never re-enter a run that is still in flight. Detaching means the caller
+      // gets control back while the record stays "running", so nothing stops the
+      // model from resuming the same agent again. Starting a second run would
+      // overwrite record.abortController — orphaning the live run beyond the
+      // reach of `/agents` stop and abortAll() — double-count the pool slot, and
+      // then reject from session.prompt() with "Agent is already processing",
+      // whose settle path would abort the LIVE run's children and report a
+      // failure for a run that is still going. Refuse instead, leaving the
+      // record untouched; the caller decides whether to wait or steer.
+      if (record.status === "running" || record.status === "queued") return undefined;
+
+      record.isBackground = true;
+      record.resultConsumed = false;
+      record.consumedAt = undefined;
+      record.result = undefined;
+      record.error = undefined;
+      record.completedAt = undefined;
+      record.status = "queued";
+
+      const start = () => this.startResume(id, record, prompt, signal, options);
+      if (occupiesPoolSlot(record) && !this.poolHasRoom("background")) {
+        // At the concurrency limit — queue it, drains when a slot frees. A
+        // detached resume has no inline caller, hence nothing to release. The
+        // queue is shared with spawns, whose startup is async, so entries are
+        // promise-shaped even though a resume starts synchronously; failures
+        // land on the record here, since drainQueue no longer catches.
+        this.queue.push({
+          id,
+          pool: "background",
+          start: async () => {
+            try {
+              start();
+            } catch (err) {
+              record.status = "error";
+              record.error = err instanceof Error ? err.message : String(err);
+              record.completedAt = this.clock.now();
+              this.onComplete?.(record);
+            }
+          },
+          release: () => {},
+        });
+      } else {
+        start();
+      }
+      return record;
+    }
+
+    // Foreground resume: run inline and return the settled record.
+    record.status = "running";
+    record.startedAt = this.clock.now();
+    record.completedAt = undefined;
+    record.result = undefined;
+    record.error = undefined;
+    // Fresh controller, as on the detached path: `abort()` fires whatever
+    // `record.abortController` holds, and the spawn's has long settled — so a
+    // stop from `/agents` or the caller's Esc would mark the record stopped
+    // while the resumed prompt ran on to completion.
+    const abortController = new AbortController();
+    record.abortController = abortController;
+    this.armParentAbort(record, signal);
+    // `addEventListener` never fires on a signal that is already aborted, and
+    // the runner skips the prompt only when ITS signal is — stop now, so that
+    // holds for a resume issued on a call that was cancelled first.
+    if (signal?.aborted) this.abort(id);
+
+    try {
+      this.settleOutcome(record, await resumeAgent(record.session, prompt, {
+        type: record.type,
+        maxTurns: options?.maxTurns ?? record.invocation?.maxTurns,
+        onTurnEnd: options?.onTurnEnd,
+        onTextDelta: options?.onTextDelta,
+        onModelCall: options?.onModelCall,
+        onToolActivity: (activity) => {
+          if (activity.type === "end") record.toolUses++;
+          options?.onToolActivity?.(activity);
+        },
+        onAssistantUsage: (usage, message) => {
+          relabelFromResponse(record, message);
+          addUsage(record.lifetimeUsage, usage);
+          this.onUsage?.(record, usage);
+          options?.onAssistantUsage?.(usage);
+        },
+        onCompaction: (info) => {
+          record.compactionCount++;
+          this.onCompact?.(record, info);
+          options?.onCompaction?.(info);
+        },
+        signal: abortController.signal,
+      }));
+    } catch (err) {
+      this.failResume(record, err);
+    }
+    this.detachParentAbort(record);
+
+    // Same contract as the spawn settle paths: children spawned during the
+    // resumed turn must not outlive it — nothing else can see or reach them.
+    this.abortOwnedChildren(id);
+
+    return record;
+  }
+
+  /**
+   * Write a run's outcome onto its record — a spawn's or a resume's, with one
+   * precedence: an external stop keeps "stopped"; a hard abort (turn limit +
+   * grace) is "aborted"; a failed final turn (a provider error pi resolved
+   * instead of rejecting, #144) is an "error", not a completion with an empty
+   * or stale result; otherwise "steered" or "completed". The run's text stays
+   * available in every case.
+   */
+  private settleOutcome(
+    record: AgentRecord,
+    { text, failure, aborted, steered }: Awaited<ReturnType<typeof resumeAgent>>,
+  ): void {
+    if (record.status !== "stopped") {
+      if (aborted) {
+        record.status = "aborted";
+      } else if (failure) {
+        record.status = "error";
+        record.error = failure;
+      } else {
+        record.status = steered ? "steered" : "completed";
+      }
+    }
+    record.result = text;
+    record.completedAt ??= this.clock.now();
+  }
+
+  /** A resume whose runner rejected: an error, unless an external stop got there first. */
+  private failResume(record: AgentRecord, err: unknown): void {
+    if (record.status !== "stopped") {
+      record.status = "error";
+      record.error = err instanceof Error ? err.message : String(err);
+    }
+    record.completedAt ??= this.clock.now();
+  }
+
+  /**
+   * Start a background resume run: detached, settling and notifying like
+   * startAgent's background path. Invoked immediately, or from drainQueue when
+   * a concurrency slot frees. The session already exists (resume reuses it), so
+   * there is no onSessionCreated to hang per-run wiring off — callers use
+   * `options.onStarted`, which fires on both the immediate and the drained path.
+   */
+  private startResume(
+    id: string,
+    record: AgentRecord,
+    prompt: string,
+    parentSignal: AbortSignal | undefined,
+    options: ResumeOptions,
+  ) {
+    if (!record.session) return;
+
+    record.status = "running";
+    record.startedAt = this.clock.now();
+    if (occupiesPoolSlot(record)) this.runningBackground++;
+    this.onStart?.(record);
+
+    // Fresh abort controller so /agents stop and steering target THIS run rather
+    // than the previous one's settled controller.
+    const abortController = new AbortController();
+    record.abortController = abortController;
+    // Optional, and NOT what the Agent tool passes for a detached resume: a
+    // parent signal aborts on the parent's own interrupt (user Esc), which is
+    // right for a foreground run whose result the caller is awaiting, and wrong
+    // for a detached one — background spawns omit it for exactly this reason.
+    let detachParentSignal: (() => void) | undefined;
+    if (parentSignal) {
+      const onParentAbort = () => this.abort(id);
+      parentSignal.addEventListener("abort", onParentAbort, { once: true });
+      detachParentSignal = () => parentSignal.removeEventListener("abort", onParentAbort);
+    }
+
+    // Per-run side effects (output streaming) — see ResumeOptions.onStarted.
+    // After the record is in its running shape, before the run is kicked off.
+    try { options.onStarted?.(); } catch { /* ignore caller wiring errors */ }
+
+    const settle = () => {
+      detachParentSignal?.();
+      detachParentSignal = undefined;
+      // Final flush of streaming output file
+      if (record.outputCleanup) {
+        try { record.outputCleanup(); } catch { /* ignore */ }
+        record.outputCleanup = undefined;
+      }
+      // Children spawned during the resumed turn must not outlive it.
+      this.abortOwnedChildren(id);
+      if (occupiesPoolSlot(record)) this.runningBackground--;
+      try { this.onComplete?.(record); } catch { /* ignore completion side-effect errors */ }
+      this.drainQueue();
+    };
+
+    const promise = resumeAgent(record.session, prompt, {
+      type: record.type,
+      maxTurns: options.maxTurns ?? record.invocation?.maxTurns,
+      onTurnEnd: options.onTurnEnd,
+      onTextDelta: options.onTextDelta,
+      onModelCall: options.onModelCall,
+      onToolActivity: (activity) => {
+        if (activity.type === "end") record.toolUses++;
+        options.onToolActivity?.(activity);
+      },
+      onAssistantUsage: (usage, message) => {
+        relabelFromResponse(record, message);
+        addUsage(record.lifetimeUsage, usage);
+        this.onUsage?.(record, usage);
+        options.onAssistantUsage?.(usage);
+      },
+      onCompaction: (info) => {
+        record.compactionCount++;
+        this.onCompact?.(record, info);
+        options.onCompaction?.(info);
+      },
+      signal: abortController.signal,
+    })
+      .then((outcome) => {
+        this.settleOutcome(record, outcome);
+        settle();
+        return outcome.text;
+      })
+      .catch((err) => {
+        this.failResume(record, err);
+        settle();
+        return "";
+      });
+
+    record.promise = promise;
+  }
+
+  /**
+   * Send a steering message to an agent from the UI (mirrors the steer_subagent
+   * tool). A live session delivers it now — it interrupts the agent after its
+   * current tool execution and appears as a user message. If the session isn't
+   * ready yet, the message is queued on `pendingSteers` and flushed when the
+   * session is created. Returns false if the agent can't accept steering
+   * (unknown id, or no longer running/queued). Fires `onSteered` on either
+   * accepted path, so the UI viewers, an `@handle` and the RPC all report
+   * the steer the same way without each emitting by hand. The steer tools
+   * bypass this method because they await delivery.
+   */
+  steer(id: string, message: string): boolean {
+    const record = this.agents.get(id);
+    if (!record) return false;
+    if (record.status !== "running" && record.status !== "queued") return false;
+    if (record.session) {
+      record.session.steer(message).catch(() => {});
+    } else {
+      if (!record.pendingSteers) record.pendingSteers = [];
+      record.pendingSteers.push(message);
+    }
+    this.onSteered?.(record, message);
+    return true;
+  }
+
+  getRecord(id: string): AgentRecord | undefined {
+    return this.agents.get(id);
+  }
+
+  /** Handles already in use, so a fresh spawn can pick an unclaimed one. */
+  private takenHandles(): Set<string> {
+    const taken = new Set<string>();
+    for (const record of this.agents.values()) {
+      if (record.handle) taken.add(record.handle);
+      if (record.alias) taken.add(record.alias);
+    }
+    // Tombstones hold their names too: an evicted `@explore` is still
+    // resurrectable, so a later Explore must become `explore-2` rather than
+    // shadowing a conversation the user can still reach.
+    for (const entry of this.tombstones.values()) {
+      taken.add(entry.handle);
+      if (entry.alias) taken.add(entry.alias);
+    }
+    return taken;
+  }
+
+  /**
+   * Resolve an `@name` from the prompt. Matches a top-level agent's handle
+   * case-insensitively, preferring one that can still be steered and otherwise
+   * the most recently started (which is the one a resume should continue), then
+   * falls back to an exact agent id so `@<agentId>` works too.
+   */
+  resolveMention(name: string): MentionResolution | undefined {
+    const wanted = name.toLowerCase();
+    let fallback: AgentRecord | undefined;
+    for (const record of this.agents.values()) {
+      if (record.parentAgentId !== undefined) continue;
+      // Handle and alias share one namespace, so at most one agent answers a
+      // name and it makes no difference which of the two matched.
+      if (record.handle?.toLowerCase() !== wanted && record.alias?.toLowerCase() !== wanted) continue;
+      if (record.status === "running" || record.status === "queued") return { kind: "live", record };
+      if (!fallback || record.startedAt > fallback.startedAt) fallback = record;
+    }
+    if (fallback) return { kind: "live", record: fallback };
+    const byId = this.agents.get(name);
+    if (byId?.parentAgentId === undefined && byId !== undefined) return { kind: "live", record: byId };
+    // Only once nothing live answers: a tombstone is a conversation to reopen,
+    // and reopening one while its record still exists would fork the session.
+    for (const entry of this.tombstones.values()) {
+      if (entry.handle.toLowerCase() === wanted || entry.alias?.toLowerCase() === wanted || entry.id === name) {
+        return { kind: "tombstone", entry };
+      }
+    }
+    return undefined;
+  }
+
+  /**
+   * Forget an evicted agent, by handle. For the case where its session file has
+   * gone: the entry can then only ever fail, while still holding the name
+   * against the type that would otherwise start a fresh agent under it.
+   *
+   * A *successful* resume does not drop its tombstone — the live record it
+   * creates already wins in `resolveMention`, and overwrites the entry in place
+   * when it is itself evicted.
+   */
+  dropTombstone(handle: string): void {
+    this.tombstones.delete(handle);
+  }
+
+  /** Evicted agents whose conversation can still be reopened, newest first. */
+  listTombstones(): AgentTombstone[] {
+    return [...this.tombstones.values()].sort((a, b) => b.completedAt - a.completedAt);
+  }
+
+  listAgents(): AgentRecord[] {
+    return [...this.agents.values()].sort(
+      (a, b) => b.startedAt - a.startedAt,
+    );
+  }
+
+  abort(id: string): boolean {
+    const record = this.agents.get(id);
+    if (!record) return false;
+
+    // Remove from queue if queued. No decrement — the slot was never taken —
+    // and no onComplete, matching what a queued background abort has always
+    // done; a blocking caller learns of the stop from its own tool result.
+    if (record.status === "queued") {
+      this.dequeue(q => q.id === id);
+      record.status = "stopped";
+      record.completedAt = this.clock.now();
+      return true;
+    }
+
+    if (record.status !== "running") return false;
+    record.abortController?.abort();
+    record.status = "stopped";
+    record.completedAt = this.clock.now();
+    return true;
+  }
+
+  /** Dispose a record's session and remove it from the map. */
+  private removeRecord(id: string, record: AgentRecord): void {
+    this.tombstone(record);
+    this.noteEviction(record);
+    const session = record.session;
+    // Detached before the shutdown starts, so the record leaves the map at once and
+    // nothing can observe a session that is half torn down.
+    record.session = undefined;
+    this.agents.delete(id);
+    // A failed startup keeps its (rejected) entry so a late awaitStartup still
+    // sees it; drop it with the record so the map can't grow unbounded.
+    this.startups.delete(id);
+    // Fire-and-forget is right here and only here: this runs from the 60s cleanup timer
+    // and from `clearCompleted()` on session boundaries, with the process staying alive,
+    // so handlers get their full window. The quit path awaits instead — see dispose().
+    void shutdownChildSession(session);
+  }
+
+  /**
+   * Preserve enough of a departing record for `@handle` to reopen its
+   * conversation later. Nothing to keep unless it has both a handle to be
+   * addressed by and a session file to reopen — an in-memory session leaves no
+   * transcript, so the mention would have nothing to continue from.
+   */
+  private tombstone(record: AgentRecord): void {
+    if (!record.handle || !record.sessionFile) return;
+    this.tombstones.set(record.handle, {
+      handle: record.handle,
+      alias: record.alias,
+      id: record.id,
+      type: record.type,
+      description: record.description,
+      sessionFile: record.sessionFile,
+      completedAt: record.completedAt ?? this.clock.now(),
+    });
+    // Bound the memory a long session can accumulate. Oldest first, since the
+    // agent someone still wants to reach is the one they used most recently.
+    while (this.tombstones.size > MAX_TOMBSTONES) {
+      const oldest = [...this.tombstones.values()].reduce((a, b) => (a.completedAt <= b.completedAt ? a : b));
+      this.tombstones.delete(oldest.handle);
+    }
+  }
+
+  /** Remember that `record` existed, so its id still answers with something better than "not found". */
+  private noteEviction(record: AgentRecord): void {
+    const completedAt = record.completedAt ?? this.clock.now();
+    this.evictedNotes.delete(record.id);
+    this.evictedNotes.set(record.id, {
+      id: record.id,
+      type: record.type,
+      description: record.description,
+      status: record.status,
+      evictedAfterMin: Math.max(0, Math.round((this.clock.now() - completedAt) / 60_000)),
+      transcript: record.outputFile ?? record.sessionFile,
+    });
+    while (this.evictedNotes.size > MAX_EVICTED_NOTES) {
+      const oldest = this.evictedNotes.keys().next().value;
+      if (oldest === undefined) break;
+      this.evictedNotes.delete(oldest);
+    }
+  }
+
+  /** What became of an evicted agent, by id; undefined for an id this session never evicted. */
+  getEvictedNote(id: string): EvictedNote | undefined {
+    return this.evictedNotes.get(id);
+  }
+
+  /**
+   * Evict finished records past their window: 10 minutes from the moment the
+   * result was consumed, 60 from completion while nobody has read it. A consumed
+   * record with no stamp (set directly, not through `markConsumed`) is aged from
+   * completion.
+   */
+  private cleanup() {
+    const now = this.clock.now();
+    for (const [id, record] of this.agents) {
+      if (record.status === "running" || record.status === "queued") continue;
+      const completedAt = record.completedAt ?? 0;
+      const expired = record.resultConsumed
+        ? now - (record.consumedAt ?? completedAt) >= this.consumedTtlMs
+        : now - completedAt >= this.unreadTtlMs;
+      if (expired) this.removeRecord(id, record);
+    }
+  }
+
+  /**
+   * Remove all completed/stopped/errored records immediately.
+   * Called on session start/switch so tasks from a prior session don't persist.
+   * Pass skipUnconsumed=true to preserve records the LLM hasn't read yet
+   * (resultConsumed=false) — they will be evicted by the cleanup timer instead (60 minutes).
+   */
+  clearCompleted(skipUnconsumed = false): void {
+    for (const [id, record] of this.agents) {
+      if (record.status === "running" || record.status === "queued") continue;
+      if (skipUnconsumed && !record.resultConsumed) continue;
+      this.removeRecord(id, record);
+    }
+    // Unconditional: both callers are session boundaries (`session_start` and
+    // `session_before_switch`), and `skipUnconsumed` only spares records whose
+    // results the LLM has yet to read — it does not make the sweep partial in
+    // the sense that matters here. A new session means new handles, or
+    // `@explore` would silently reach an agent the user never started. Claude
+    // Code resets its registry on `/clear` for the same reason.
+    this.tombstones.clear();
+    // Ids are only meaningful within the session that issued them.
+    this.evictedNotes.clear();
+  }
+
+  /** Whether any agents are still running or queued. */
+  hasRunning(): boolean {
+    return [...this.agents.values()].some(
+      r => r.status === "running" || r.status === "queued",
+    );
+  }
+
+  /**
+   * Abort everything that cannot outlive its session and let the rest finish.
+   * Survivors are running top-level background agents and their nested children:
+   * they own their own session and report through a notification. Queued agents
+   * would start later against a stale context, foreground agents belong to a
+   * tool call that is being torn down, and workflow children die with their
+   * workflow, so all of those are aborted.
+   *
+   * @returns how many agents keep running.
+   */
+  detach(): number {
+    for (const queued of this.queue) {
+      const record = this.agents.get(queued.id);
+      if (record) {
+        record.status = "stopped";
+        record.completedAt = this.clock.now();
+      }
+    }
+    this.dequeue(() => true);
+    let survivors = 0;
+    for (const record of this.agents.values()) {
+      if (record.status !== "running") continue;
+      const outlives = record.workflowId === undefined && (!!record.isBackground || record.parentAgentId !== undefined);
+      if (outlives) {
+        survivors++;
+        continue;
+      }
+      record.abortController?.abort();
+      record.status = "stopped";
+      record.completedAt = this.clock.now();
+    }
+    return survivors;
+  }
+
+  /** Abort all running and queued agents immediately. */
+  abortAll(): number {
+    let count = 0;
+    // Clear queued agents first
+    for (const queued of this.queue) {
+      const record = this.agents.get(queued.id);
+      if (record) {
+        record.status = "stopped";
+        record.completedAt = this.clock.now();
+        count++;
+      }
+    }
+    this.dequeue(() => true);
+    // Abort running agents
+    for (const record of this.agents.values()) {
+      if (record.status === "running") {
+        record.abortController?.abort();
+        record.status = "stopped";
+        record.completedAt = this.clock.now();
+        count++;
+      }
+    }
+    return count;
+  }
+
+  /** Wait for all running and queued agents to complete (including queued ones). */
+  async waitForAll(): Promise<void> {
+    // Loop because drainQueue respects the concurrency limit — as running
+    // agents finish they start queued ones, which need awaiting too.
+    while (true) {
+      this.drainQueue();
+      const pending: Promise<unknown>[] = [];
+      for (const record of this.agents.values()) {
+        if (record.status !== "running" && record.status !== "queued") continue;
+        // An agent whose worktree is still being created is "running" with no
+        // `promise` yet — without its startup the wait would return too early.
+        const startup = this.startups.get(record.id);
+        if (startup) pending.push(startup);
+        if (record.promise) pending.push(record.promise);
+      }
+      if (pending.length === 0) break;
+      await Promise.allSettled(pending);
+    }
+  }
+
+  /**
+   * @param pi - Needed to run `git worktree prune`, which is async now and so
+   *   cannot be reached through a stored spawn argument at shutdown. Omitting
+   *   it (tests, teardown of a manager that never spawned) skips the prune.
+   */
+  async dispose(pi?: ExtensionAPI): Promise<void> {
+    this.cleanupInterval.cancel();
+    // Clear queue — via dequeue, so anyone blocked in spawnAndWait is woken
+    // rather than left awaiting a gate nothing will ever resolve.
+    this.dequeue(() => true);
+    const sessions = [...this.agents.values()].map(record => record.session);
+    this.agents.clear();
+    this.startups.clear();
+    if (pi) {
+      // Prune any orphaned git worktrees (crash recovery). Detached: dispose runs
+      // on the shutdown path, which cannot wait for git. Started before the awaited
+      // shutdown below rather than after it, so the git calls have that window to
+      // finish in instead of racing the process exit that follows.
+      const prune = (repo: string) => { pruneWorktrees(pi, repo).catch(() => {}); };
+      prune(process.cwd());
+      // Also prune repos that caller-supplied cwds created worktrees in — a clean
+      // exit with in-flight agents would otherwise leave stale registrations there.
+      for (const repo of this.worktreeRepos) prune(repo);
+    }
+    // Awaited, unlike the eviction path: pi awaits this extension's `session_shutdown`
+    // handler and the process exits right after it returns, so anything left unawaited
+    // here never runs at all. Bounded — each call carries its own ceiling, concurrently.
+    await Promise.all(sessions.map(session => shutdownChildSession(session)));
+  }
+}