pi-subagents 0.67.0 → 0.68.0

package/src/runs/background/subagent-wait.ts

@@ -39,6 +39,7 @@
  */
 
 import * as fs from "node:fs";
+import * as path from "node:path";
 import type { AgentToolResult } from "@earendil-works/pi-agent-core";
 import {
 	listBackgroundWorkWakeChannels,
@@ -61,6 +62,7 @@ import {
 	type Usage,
 	type WaitCompletion,
 } from "../../shared/types.ts";
+import { nestedRunScope } from "../shared/nested-events.ts";
 import { formatDuration, shortenPath } from "../../shared/formatters.ts";
 import { toAgentToolUsage } from "../../shared/utils.ts";
 import { collectWaitCompletions } from "./wait-completions.ts";
@@ -103,6 +105,8 @@ export interface SubagentWaitDeps {
 	onUpdate?: (result: AgentToolResult<Details>) => void;
 	asyncDirRoot?: string;
 	resultsDir?: string;
+	/** Root from the child's validated inherited route, not a tool argument. */
+	nestedRootRunId?: string;
 	kill?: (pid: number, signal?: NodeJS.Signals | 0) => boolean;
 	now?: () => number;
 	pollIntervalMs?: number;
@@ -118,6 +122,8 @@ export interface SubagentWaitDeps {
 	failOnFailedRuns?: boolean;
 	/** Internal auto-drain mode surfaces actionable attention as an error. */
 	failOnAttention?: boolean;
+	/** Durable owned supervisor-request barrier used by headless auto-drain. */
+	hasPendingSupervisorRequest?: () => boolean;
 	/** Arm a durable exact-target wait subscription in a long-lived interactive runtime. */
 	subscribe?: (input: { targetKind: "async" | "foreground"; runId: string; requestedId: string; timeoutMs: number }) => { token: string; expiresAt: number };
 	/** Injectable provider protocol surfaces for deterministic tests. */
@@ -264,22 +270,39 @@ function backgroundWorkForSession(deps: SubagentWaitDeps, nowMs: number): Backgr
 	return deps.backgroundWork?.snapshot(sessionId, nowMs) ?? snapshotBackgroundWork(sessionId, nowMs);
 }
 
-/** Queued/running runs from this session, including runs that need attention. */
-function activeRunsForSession(params: SubagentWaitParams, deps: SubagentWaitDeps): AsyncRunSummary[] {
-	const asyncDirRoot = deps.asyncDirRoot ?? DIRS.async;
-	const resultsDir = deps.resultsDir ?? DIRS.results;
-	const runs = listAsyncRuns(asyncDirRoot, {
-		states: [...ACTIVE_STATES],
+export function waitRunScopes(deps: Pick<SubagentWaitDeps, "asyncDirRoot" | "resultsDir" | "nestedRootRunId">): Array<{ asyncDirRoot: string; resultsDir: string }> {
+	return [
+		{ asyncDirRoot: deps.asyncDirRoot ?? DIRS.async, resultsDir: deps.resultsDir ?? DIRS.results },
+		...(deps.nestedRootRunId ? [nestedRunScope(deps.nestedRootRunId)] : []),
+	];
+}
+
+function collectScopedCompletions(terminal: AsyncRunSummary[], deps: SubagentWaitDeps, references: string[]): WaitCompletion[] | undefined {
+	const completions = waitRunScopes(deps).flatMap((scope) => collectWaitCompletions(
+		terminal.filter((run) => path.resolve(path.dirname(run.asyncDir)) === path.resolve(scope.asyncDirRoot)),
+		deps.state, scope.resultsDir, (reference) => references.push(reference),
+	) ?? []);
+	return completions.length ? completions : undefined;
+}
+
+/** Immediate-session runs only; named lookups can also recover already-terminal results. */
+function runsForSession(params: SubagentWaitParams, deps: SubagentWaitDeps, activeOnly = true): AsyncRunSummary[] {
+	const runs = waitRunScopes(deps).flatMap(({ asyncDirRoot, resultsDir }) => listAsyncRuns(asyncDirRoot, {
+		states: activeOnly ? [...ACTIVE_STATES] : undefined,
 		sessionId: deps.state.currentSessionId ?? undefined,
 		resultsDir,
 		kill: deps.kill,
 		now: deps.now,
 		includeNested: false,
 		...(params.id ? { runId: params.id } : {}),
-	});
+	}));
 	return params.id ? runs.filter((run) => matchesId(run, params.id!)) : runs;
 }
 
+function activeRunsForSession(params: SubagentWaitParams, deps: SubagentWaitDeps): AsyncRunSummary[] {
+	return runsForSession(params, deps);
+}
+
 /** Runs (from the initial set) currently flagged needs_attention, for reporting. */
 function attentionRunsForSession(params: SubagentWaitParams, deps: SubagentWaitDeps, initialIds: Set<string>): AsyncRunSummary[] {
 	return activeRunsForSession(params, deps).filter((run) => needsAttention(run) && initialIds.has(run.id));
@@ -287,9 +310,7 @@ function attentionRunsForSession(params: SubagentWaitParams, deps: SubagentWaitD
 
 /** Exact initial runs in any state, for the final summary. */
 function runsForIds(runIds: Iterable<string>, deps: SubagentWaitDeps): AsyncRunSummary[] {
-	const asyncDirRoot = deps.asyncDirRoot ?? DIRS.async;
-	const resultsDir = deps.resultsDir ?? DIRS.results;
-	return [...runIds].flatMap((runId) => listAsyncRuns(asyncDirRoot, {
+	return waitRunScopes(deps).flatMap(({ asyncDirRoot, resultsDir }) => [...runIds].flatMap((runId) => listAsyncRuns(asyncDirRoot, {
 		sessionId: deps.state.currentSessionId ?? undefined,
 		resultsDir,
 		kill: deps.kill,
@@ -297,7 +318,7 @@ function runsForIds(runIds: Iterable<string>, deps: SubagentWaitDeps): AsyncRunS
 		includeNested: false,
 		runId,
 		exactRunId: true,
-	}));
+	})));
 }
 
 function summarizeTerminalRuns(runs: AsyncRunSummary[], providerFinishedCount = 0): string {
@@ -377,6 +398,49 @@ function windowElapsedResult(
 	};
 }
 
+function supervisorYieldResult(
+	activeRunIds: string[],
+	activeProviderItems: readonly RegisteredBackgroundWorkItem[] = [],
+): AgentToolResult<Details> {
+	return {
+		content: [{ type: "text", text: "Wait yielded for a pending supervisor request. Background work remains active and will continue after the supervisor reply." }],
+		details: {
+			mode: "management",
+			results: [],
+			wait: {
+				reason: "supervisor_request",
+				timedOut: false,
+				activeRunIds,
+				activeProviderItems: activeProviderItems.map(({ provider, id }) => ({ provider, id })),
+			},
+		},
+	};
+}
+
+interface InitialWaitScope {
+	asyncRunIds: Set<string>;
+	foregroundRuns: Array<{ runId: string; sessionId?: string; detachedIndices: Set<number> }>;
+	providerIds: Set<string>;
+}
+
+function supervisorYieldForScope(scope: InitialWaitScope, params: SubagentWaitParams, deps: SubagentWaitDeps, nowMs: number): AgentToolResult<Details> {
+	try {
+		const activeAsyncIds = new Set(activeRunsForSession(params.id ? { id: params.id } : {}, deps).map((run) => run.id));
+		const activeRunIds = new Set([...scope.asyncRunIds].filter((id) => activeAsyncIds.has(id)));
+		for (const initial of scope.foregroundRuns) {
+			const current = deps.state.foregroundRuns?.get(initial.runId);
+			if (!current || current.sessionId !== initial.sessionId) continue;
+			if (current.children.some((child) => initial.detachedIndices.has(child.index) && child.status === "detached")) activeRunIds.add(initial.runId);
+		}
+		const activeProviderItems = scope.providerIds.size > 0
+			? backgroundWorkForSession(deps, nowMs).items.filter((item) => scope.providerIds.has(backgroundWorkIdentity(item)))
+			: [];
+		return supervisorYieldResult([...activeRunIds], activeProviderItems);
+	} catch (error) {
+		return result(error instanceof Error ? error.message : String(error), true);
+	}
+}
+
 /** Build the live status shown while async work keeps bg_wait blocked. */
 function asyncWaitUpdate(runs: AsyncRunSummary[], providerCount: number, elapsedMs: number): AgentToolResult<Details> {
 	const activity = runs.flatMap((run) => {
@@ -499,6 +563,7 @@ async function waitForDetachedForegroundRun(
 	now: () => number,
 	pollIntervalMs: number,
 	timeoutMs: number,
+	supervisorYield: () => AgentToolResult<Details>,
 ): Promise<AgentToolResult<Details>> {
 	const initialDetachedIndices = new Set(run.children.filter((child) => child.status === "detached").map((child) => child.index));
 	while (true) {
@@ -509,6 +574,7 @@ async function waitForDetachedForegroundRun(
 		if (!current || current.sessionId !== run.sessionId) {
 			return result(`Remembered foreground run "${run.runId}" disappeared before a terminal child result was recorded. Completion cannot be confirmed; do not launch a replacement without checking the originating child session.`, true);
 		}
+		if (deps.hasPendingSupervisorRequest?.()) return supervisorYield();
 		const pending = current.children.filter((child) => initialDetachedIndices.has(child.index) && child.status === "detached");
 		const attention = foregroundChildrenNeedingAttention(current, initialDetachedIndices);
 		if (attention.length > 0) return formatForegroundAttention(current, attention, now() - startedAt);
@@ -541,11 +607,13 @@ async function waitForSessionDetachedForegroundRuns(
 	now: () => number,
 	pollIntervalMs: number,
 	timeoutMs: number,
+	supervisorYield: () => AgentToolResult<Details>,
 ): Promise<AgentToolResult<Details>> {
 	const texts: string[] = [];
 	for (const run of runs) {
-		const one = await waitForDetachedForegroundRun(run, signal, deps, startedAt, now, pollIntervalMs, timeoutMs);
+		const one = await waitForDetachedForegroundRun(run, signal, deps, startedAt, now, pollIntervalMs, timeoutMs, supervisorYield);
 		if (one.isError) return one;
+		if (one.details.wait?.reason === "supervisor_request") return one;
 		if (one.details.wait?.reason === "window_elapsed") {
 			const activeRunIds = runs.filter((initial) => {
 				const current = deps.state.foregroundRuns?.get(initial.runId);
@@ -586,6 +654,7 @@ export async function waitForSubagents(
 		? params.timeoutMs
 		: deps.defaultTimeoutMs ?? DEFAULT_TIMEOUT_MS;
 	const startedAt = now();
+	const sessionId = deps.state.currentSessionId;
 	const waitForAll = params.id ? true : params.all === true;
 	if (params.nonBlocking && !params.id) {
 		return result("Non-blocking wait subscriptions require id so the registration can bind one exact run identity.", true);
@@ -598,7 +667,7 @@ export async function waitForSubagents(
 	let foreground: ForegroundResumeRun[];
 	let providerSnapshot: BackgroundWorkSnapshot;
 	try {
-		active = activeRunsForSession(params, deps);
+		active = runsForSession(params, deps, !params.id);
 		foreground = activeDetachedForegroundRuns(params, deps).map((run) => ({
 			...run,
 			children: run.children.map((child) => ({ ...child })),
@@ -608,6 +677,7 @@ export async function waitForSubagents(
 		return result(error instanceof Error ? error.message : String(error), true);
 	}
 
+	let selectedForeground: ForegroundResumeRun | undefined;
 	if (params.id) {
 		const candidates = [
 			...active.map((run) => ({ kind: "async" as const, id: run.id, run })),
@@ -616,9 +686,21 @@ export async function waitForSubagents(
 		const exact = candidates.filter((candidate) => candidate.id === params.id);
 		const matches = exact.length > 0 ? exact : candidates;
 		if (matches.length > 1) {
-			return result(`Ambiguous subagent run id prefix "${params.id}" matched ${matches.length} active runs: ${matches.map((candidate) => candidate.id).join(", ")}. Pass a longer id.`, true);
+			return result(`Ambiguous subagent run id prefix "${params.id}" matched ${matches.length} runs: ${matches.map((candidate) => candidate.id).join(", ")}. Pass a longer id.`, true);
 		}
 		const selected = matches[0];
+		if (selected?.kind === "async" && !ACTIVE_STATES.includes(selected.run.state)) {
+			try {
+				const references: string[] = [];
+				const completions = collectScopedCompletions([selected.run], deps, references);
+				return result(
+					`Run "${selected.id}" is terminal. Outcome: ${summarizeTerminalRuns([selected.run])}.${formatCompletionRecovery(completions)}${references.length ? `\n${references.join("\n")}` : ""}`,
+					deps.failOnFailedRuns === true && (selected.run.state === "failed" || selected.run.state === "partial"), completions,
+				);
+			} catch (error) {
+				return result(error instanceof Error ? error.message : String(error), true);
+			}
+		}
 		if (selected && params.nonBlocking) {
 			if (!deps.subscribe) {
 				return result("Non-blocking wait subscriptions require a long-lived interactive subagent runtime; this runtime can only use blocking bg_wait calls.", true);
@@ -630,30 +712,47 @@ export async function waitForSubagents(
 				return result(error instanceof Error ? error.message : String(error), true);
 			}
 		}
-		if (selected?.kind === "foreground") {
-			return waitForDetachedForegroundRun(selected.run, signal, deps, startedAt, now, pollIntervalMs, timeoutMs);
-		}
+		selectedForeground = selected?.kind === "foreground" ? selected.run : undefined;
 		active = selected?.kind === "async" ? [selected.run] : [];
 	}
 
 	let providerActive = providerSnapshot.items;
+	const scopedForeground = selectedForeground ? [selectedForeground] : params.id ? [] : foreground;
+	const initialAsyncIds = new Set(active.map((run) => run.id));
+	const initialProviderIds = new Set(providerActive.map(backgroundWorkIdentity));
+	const initialScope: InitialWaitScope = {
+		asyncRunIds: initialAsyncIds,
+		foregroundRuns: scopedForeground.map((run) => ({
+			runId: run.runId,
+			sessionId: run.sessionId,
+			detachedIndices: new Set(run.children.filter((child) => child.status === "detached").map((child) => child.index)),
+		})),
+		providerIds: initialProviderIds,
+	};
+	const supervisorYield = () => supervisorYieldForScope(initialScope, params, deps, now());
+	if (selectedForeground) {
+		return waitForDetachedForegroundRun(selectedForeground, signal, deps, startedAt, now, pollIntervalMs, timeoutMs, supervisorYield);
+	}
 	if (active.length === 0 && providerActive.length === 0) {
 		if (waitForAll && !params.id && foreground.length > 0) {
-			return waitForSessionDetachedForegroundRuns(foreground, signal, deps, startedAt, now, pollIntervalMs, timeoutMs);
+			return waitForSessionDetachedForegroundRuns(foreground, signal, deps, startedAt, now, pollIntervalMs, timeoutMs, supervisorYield);
 		}
 		return result(params.id
 			? `No active run matched "${params.id}". Nothing to wait for.`
 			: "No active async runs or registered provider work in this session. Nothing to wait for.");
 	}
 	const waitParams = params.id ? { ...params, id: active[0]!.id } : params;
-	const initialAsyncIds = new Set(active.map((run) => run.id));
-	const initialProviderIds = new Set(providerActive.map(backgroundWorkIdentity));
 	const initialProviderNames = new Set(providerActive.map((item) => item.provider));
 	const initialCount = initialAsyncIds.size + initialProviderIds.size;
 	const stopOnAttention = params.stopOnAttention ?? deps.stopOnAttention !== false;
 	let attention = active.filter((run) => needsAttention(run));
+	let supervisorBarrier = false;
 
 	const isDone = (): boolean => {
+		if (deps.hasPendingSupervisorRequest?.()) {
+			supervisorBarrier = true;
+			return true;
+		}
 		if (attention.some((run) => initialAsyncIds.has(run.id) && (stopOnAttention || hasSupervisorTool(run)))) return true;
 		const activeAsyncIds = new Set(active.map((run) => run.id));
 		const activeProviderIds = new Set(providerActive.map(backgroundWorkIdentity));
@@ -685,6 +784,7 @@ export async function waitForSubagents(
 		}
 		try {
 			await waitForWake(pollIntervalMs, signal, deps);
+			if (deps.state.currentSessionId !== sessionId) return result("Wait stopped because the active session changed.", true);
 			active = activeRunsForSession(waitParams, deps);
 			attention = attentionRunsForSession(waitParams, deps, initialAsyncIds);
 			providerSnapshot = params.id ? providerSnapshot : backgroundWorkForSession(deps, now());
@@ -698,12 +798,16 @@ export async function waitForSubagents(
 			return result(error instanceof Error ? error.message : String(error), true);
 		}
 	}
+	if (supervisorBarrier) {
+		return supervisorYield();
+	}
 
 	let terminalSummary: string;
 	let finishedAsyncCount: number;
 	let failedAsyncCount: number;
 	let completions: WaitCompletion[] | undefined;
 	let resumeGuidance = "";
+	const references: string[] = [];
 	const activeProviderIds = new Set(providerActive.map(backgroundWorkIdentity));
 	const providerFinishedCount = [...initialProviderIds].filter((id) => !activeProviderIds.has(id)).length;
 	try {
@@ -713,7 +817,7 @@ export async function waitForSubagents(
 		failedAsyncCount = terminal.filter((run) => run.state === "failed" || run.state === "partial").length;
 		terminalSummary = summarizeTerminalRuns(terminal, providerFinishedCount);
 		resumeGuidance = formatResumeFirstFailedRunsNote(terminal);
-		completions = collectWaitCompletions(terminal, deps.state, deps.resultsDir ?? DIRS.results);
+		completions = collectScopedCompletions(terminal, deps, references);
 	} catch (error) {
 		return result(error instanceof Error ? error.message : String(error), true);
 	}
@@ -729,13 +833,14 @@ export async function waitForSubagents(
 		+ providerActive.filter((item) => initialProviderIds.has(backgroundWorkIdentity(item))).length;
 	const elapsed = formatDuration(now() - startedAt);
 	const outcome = terminalSummary ? ` Outcome: ${terminalSummary}.` : "";
-	const recoveryNote = formatCompletionRecovery(completions);
+	const recoveryNote = formatCompletionRecovery(completions) + (references.length ? `\n${references.join("\n")}\n` : "");
 
 	if (waitForAll) {
 		const foregroundResult = !params.id && foreground.length > 0 && relevantAttention.length === 0
-			? await waitForSessionDetachedForegroundRuns(foreground, signal, deps, startedAt, now, pollIntervalMs, timeoutMs)
+			? await waitForSessionDetachedForegroundRuns(foreground, signal, deps, startedAt, now, pollIntervalMs, timeoutMs, supervisorYield)
 			: undefined;
 		if (foregroundResult?.isError) return foregroundResult;
+		if (foregroundResult?.details.wait?.reason === "supervisor_request") return foregroundResult;
 		if (foregroundResult?.details.wait?.reason === "window_elapsed") return foregroundResult;
 		const foregroundNote = foregroundResult
 			? `\n${foregroundResult.content.map((part) => part.type === "text" ? part.text : "").join("\n")}`

package/src/runs/background/wait-completions.ts

@@ -2,7 +2,7 @@ import * as fs from "node:fs";
 import type { ArtifactPaths, SubagentState, Usage, WaitCompletion, WaitCompletionChild } from "../../shared/types.ts";
 import type { AsyncRunSummary } from "./async-status.ts";
 import { readCompletionReplay, writeCompletionReplay } from "./completion-replay.ts";
-import { fallbackResultPayloadPathForSessionRun, resultFilePath, resultPayloadPathForSessionRun } from "./result-files.ts";
+import { fallbackResultPayloadPathForSessionRun, resultFilePath, resultPayloadMatchesSessionRun, resultPayloadPathForSessionRun } from "./result-files.ts";
 import { parseWorkflowChildSummary } from "../../workflows/workflow-child-summary.ts";
 import { projectTimeoutRecovery } from "../shared/mutation-evidence.ts";
 
@@ -44,6 +44,27 @@ function errorMessage(error: unknown): string {
 
 const STRUCTURED_OUTPUT_INLINE_LIMIT_BYTES = 4 * 1024;
 
+function waitResultPayloadCandidate(resultsDir: string, run: AsyncRunSummary): string {
+	const publicPath = resultFilePath(resultsDir, run.id);
+	if (!run.sessionId) return publicPath;
+	try {
+		const indexedPath = resultPayloadPathForSessionRun(resultsDir, run.sessionId, run.id);
+		return indexedPath ?? publicPath;
+	} catch (error) {
+		if (!isAccessDenied(error)) throw error;
+		const pendingPath = fallbackResultPayloadPathForSessionRun(resultsDir, run.sessionId, run.id);
+		return pendingPath ?? publicPath;
+	}
+}
+
+function readWaitResultPayload(candidatePath: string, run: AsyncRunSummary): Record<string, unknown> | undefined {
+	// A terminal run without session ownership cannot safely claim a payload.
+	if (!run.sessionId) return undefined;
+	const payload: unknown = JSON.parse(fs.readFileSync(candidatePath, "utf-8"));
+	if (!resultPayloadMatchesSessionRun(payload, run.sessionId, run.id)) return undefined;
+	return payload as Record<string, unknown>;
+}
+
 export function projectStructuredOutput(value: unknown): unknown {
 	if (value === undefined) return undefined;
 	const serialized = JSON.stringify(value);
@@ -118,8 +139,9 @@ export function toWaitCompletion(data: Record<string, unknown>, runId: string):
 /**
  * Record a consumed terminal payload for later surfacing by bg_wait, pruning
  * stale entries with the same TTL that dedupes completion notifications. The result
- * file is deleted after delivery, so this record is the only in-process source once
- * the watcher has consumed it.
+ * file is deleted after durable replay succeeds, so this record is the in-process
+ * source once the watcher has consumed it. Payload ownership must be explicit and
+ * agree with persistence ownership. Returns whether that replay is durable.
  */
 export function recordWaitCompletion(
 	state: SubagentState,
@@ -128,7 +150,10 @@ export function recordWaitCompletion(
 	now: number,
 	ttlMs: number,
 	persistence?: { resultsDir: string; sessionId: string },
-): void {
+): boolean {
+	const sessionId = asNonEmptyString(data.sessionId);
+	if (!sessionId || !resultPayloadMatchesSessionRun(data, sessionId, runId)) return false;
+	if (persistence && persistence.sessionId !== sessionId) return false;
 	const store = state.completedResults ??= new Map();
 	for (const [key, entry] of store) {
 		if (now - entry.seenAt > ttlMs) store.delete(key);
@@ -148,7 +173,8 @@ export function recordWaitCompletion(
 			console.error(`Failed to persist completion replay for '${runId}':`, error);
 		}
 	}
-	store.set(runId, { seenAt: now, completion });
+	store.set(runId, { sessionId, seenAt: now, completion });
+	return completion.archivePath !== undefined;
 }
 
 /**
@@ -157,37 +183,59 @@ export function recordWaitCompletion(
  * so a direct read never observes a torn write; the read is deliberately read-only —
  * the watcher owns notification and cleanup.
  */
-export function collectWaitCompletions(terminal: AsyncRunSummary[], state: SubagentState, resultsDir: string): WaitCompletion[] | undefined {
+export function collectWaitCompletions(terminal: AsyncRunSummary[], state: SubagentState, resultsDir: string, onReference?: (text: string) => void): WaitCompletion[] | undefined {
 	if (terminal.length === 0) return undefined;
 	const completions: WaitCompletion[] = [];
+	const add = (completion: WaitCompletion, resultPath?: string): void => {
+		completions.push(completion);
+		// Tool details are not model context. Publish the evidence address in content
+		// without copying potentially unbounded reviewer output into every wait.
+		const reference = completion.archivePath ?? resultPath;
+		if (reference) onReference?.(`Result [${completion.runId}]: ${reference}`);
+	};
 	for (const run of terminal) {
 		const recorded = state.completedResults?.get(run.id);
-		if (recorded) {
-			completions.push(recorded.completion);
+		if (recorded && recorded.sessionId === run.sessionId) {
+			if (recorded.completion.archivePath) {
+				add(recorded.completion);
+				continue;
+			}
+			try {
+				const candidatePath = waitResultPayloadCandidate(resultsDir, run);
+				if (!readWaitResultPayload(candidatePath, run)) {
+					const replay = readCompletionReplay(resultsDir, run.id, { sessionId: run.sessionId });
+					add(replay?.completion ?? recorded.completion);
+					continue;
+				}
+				add(recorded.completion, candidatePath);
+				continue;
+			} catch (error) {
+				if (errorCode(error) !== "ENOENT") throw error;
+			}
+			const replay = readCompletionReplay(resultsDir, run.id, { sessionId: run.sessionId });
+			add(replay?.completion ?? recorded.completion);
 			continue;
 		}
-		const publicResultPath = resultFilePath(resultsDir, run.id);
-		let resultPath = publicResultPath;
+		let candidatePath: string;
 		try {
-			resultPath = run.sessionId
-				? resultPayloadPathForSessionRun(resultsDir, run.sessionId, run.id) ?? publicResultPath
-				: publicResultPath;
+			candidatePath = waitResultPayloadCandidate(resultsDir, run);
 		} catch (error) {
-			if (!isAccessDenied(error) || !run.sessionId) throw error;
-			try {
-				resultPath = fallbackResultPayloadPathForSessionRun(resultsDir, run.sessionId, run.id) ?? publicResultPath;
-			} catch (fallbackError) {
-				throw new Error(`Failed to read subagent result '${publicResultPath}': ${errorMessage(fallbackError)}`, {
-					cause: fallbackError instanceof Error ? fallbackError : undefined,
-				});
-			}
+			const publicResultPath = resultFilePath(resultsDir, run.id);
+			throw new Error(`Failed to read subagent result '${publicResultPath}': ${errorMessage(error)}`, {
+				cause: error instanceof Error ? error : undefined,
+			});
 		}
 		try {
-			const raw = JSON.parse(fs.readFileSync(resultPath, "utf-8")) as Record<string, unknown>;
-			completions.push(toWaitCompletion(raw, run.id));
+			const payload = readWaitResultPayload(candidatePath, run);
+			if (!payload) {
+				const replay = readCompletionReplay(resultsDir, run.id, { sessionId: run.sessionId });
+				if (replay) add(replay.completion);
+				continue;
+			}
+			add(toWaitCompletion(payload, run.id), candidatePath);
 		} catch (error) {
 			if (errorCode(error) !== "ENOENT") {
-				throw new Error(`Failed to read subagent result '${resultPath}': ${errorMessage(error)}`, {
+				throw new Error(`Failed to read subagent result '${candidatePath}': ${errorMessage(error)}`, {
 					cause: error instanceof Error ? error : undefined,
 				});
 			}
@@ -195,13 +243,13 @@ export function collectWaitCompletions(terminal: AsyncRunSummary[], state: Subag
 			// read. Prefer its in-memory record, then the durable replay written before
 			// result cleanup so watcher reloads do not lose completion details.
 			const late = state.completedResults?.get(run.id);
-			if (late) {
-				completions.push(late.completion);
+			if (late && late.sessionId === run.sessionId) {
+				add(late.completion);
 				continue;
 			}
 			try {
 				const replay = readCompletionReplay(resultsDir, run.id, { sessionId: run.sessionId });
-				if (replay) completions.push(replay.completion);
+				if (replay) add(replay.completion);
 			} catch (replayError) {
 				throw new Error(`Failed to read completion replay for '${run.id}': ${errorMessage(replayError)}`, {
 					cause: replayError instanceof Error ? replayError : undefined,

package/src/runs/background/wait-subscriptions.ts

@@ -44,6 +44,7 @@ export interface ArmWaitSubscriptionInput {
 }
 
 export interface WaitSubscriptionManager {
+	start(): void;
 	arm(input: ArmWaitSubscriptionInput): WaitSubscriptionRecord;
 	restore(): void;
 	reconcile(): void;
@@ -109,6 +110,7 @@ export function createWaitSubscriptionManager(
 	state.waitSubscriptions = subscriptions;
 	const unresolvedRestoredForegroundTokens = new Set<string>();
 	let disposed = false;
+	let interval: ReturnType<typeof setInterval> | undefined;
 	let lastForeignSweepAt = 0;
 
 	/**
@@ -283,10 +285,13 @@ export function createWaitSubscriptionManager(
 		SUBAGENT_RESULT_INTERCOM_EVENT,
 	];
 	const unsubscribes = wakeChannels.map((channel) => pi.events.on(channel, reconcile));
-	const interval = setInterval(reconcile, options.pollIntervalMs ?? RECONCILE_INTERVAL_MS);
-	interval.unref?.();
 
 	return {
+		start() {
+			if (disposed || interval) return;
+			interval = setInterval(reconcile, options.pollIntervalMs ?? RECONCILE_INTERVAL_MS);
+			interval.unref?.();
+		},
 		arm(input) {
 			const sessionId = state.currentSessionId;
 			if (!sessionId) throw new Error("A wait subscription requires an active session identity.");
@@ -338,7 +343,8 @@ export function createWaitSubscriptionManager(
 		dispose() {
 			if (disposed) return;
 			disposed = true;
-			clearInterval(interval);
+			if (interval) clearInterval(interval);
+			interval = undefined;
 			for (const unsubscribe of unsubscribes) {
 				try { unsubscribe(); } catch { /* best effort */ }
 			}

package/src/runs/background/wait-tool.ts

@@ -11,14 +11,15 @@ export function registerWaitTool(
 	enabled = resolveWaitToolConfig().enabled,
 	subscriptions?: Pick<WaitSubscriptionManager, "arm">,
 	defaultTimeoutMs?: number,
+	child?: { nestedRootRunId?: string },
 ): void {
 	const description = `Wait for background, provider, or detached work that has no native completion notification, then return.
 
-Ordinary async subagent runs already notify this session natively when they complete or need attention. In an interactive chat, return control instead of calling this merely to wait. Use this tool for provider jobs, remembered detached foreground runs, or other background work without a native notification path. Headless runs auto-drain current-session subagent work at agent_end; use this tool only when the current turn must receive non-notifying background work results.
+${child ? "This child runtime does not install the root session's native completion notifier. Use blocking bg_wait to collect your owned descendants during this turn, then read the returned result references before synthesizing findings. Automatic draining at agent_end keeps owned work alive but does not synthesize its results." : "Ordinary async subagent runs already notify this session natively when they complete or need attention. In an interactive chat, return control instead of calling this merely to wait. Use this tool for provider jobs, remembered detached foreground runs, or other background work without a native notification path. Headless runs auto-drain current-session subagent work at agent_end; use this tool only when the current turn must receive non-notifying background work results."}
 
 • { } — return when the first initially active async run or registered provider item finishes, or when a subagent needs attention.
 • { all: true } — wait for every async run, provider item, and remembered detached foreground descendant that was active when the call began.
-• { id: "..." } — wait for one async or remembered detached foreground subagent run (id or prefix).
+• { id: "..." } — wait for one async or remembered detached foreground subagent run (id or prefix). Named async runs that already finished return their terminal result references.
 • { id: "...", nonBlocking: true } — resolve the prefix once, persist an exact-run wake subscription, and return immediately. Use this for detached work without native completion delivery; the originating interactive session wakes on completion, failure, attention, reconciliation failure, or timeout.
 • { stopOnAttention: false } — for blocking waits only, keep waiting through idle or long-thinking attention; supervisor/contact requests still stop the wait.
 • { timeoutMs: 600000 } — stop waiting after N ms; active work keeps running. Omitted values use waitTool.defaultTimeoutMs, then 30 minutes. Window expiry returns a non-error window_elapsed result with active work identities.
@@ -26,6 +27,7 @@ Ordinary async subagent runs already notify this session natively when they comp
 Non-blocking subscriptions are visible in subagent status and differ from disabling waitTool: waitTool.enabled=false returns immediately without registering any future wake. Provider jobs are session-scoped and identified exactly, so replacing one job with another cannot hide a completion. Provider extensions must be explicitly loaded in this process. In a child agent, keep \`bg_wait\` in the child tool allowlist and load each provider through the agent's extensions or subagentOnlyExtensions; this tool never loads providers or grants tools itself.${enabled ? "" : "\n\nConfigured behavior: bg_wait is disabled by config.waitTool or PI_SUBAGENT_WAIT_TOOL_ENABLED and returns immediately without blocking."}`;
 	const execute: ToolDefinition<typeof SubagentWaitParams, Details>["execute"] = async (_id, params, signal, onUpdate, ctx) => finalizeToolResult(await waitForSubagents(params, signal, {
 		state,
+		nestedRootRunId: child?.nestedRootRunId,
 		events: pi.events,
 		enabled,
 		...(defaultTimeoutMs !== undefined ? { defaultTimeoutMs } : {}),