pi-subagents 0.55.0 → 0.56.0

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

@@ -80,9 +80,10 @@ import {
 import { applyThinkingSuffix, buildPiArgs, cleanupTempDir, projectLaunchResolvedChildExtensions, resolvePiLaunchToolPlan, type SubagentTaskDelivery } from "../shared/pi-args.ts";
 import { readRuntimeAcknowledgedExtensions } from "../shared/runtime-acknowledged-extensions.ts";
 import { outputEntryFromAsyncResult, resolveOutputReferences } from "../shared/chain-outputs.ts";
-import { createStructuredOutputRuntime, MISSING_STRUCTURED_OUTPUT_CALL_ERROR, readStructuredOutput } from "../shared/structured-output.ts";
+import { createStructuredOutputRuntime, MISSING_STRUCTURED_OUTPUT_CALL_ERROR, readStructuredOutput, readStructuredOutputAcceptanceReport } from "../shared/structured-output.ts";
 import { formatProcessSignalError, isUnexplainedProcessSignal } from "../shared/process-signal.ts";
 import { readChildToolDiagnosticError } from "../shared/tool-availability.ts";
+import { buildTimeoutRecoverySummary, collectTrackedMutationEvidence, snapshotTrackedMutations } from "../shared/mutation-evidence.ts";
 import { collectDynamicResults, DynamicFanoutError, materializeDynamicParallelStep, validateDynamicCollection } from "../shared/dynamic-fanout.ts";
 import { claimRunFanoutBatch, getRunFanoutBudgetSnapshot } from "../shared/run-fanout-budget.ts";
 import { nestedSummaryFromAsyncStatus, projectNestedEvents, resolveNestedAsyncDir, writeNestedEvent } from "../shared/nested-events.ts";
@@ -131,6 +132,7 @@ import { resolveSubagentIntercomTarget } from "../../intercom/intercom-bridge.ts
 import { acceptanceFailureMessage, aggregateAcceptanceReport, buildSkippedAcceptanceLedger, evaluateAcceptance, formatAcceptancePrompt, resolveEffectiveAcceptance, stripAcceptanceReport } from "../shared/acceptance.ts";
 import { attachContractProjections, isAgentContractV1 } from "../shared/agent-contract.ts";
 import { waitForImportedAsyncRoot } from "./chain-root-attachment.ts";
+import { normalizeExtensionBindings } from "../shared/extension-bindings.ts";
 import { appendRunnerStepsToStatus, consumeChainAppendRequests, countPendingChainAppendRequests, statusStepDescription } from "./chain-append.ts";
 import { asyncStatusChildIdentity } from "../shared/child-identity.ts";
 import { appendTurnBudgetSystemPrompt, formatTurnBudgetOutput, initialTurnBudgetState, turnBudgetDecision, turnBudgetDeferredNote, turnBudgetDeferredState, turnBudgetExceededMessage, turnBudgetSoftNote, turnBudgetState } from "../shared/turn-budget.ts";
@@ -229,6 +231,7 @@ interface StepResult {
 	timedOut?: boolean;
 	stopped?: boolean;
 	processSignal?: string | null;
+	timeoutRecovery?: import("../../shared/types.ts").TimeoutRecoverySummary;
 	turnBudget?: TurnBudgetState;
 	turnBudgetExceeded?: boolean;
 	wrapUpRequested?: boolean;
@@ -540,6 +543,9 @@ interface RunPiStreamingResult {
 	processCloseObservedAt?: number;
 	processSignal?: string | null;
 	processTree: ProcessTreeTerminalV1;
+	currentTool?: string;
+	currentToolArgs?: string;
+	currentPath?: string;
 }
 
 function runPiStreaming(
@@ -610,6 +616,9 @@ function runPiStreaming(
 		let observedMutationAttempt = false;
 		let structuredOutputToolInvoked = false;
 		let structuredOutputMessageStartIndex: number | undefined;
+		let currentTool: string | undefined;
+		let currentToolArgs: string | undefined;
+		let currentPath: string | undefined;
 		let toolCount = 0;
 		const childWatchdogConfig = decodeChildWatchdogConfig(env?.[CHILD_WATCHDOG_CONFIG_ENV]);
 		let childWatchdogState: ChildWatchdogStateSnapshot | undefined;
@@ -701,12 +710,18 @@ function runPiStreaming(
 
 			if (event.type === "tool_execution_end") {
 				clearActiveToolTimeout(event);
+				currentTool = undefined;
+				currentToolArgs = undefined;
+				currentPath = undefined;
 				return;
 			}
 
 			if (event.type === "tool_execution_start" && event.toolName) {
 				toolCount += 1;
 				armToolTimeout({ toolCallId: (event as { toolCallId?: unknown }).toolCallId, toolName: event.toolName });
+				currentTool = event.toolName;
+				currentToolArgs = extractToolArgsPreview(event.args ?? {});
+				currentPath = resolveCurrentPath(event.toolName, event.args);
 				if (event.toolName === "structured_output") {
 					structuredOutputToolInvoked = true;
 					structuredOutputMessageStartIndex = messages.length;
@@ -1029,6 +1044,9 @@ function runPiStreaming(
 				processCloseObservedAt,
 				processSignal: signal,
 				processTree,
+				currentTool,
+				currentToolArgs,
+				currentPath,
 			}));
 		});
 
@@ -1207,6 +1225,8 @@ interface SingleStepContext {
 	onExternalProcess?: (process: ExternalProcessStatus) => void;
 	onExternalJob?: (status: ExternalJobStatus) => void;
 	skipAcceptance?: () => boolean;
+	/** False when sibling work in the same Git worktree could have caused the tracked diff. */
+	trackedMutationEvidenceForCompletionGuard?: boolean;
 	orcaProgressTab?: OrcaProgressTab;
 }
 
@@ -1297,6 +1317,9 @@ async function runSingleStepInner(
 			tools: step.tools,
 			extensions: step.extensions,
 			subagentOnlyExtensions: step.subagentOnlyExtensions,
+			fast: step.fast,
+			model: step.model,
+			modelCandidates: step.modelCandidates,
 			mcpDirectTools: step.mcpDirectTools,
 			cwd: step.cwd ?? ctx.cwd,
 			requireReadTool: Boolean(step.skills?.length),
@@ -1329,7 +1352,7 @@ async function runSingleStepInner(
 		}
 	}
 	if (step.effectiveAcceptance) {
-		const acceptancePrompt = formatAcceptancePrompt(step.effectiveAcceptance, { reportOptional: isAgentContractV1(step.agentContract) });
+		const acceptancePrompt = formatAcceptancePrompt(step.effectiveAcceptance, { reportOptional: isAgentContractV1(step.agentContract), structuredOutput: Boolean(step.structuredOutput?.acceptanceReportPath) });
 		if (acceptancePrompt) task = `${task}\n${acceptancePrompt}`;
 	}
 	const sessionEnabled = Boolean(step.sessionFile) || ctx.sessionEnabled;
@@ -1500,11 +1523,15 @@ async function runSingleStepInner(
 	const eventsPath = path.join(path.dirname(ctx.outputFile), "events.jsonl");
 	let finalResult: RunPiStreamingResult | undefined;
 	let finalOutputSnapshot: SingleOutputSnapshot | undefined;
+	let structuredAcceptanceReport: unknown;
+	let structuredAcceptanceReportError: string | undefined;
 	let completionGuardTriggeredFinal = false;
 	let turnBudget = ctx.turnBudget ? initialTurnBudgetState(ctx.turnBudget) : undefined;
 	let toolBudget = step.toolBudget ? initialToolBudgetState(step.toolBudget) : undefined;
 	let toolBudgetBlocked = false;
 	let actualLaunchContractDigest = step.launchContractDigest;
+	const mutationSnapshot = snapshotTrackedMutations(step.cwd ?? ctx.cwd);
+	let finalMutationEvidence = collectTrackedMutationEvidence(mutationSnapshot, step.cwd ?? ctx.cwd);
 
 	let modelIndex = 0;
 	let startupAttemptIndex = 0;
@@ -1533,6 +1560,7 @@ async function runSingleStepInner(
 			}
 		}
 		const watchdogConfig = resolveWatchdogConfig(step.cwd ?? ctx.cwd);
+		const extensionBindings = normalizeExtensionBindings(step.extensionBindings)?.value;
 		const childWatchdog = watchdogConfig.ok
 			? resolveChildWatchdogConfig({
 				config: watchdogConfig.config,
@@ -1556,6 +1584,8 @@ async function runSingleStepInner(
 			tools: step.tools,
 			extensions: step.extensions,
 			subagentOnlyExtensions: step.subagentOnlyExtensions,
+			fast: step.fast,
+			modelCandidates: step.modelCandidates,
 			systemPrompt: appendTurnBudgetSystemPrompt(step.systemPrompt ?? "", ctx.turnBudget),
 			systemPromptMode: step.systemPromptMode,
 			mcpDirectTools: step.mcpDirectTools,
@@ -1587,6 +1617,7 @@ async function runSingleStepInner(
 			childWatchdog,
 			waitToolEnabled: step.waitToolEnabled,
 			thinkingCeiling: step.thinkingCeiling,
+			extensionBindings,
 		}));
 		if (!launchWarningsEmitted && warnings.length > 0) {
 			for (const warning of warnings) console.warn(`[pi-subagents] ${warning}`);
@@ -1597,6 +1628,9 @@ async function runSingleStepInner(
 				tools: step.tools,
 				extensions: step.extensions,
 				subagentOnlyExtensions: step.subagentOnlyExtensions,
+				fast: step.fast,
+				model: step.model,
+				modelCandidates: step.modelCandidates,
 				mcpDirectTools: step.mcpDirectTools,
 				cwd: step.cwd ?? ctx.cwd,
 				requireReadTool: Boolean(step.skills?.length),
@@ -1611,6 +1645,7 @@ async function runSingleStepInner(
 				task: step.launchBindingTask ?? task,
 				...(candidate ? { model: candidate } : {}),
 				modelCandidates: candidates as string[],
+				...(step.fast !== undefined ? { fast: step.fast } : {}),
 				...(resolveEffectiveThinking(candidate, step.thinking) ? { thinking: resolveEffectiveThinking(candidate, step.thinking) } : {}),
 				...(step.thinkingCeiling ? { thinkingCeiling: step.thinkingCeiling } : {}),
 				systemPrompt: appendTurnBudgetSystemPrompt(step.systemPrompt ?? "", ctx.turnBudget),
@@ -1624,6 +1659,7 @@ async function runSingleStepInner(
 				...(step.outputPath ? { outputPath: step.outputPath } : {}),
 				...(step.outputMode ? { outputMode: step.outputMode } : {}),
 				...(step.structuredOutputSchema ? { structuredOutputSchema: step.structuredOutputSchema } : {}),
+				...(extensionBindings ? { extensionBindings } : {}),
 			}));
 		}
 		capabilityAudit = attemptCapabilityAudit;
@@ -1703,6 +1739,9 @@ async function runSingleStepInner(
 				if (structured.error) structuredError = structured.error;
 				else {
 					structuredOutput = structured.value;
+					const acceptanceReport = readStructuredOutputAcceptanceReport(effectiveStructuredOutput);
+					structuredAcceptanceReport = acceptanceReport.value;
+					structuredAcceptanceReportError = acceptanceReport.error;
 					validatedStructuredOutput = true;
 				}
 			}
@@ -1724,6 +1763,9 @@ async function runSingleStepInner(
 			: undefined;
 		const completionGuardEnabled = isAgentContractV1(step.agentContract) ? step.completionGuard === true : step.completionGuard !== false;
 		const completionToolPlan = resolvedTaskToolPlan;
+		const mutationEvidence = collectTrackedMutationEvidence(mutationSnapshot, step.cwd ?? ctx.cwd);
+		finalMutationEvidence = mutationEvidence;
+		const completionMutationEvidence = ctx.trackedMutationEvidenceForCompletionGuard === false ? undefined : mutationEvidence;
 		const completionGuard = run.exitCode === 0 && !run.error && !structuredError && !hiddenError?.hasError && !emptyOutputError && completionGuardEnabled
 			? evaluateCompletionMutationGuard(omitUndefinedProperties({
 				agent: step.agent,
@@ -1732,15 +1774,18 @@ async function runSingleStepInner(
 				tools: completionToolPlan ? (completionToolPlan.explicitToolAllowlist ? completionToolPlan.effectiveToolAllowlist : undefined) : step.tools,
 				mcpDirectTools: completionToolPlan?.effectiveMcpTools ?? step.mcpDirectTools,
 				toolAvailabilityError,
+				mutationEvidence: completionMutationEvidence,
 			}))
 			: undefined;
-		const completionGuardTriggered = completionGuard?.triggered === true && !run.observedMutationAttempt;
+		const mutationAttemptObserved = run.observedMutationAttempt === true || completionMutationEvidence?.attemptedMutation === true;
+		const completionGuardTriggered = completionGuard?.triggered === true && !mutationAttemptObserved;
 		const completionGuardBlocked = completionGuard?.blocked === true;
 		const fileMutationEffect = completionGuard
 			? {
 				status: completionGuardBlocked ? "blocked" as const : completionGuard.expectedMutation ? completionGuardTriggered ? "missing" as const : "observed" as const : "not-applicable" as const,
 				expected: completionGuard.expectedMutation,
-				attempted: completionGuardBlocked ? false : completionGuard.attemptedMutation || run.observedMutationAttempt === true,
+				attempted: completionGuardBlocked ? false : completionGuard.attemptedMutation || mutationAttemptObserved,
+				...(completionMutationEvidence ? { evidence: completionMutationEvidence } : {}),
 				...(completionGuardBlocked && completionGuard.message ? { message: completionGuard.message } : {}),
 				...(completionGuardTriggered ? { message: "Subagent completed without making edits for an implementation task." } : {}),
 			}
@@ -1890,6 +1935,21 @@ async function runSingleStepInner(
 		: resolvedOutput.savedPath
 			? "unknown"
 			: finalResult?.outputState ?? "unknown";
+	const timeoutRecovery = finalResult?.timedOut === true || ctx.timeoutSignal?.aborted === true
+		? buildTimeoutRecoverySummary({
+			termination: "timed-out",
+			evidence: finalMutationEvidence,
+			currentTool: finalResult?.currentTool,
+			currentToolArgs: finalResult?.currentToolArgs,
+			currentPath: finalResult?.currentPath,
+			sessionFile: step.sessionFile,
+			transcriptPath: transcriptWriter ? artifactPaths?.transcriptPath : undefined,
+			artifactPaths,
+		})
+		: undefined;
+	if (timeoutRecovery) outputForSummary = outputForSummary.trim()
+		? `${outputForSummary}\n\n${timeoutRecovery.message}`
+		: timeoutRecovery.message;
 	const finalizedOutput = finalizeSingleOutput(omitUndefinedProperties({
 		fullOutput: outputForSummary,
 		outputPath: step.outputPath,
@@ -1904,6 +1964,8 @@ async function runSingleStepInner(
 		? await evaluateAcceptance(omitUndefinedProperties({
 			acceptance: step.effectiveAcceptance,
 			output: outputForAcceptance,
+			report: structuredAcceptanceReport as import("../../shared/types.ts").AcceptanceReport | undefined,
+			reportError: structuredAcceptanceReportError,
 			fileOutput: childWrittenOutput !== undefined && step.outputPath
 				? { content: childWrittenOutput, path: step.outputPath, authoritative: step.outputMode === "file-only" }
 				: undefined,
@@ -1999,6 +2061,7 @@ async function runSingleStepInner(
 		timedOut: timedOutAfterAcceptance ? true : finalResult?.timedOut,
 		stopped: stoppedAfterAcceptance ? true : finalResult?.stopped,
 		processSignal: finalResult?.processSignal,
+		timeoutRecovery,
 		turnBudget,
 		turnBudgetExceeded: turnBudgetExceeded || undefined,
 		wrapUpRequested: finalResult?.wrapUpRequested || turnBudget?.outcome === "wrap-up-requested" || turnBudget?.outcome === "termination-deferred" || turnBudgetExceeded || undefined,
@@ -4013,6 +4076,7 @@ async function runSubagent(
 					registerTurnBudgetAbort: (abort) => registerStepTurnBudgetAbort(fi, abort),
 					timeoutSignal: timeoutAbortController.signal,
 					stopSignal: stopAbortController.signal,
+					trackedMutationEvidenceForCompletionGuard: false,
 					timeoutMessage,
 					stopMessage,
 					turnBudget: config.turnBudget,
@@ -4072,6 +4136,7 @@ async function runSubagent(
 				setOptionalProperty(requiredStatusStep(statusPayload, fi), "structuredOutputPath", singleResult.structuredOutputPath);
 				setOptionalProperty(requiredStatusStep(statusPayload, fi), "structuredOutputSchemaPath", singleResult.structuredOutputSchemaPath);
 				setOptionalProperty(requiredStatusStep(statusPayload, fi), "acceptance", singleResult.acceptance);
+				setOptionalProperty(requiredStatusStep(statusPayload, fi), "timeoutRecovery", singleResult.timeoutRecovery);
 				setOptionalProperty(requiredStatusStep(statusPayload, fi), "watchdog", singleResult.watchdog);
 				setOptionalProperty(requiredStatusStep(statusPayload, fi), "capabilityCeiling", singleResult.capabilityCeiling);
 				setOptionalProperty(requiredStatusStep(statusPayload, fi), "capabilityAudit", singleResult.capabilityAudit);
@@ -4087,7 +4152,7 @@ async function runSubagent(
 		}));
 		if (stopped || childStopped) appendTerminalChildStatusEvent(fi, taskEndTime);
 		if (singleResult.exitCode !== 0 && failFast && !childStopped) aborted = true;
-				return stopped || childStopped ? { ...singleResult, output: stopMessage, error: stopMessage, exitCode: 1, interrupted: false, timedOut: false, stopped: true, skipped: false } : timedOut ? { ...singleResult, output: timeoutMessage ?? "Subagent timed out.", error: timeoutMessage ?? "Subagent timed out.", exitCode: 1, interrupted: false, timedOut: true, skipped: false } : { ...singleResult, skipped: false };
+				return stopped || childStopped ? { ...singleResult, output: stopMessage, error: stopMessage, exitCode: 1, interrupted: false, timedOut: false, stopped: true, skipped: false } : timedOut ? { ...singleResult, output: singleResult.output || (timeoutMessage ?? "Subagent timed out."), error: singleResult.error ?? timeoutMessage ?? "Subagent timed out.", exitCode: 1, interrupted: false, timedOut: true, skipped: false } : { ...singleResult, skipped: false };
 			}, globalSemaphore);
 
 			flatIndex += dynamicSteps.length;
@@ -4127,6 +4192,7 @@ async function runSubagent(
 					effects: pr.effects,
 					execution: pr.execution,
 					review: pr.review,
+					timeoutRecovery: pr.timeoutRecovery,
 					structuredOutput: pr.structuredOutput,
 					structuredOutputPath: pr.structuredOutputPath,
 					structuredOutputSchemaPath: pr.structuredOutputSchemaPath,
@@ -4406,6 +4472,7 @@ async function runSubagent(
 							registerTurnBudgetAbort: (abort) => registerStepTurnBudgetAbort(fi, abort),
 							timeoutSignal: timeoutAbortController.signal,
 							stopSignal: stopAbortController.signal,
+							trackedMutationEvidenceForCompletionGuard: Boolean(worktreeSetup),
 							timeoutMessage,
 							stopMessage,
 							turnBudget: config.turnBudget,
@@ -4470,6 +4537,7 @@ async function runSubagent(
 						setOptionalProperty(requiredStatusStep(statusPayload, fi), "structuredOutputPath", singleResult.structuredOutputPath);
 						setOptionalProperty(requiredStatusStep(statusPayload, fi), "structuredOutputSchemaPath", singleResult.structuredOutputSchemaPath);
 						setOptionalProperty(requiredStatusStep(statusPayload, fi), "acceptance", singleResult.acceptance);
+						setOptionalProperty(requiredStatusStep(statusPayload, fi), "timeoutRecovery", singleResult.timeoutRecovery);
 						setOptionalProperty(requiredStatusStep(statusPayload, fi), "watchdog", singleResult.watchdog);
 						setOptionalProperty(requiredStatusStep(statusPayload, fi), "capabilityCeiling", singleResult.capabilityCeiling);
 						setOptionalProperty(requiredStatusStep(statusPayload, fi), "capabilityAudit", singleResult.capabilityAudit);
@@ -4500,7 +4568,7 @@ async function runSubagent(
 						}
 
 						if (singleResult.exitCode !== 0 && failFast && !childStopped) aborted = true;
-						return stopped || childStopped ? { ...singleResult, output: stopMessage, error: stopMessage, exitCode: 1, interrupted: false, timedOut: false, stopped: true, skipped: false } : timedOut ? { ...singleResult, output: timeoutMessage ?? "Subagent timed out.", error: timeoutMessage ?? "Subagent timed out.", exitCode: 1, interrupted: false, timedOut: true, skipped: false } : { ...singleResult, skipped: false };
+						return stopped || childStopped ? { ...singleResult, output: stopMessage, error: stopMessage, exitCode: 1, interrupted: false, timedOut: false, stopped: true, skipped: false } : timedOut ? { ...singleResult, output: singleResult.output || (timeoutMessage ?? "Subagent timed out."), error: singleResult.error ?? timeoutMessage ?? "Subagent timed out.", exitCode: 1, interrupted: false, timedOut: true, skipped: false } : { ...singleResult, skipped: false };
 					},
 					globalSemaphore,
 				);
@@ -4557,10 +4625,11 @@ async function runSubagent(
 						artifactPaths: pr.artifactPaths,
 						transcriptPath: pr.transcriptPath,
 						transcriptError: pr.transcriptError,
-						effects: pr.effects,
-						execution: pr.execution,
-						review: pr.review,
-						structuredOutput: pr.structuredOutput,
+							effects: pr.effects,
+							execution: pr.execution,
+							review: pr.review,
+							timeoutRecovery: pr.timeoutRecovery,
+							structuredOutput: pr.structuredOutput,
 						structuredOutputPath: pr.structuredOutputPath,
 						structuredOutputSchemaPath: pr.structuredOutputSchemaPath,
 						acceptance: pr.acceptance,
@@ -4780,7 +4849,7 @@ async function runSubagent(
 				launchContractDigest: singleResult.launchContractDigest,
 				launchResolvedExtensions: singleResult.launchResolvedExtensions,
 				runtimeAcknowledgedExtensions: singleResult.runtimeAcknowledgedExtensions,
-				output: stopped || childStopped ? stopMessage : timedOut ? (timeoutMessage ?? "Subagent timed out.") : singleResult.output,
+				output: stopped || childStopped ? stopMessage : timedOut ? singleResult.output || (timeoutMessage ?? "Subagent timed out.") : singleResult.output,
 				outputState: singleResult.outputState,
 				error: stopped || childStopped ? stopMessage : timedOut ? (timeoutMessage ?? "Subagent timed out.") : singleResult.error,
 				protocolError: singleResult.protocolError,
@@ -4799,6 +4868,7 @@ async function runSubagent(
 				effects: singleResult.effects,
 				execution: singleResult.execution,
 				review: singleResult.review,
+				timeoutRecovery: singleResult.timeoutRecovery,
 				structuredOutput: singleResult.structuredOutput,
 				structuredOutputPath: singleResult.structuredOutputPath,
 				structuredOutputSchemaPath: singleResult.structuredOutputSchemaPath,
@@ -4885,6 +4955,7 @@ async function runSubagent(
 			setOptionalProperty(requiredStatusStep(statusPayload, flatIndex), "structuredOutputPath", singleResult.structuredOutputPath);
 			setOptionalProperty(requiredStatusStep(statusPayload, flatIndex), "structuredOutputSchemaPath", singleResult.structuredOutputSchemaPath);
 			setOptionalProperty(requiredStatusStep(statusPayload, flatIndex), "acceptance", singleResult.acceptance);
+			setOptionalProperty(requiredStatusStep(statusPayload, flatIndex), "timeoutRecovery", singleResult.timeoutRecovery);
 			setOptionalProperty(requiredStatusStep(statusPayload, flatIndex), "watchdog", singleResult.watchdog);
 			setOptionalProperty(requiredStatusStep(statusPayload, flatIndex), "capabilityCeiling", singleResult.capabilityCeiling);
 			setOptionalProperty(requiredStatusStep(statusPayload, flatIndex), "capabilityAudit", singleResult.capabilityAudit);
@@ -5169,6 +5240,7 @@ async function runSubagent(
 				structuredOutputSchemaPath: r.structuredOutputSchemaPath,
 				acceptance: r.acceptance,
 				watchdog: r.watchdog,
+				timeoutRecovery: r.timeoutRecovery,
 				capabilityCeiling: r.capabilityCeiling,
 				capabilityAudit: r.capabilityAudit,
 			})),

package/src/runs/foreground/execution.ts

@@ -67,9 +67,10 @@ import { readRuntimeAcknowledgedExtensions } from "../shared/runtime-acknowledge
 import { assertAgentAllowedByCapabilityCeiling, decodeSubagentCapabilityCeiling, intersectSubagentCapabilityCeilings, resolveCurrentSubagentCapabilityCeiling, SUBAGENT_CAPABILITY_CEILING_ENV } from "../shared/capability-ceiling.ts";
 import { resolveEffectiveThinking } from "../../shared/model-info.ts";
 import { assertThinkingWithinCeiling, decodeThinkingCeiling, intersectThinkingCeilings, SUBAGENT_THINKING_CEILING_ENV } from "../../shared/thinking-ceiling.ts";
-import { MISSING_STRUCTURED_OUTPUT_CALL_ERROR, readStructuredOutput } from "../shared/structured-output.ts";
+import { MISSING_STRUCTURED_OUTPUT_CALL_ERROR, readStructuredOutput, readStructuredOutputAcceptanceReport } from "../shared/structured-output.ts";
 import { formatProcessSignalError, isUnexplainedProcessSignal } from "../shared/process-signal.ts";
 import { readChildToolDiagnosticError } from "../shared/tool-availability.ts";
+import { buildTimeoutRecoverySummary, collectTrackedMutationEvidence, snapshotTrackedMutations } from "../shared/mutation-evidence.ts";
 import { captureSingleOutputSnapshot, extractChildWrittenOutput, finalizeSingleOutput, formatSavedOutputReference, injectOutputPathSystemPrompt, resolveSingleOutput, validateFileOnlyOutputMode, type SingleOutputSnapshot } from "../shared/single-output.ts";
 import {
 	buildModelCandidates,
@@ -368,6 +369,8 @@ async function runSingleAttempt(
 		steerCapabilityPath: options.steerCapabilityPath,
 		steerAckDir: options.steerAckDir,
 		structuredOutput: options.structuredOutput,
+		fast: options.fast ?? agent.fast,
+		modelCandidates: shared.modelCandidates,
 		toolBudget: options.toolBudget,
 		allowZeroToolBudget: options.allowZeroToolBudget,
 		permissionRules,
@@ -376,6 +379,7 @@ async function runSingleAttempt(
 		waitToolEnabled: options.waitToolEnabled,
 		capabilityCeiling: options.capabilityCeiling,
 		thinkingCeiling: options.thinkingCeiling,
+		extensionBindings: options.extensionBindings,
 	});
 	if (!shared.launchWarnings.emitted && warnings.length > 0) {
 		for (const warning of warnings) console.warn(`[pi-subagents] ${warning}`);
@@ -391,6 +395,9 @@ async function runSingleAttempt(
 		cwd: options.cwd ?? runtimeCwd,
 		requireReadTool: Boolean(shared.resolvedSkillNames?.length),
 		structuredOutput: Boolean(options.structuredOutput),
+		fast: options.fast ?? agent.fast,
+		model: modelArg,
+		modelCandidates: shared.modelCandidates,
 		capabilityCeiling: options.capabilityCeiling,
 		inheritedCapabilityCeiling: decodeSubagentCapabilityCeiling(process.env[SUBAGENT_CAPABILITY_CEILING_ENV]),
 		agentName: agent.name,
@@ -432,6 +439,7 @@ async function runSingleAttempt(
 		task: shared.originalTask ?? task,
 		...(modelArg ? { model: modelArg } : {}),
 		modelCandidates: shared.modelCandidates,
+		...((options.fast ?? agent.fast) !== undefined ? { fast: options.fast ?? agent.fast } : {}),
 		...(resolvedThinking ? { thinking: resolvedThinking } : {}),
 		...(options.thinkingCeiling ? { thinkingCeiling: options.thinkingCeiling } : {}),
 		systemPrompt: effectiveSystemPrompt,
@@ -445,6 +453,7 @@ async function runSingleAttempt(
 		...(options.outputPath ? { outputPath: options.outputPath } : {}),
 		outputMode: options.outputMode ?? "inline",
 		...(options.structuredOutput ? { structuredOutputSchema: options.structuredOutput.schema } : {}),
+		...(options.extensionBindings ? { extensionBindings: options.extensionBindings } : {}),
 	});
 	const result: SingleResult = withRunContext({
 		index: options.index ?? 0,
@@ -524,6 +533,7 @@ async function runSingleAttempt(
 		return result;
 	}
 	const spawnEnv = { ...process.env, ...sharedEnv, ...getSubagentDepthEnv(options.maxSubagentDepth) };
+	const mutationSnapshot = snapshotTrackedMutations(options.cwd ?? runtimeCwd);
 	let observedMutationAttempt = false;
 	let structuredOutputToolInvoked = false;
 	let structuredOutputMessageStartIndex: number | undefined;
@@ -1448,6 +1458,9 @@ async function runSingleAttempt(
 				result.structuredOutputFailed = true;
 			} else {
 				result.structuredOutput = structured.value;
+				const acceptanceReport = readStructuredOutputAcceptanceReport(options.structuredOutput);
+				(result as SingleResult & { structuredAcceptanceReport?: unknown; structuredAcceptanceReportError?: string }).structuredAcceptanceReport = acceptanceReport.value;
+				(result as SingleResult & { structuredAcceptanceReport?: unknown; structuredAcceptanceReportError?: string }).structuredAcceptanceReportError = acceptanceReport.error;
 				validatedStructuredOutput = true;
 			}
 		}
@@ -1485,6 +1498,7 @@ async function runSingleAttempt(
 		tokens: progress.tokens,
 		durationMs: progress.durationMs,
 	};
+	const mutationEvidence = collectTrackedMutationEvidence(mutationSnapshot, options.cwd ?? runtimeCwd);
 
 	const acceptanceOutput = getFinalOutput(result.messages ?? []);
 	let fullOutput = stripAcceptanceReport(acceptanceOutput);
@@ -1492,9 +1506,19 @@ async function runSingleAttempt(
 	result.outputState = fullOutput.trim() || result.structuredOutput !== undefined ? "present" : "absent";
 	if (result.timedOut) {
 		const timeoutMessage = formatTimeoutMessage(options.timeoutMs ?? 0);
+		result.timeoutRecovery = buildTimeoutRecoverySummary({
+			termination: "timed-out",
+			evidence: mutationEvidence,
+			currentTool: progress.currentTool,
+			currentToolArgs: progress.currentToolArgs,
+			currentPath: progress.currentPath,
+			sessionFile: options.sessionFile,
+			transcriptPath: shared.transcriptWriter ? shared.artifactPaths?.transcriptPath : undefined,
+			artifactPaths: shared.artifactPaths,
+		});
 		fullOutput = fullOutput.trim()
-			? `${timeoutMessage}\n\nPartial output before timeout:\n${fullOutput}`
-			: timeoutMessage;
+			? `${timeoutMessage}\n\n${result.timeoutRecovery.message}\n\nPartial output before timeout:\n${fullOutput}`
+			: `${timeoutMessage}\n\n${result.timeoutRecovery.message}`;
 	} else if (result.turnBudgetExceeded && result.turnBudget) {
 		fullOutput = formatTurnBudgetOutput(turnBudgetExceededMessage(result.turnBudget, result.turnBudget.turnCount), fullOutput);
 	} else if (result.turnBudget?.outcome === "termination-deferred") {
@@ -1513,9 +1537,11 @@ async function runSingleAttempt(
 			tools: contractTools,
 			mcpDirectTools: toolPlan.effectiveMcpTools,
 			toolAvailabilityError,
+			mutationEvidence,
 		})
 		: undefined;
-	let completionGuardTriggered = completionGuard?.triggered === true && !observedMutationAttempt;
+	const mutationAttemptObserved = observedMutationAttempt || mutationEvidence.attemptedMutation;
+	let completionGuardTriggered = completionGuard?.triggered === true && !mutationAttemptObserved;
 	const completionGuardBlocked = completionGuard?.blocked === true;
 	// The classifier is deliberately narrow, so a read-only review task can
 	// still be misread as implementation. Arbitrate BEFORE any failure side
@@ -1547,7 +1573,8 @@ async function runSingleAttempt(
 							: "observed"
 					: "not-applicable",
 				expected: completionGuard.expectedMutation,
-				attempted: completionGuardBlocked ? false : completionGuard.attemptedMutation || observedMutationAttempt,
+				attempted: completionGuardBlocked ? false : completionGuard.attemptedMutation || mutationAttemptObserved,
+				evidence: mutationEvidence,
 				...(completionGuardBlocked && completionGuard.message ? { message: completionGuard.message } : {}),
 				...(completionGuardTriggered ? { message: "Subagent completed without making edits for an implementation task." } : {}),
 				...(arbiterRescued ? { resolvedBy: "llm-intent-arbiter" } : {}),
@@ -1708,7 +1735,7 @@ async function runSyncCompletionInner(
 		dynamicGroup: options.acceptanceContext?.dynamicGroup,
 		agentContract: options.agentContract,
 	});
-	const acceptancePrompt = formatAcceptancePrompt(effectiveAcceptance, { reportOptional: isAgentContractV1(options.agentContract) });
+	const acceptancePrompt = formatAcceptancePrompt(effectiveAcceptance, { reportOptional: isAgentContractV1(options.agentContract), structuredOutput: Boolean(options.structuredOutput?.acceptanceReportPath) });
 	const taskWithAcceptance = acceptancePrompt ? `${task}\n${acceptancePrompt}` : task;
 	options.onEffectivePrompt?.(taskWithAcceptance);
 	const sessionEnabled = Boolean(options.sessionFile || options.sessionDir) || shareEnabled;
@@ -2045,6 +2072,8 @@ async function runSyncCompletionInner(
 			result.acceptance = await evaluateAcceptance({
 				acceptance: effectiveAcceptance,
 				output: acceptanceOutputByResult.get(result) ?? result.finalOutput ?? "",
+				report: (result as SingleResult & { structuredAcceptanceReport?: import("../../shared/types.ts").AcceptanceReport; structuredAcceptanceReportError?: string }).structuredAcceptanceReport,
+				reportError: (result as SingleResult & { structuredAcceptanceReport?: import("../../shared/types.ts").AcceptanceReport; structuredAcceptanceReportError?: string }).structuredAcceptanceReportError,
 				fileOutput: childWrittenOutput !== undefined && options.outputPath
 					? { content: childWrittenOutput, path: options.outputPath, authoritative: options.outputMode === "file-only" }
 					: undefined,

package/src/runs/foreground/foreground-history.ts

@@ -2,7 +2,7 @@ import * as fs from "node:fs";
 import * as path from "node:path";
 import type { ForegroundResumeChild, ForegroundResumeRun, SubagentState } from "../../shared/types.ts";
 import { DIRS } from "../../shared/types.ts";
-import { writeAtomicJson } from "../../shared/atomic-json.ts";
+import { writePrivateAtomicJson } from "../../shared/atomic-json.ts";
 import { utf8Tail } from "../../shared/utf8.ts";
 
 export const MAX_REMEMBERED_FOREGROUND_RUNS = 50;
@@ -52,6 +52,7 @@ function compactChild(child: ForegroundResumeChild): ForegroundResumeChild {
 		...(child.transcriptError ? { transcriptError: child.transcriptError } : {}),
 		...(child.acceptance ? { acceptance: child.acceptance } : {}),
 		...(child.launchContractDigest ? { launchContractDigest: child.launchContractDigest } : {}),
+		...(child.extensionBindings ? { extensionBindings: child.extensionBindings } : {}),
 		...(child.capabilityCeiling ? { capabilityCeiling: child.capabilityCeiling } : {}),
 		...(child.updatedAt !== undefined ? { updatedAt: child.updatedAt } : {}),
 	};
@@ -118,7 +119,7 @@ export function persistForegroundRunHistory(state: SubagentState, options: { res
 		if (compact) merged.set(compact.runId, compact);
 	}
 	const runs = sortAndBound([...merged.values()], limit);
-	writeAtomicJson(historyPath(resultsDir), { version: HISTORY_VERSION, runs });
+	writePrivateAtomicJson(historyPath(resultsDir), { version: HISTORY_VERSION, runs });
 }
 
 export function restoreForegroundRunHistory(state: SubagentState, options: { resultsDir?: string; sessionId?: string | null; limit?: number } = {}): number {