pi-subagents 0.36.0 → 0.37.1

package/src/runs/foreground/execution.ts

@@ -42,6 +42,7 @@ import {
 	getFinalOutput,
 	findLatestSessionFile,
 	detectSubagentError,
+	hasEmptyTerminalAssistantResponse,
 	extractToolArgsPreview,
 	extractTextFromContent,
 } from "../../shared/utils.ts";
@@ -51,7 +52,8 @@ import { evaluateCompletionMutationGuard } from "../shared/completion-guard.ts";
 import { getPiSpawnCommand } from "../shared/pi-spawn.ts";
 import { createJsonlWriter } from "../../shared/jsonl-writer.ts";
 import { attachPostExitStdioGuard, trySignalChild } from "../../shared/post-exit-stdio-guard.ts";
-import { applyThinkingSuffix, buildPiArgs, cleanupTempDir } from "../shared/pi-args.ts";
+import { applyThinkingSuffix, buildPiArgs, cleanupTempDir, resolvePiLaunchToolPlan } from "../shared/pi-args.ts";
+import { decodeSubagentCapabilityCeiling, resolveCurrentSubagentCapabilityCeiling, SUBAGENT_CAPABILITY_CEILING_ENV } from "../shared/capability-ceiling.ts";
 import { resolveEffectiveThinking } from "../../shared/model-info.ts";
 import { MISSING_STRUCTURED_OUTPUT_CALL_ERROR, readStructuredOutput } from "../shared/structured-output.ts";
 import { readChildToolDiagnosticError } from "../shared/tool-availability.ts";
@@ -77,6 +79,7 @@ import { attachContractProjections, isAgentContractV1 } from "../shared/agent-co
 import { appendTurnBudgetSystemPrompt, formatTurnBudgetOutput, initialTurnBudgetState, turnBudgetDecision, turnBudgetDeferredNote, turnBudgetDeferredState, turnBudgetExceededMessage, turnBudgetSoftNote, turnBudgetState } from "../shared/turn-budget.ts";
 import { initialToolBudgetState, toolBudgetState } from "../shared/tool-budget.ts";
 import { resolveWatchdogConfig } from "../../watchdog/settings.ts";
+import { agentDefinitionDigest, launchBindingDigest } from "../../shared/launch-contract.ts";
 import { createBoundedByteTail, createBoundedLineReader, formatProtocolOutputLimit, MAX_CHILD_STDERR_BYTES, projectChildLifecycle, type ChildLifecycleAction, type ProtocolOutputLimit } from "../shared/child-protocol.ts";
 import {
 	acceptChildWatchdogEvent,
@@ -125,6 +128,7 @@ function resolveAttemptTimeout(options: RunSyncOptions): { timeoutMs: number; re
 function buildPendingAcceptanceLedger(acceptance: ResolvedAcceptanceConfig): AcceptanceLedger {
 	return {
 		status: "pending",
+		evidenceStatus: "pending",
 		explicit: acceptance.explicit,
 		effectiveAcceptance: acceptance,
 		inferredReason: acceptance.inferredReason,
@@ -194,6 +198,7 @@ async function runSingleAttempt(
 		sessionEnabled: boolean;
 		systemPrompt: string;
 		resolvedSkillNames?: string[];
+		modelCandidates?: string[];
 		skillsWarning?: string;
 		jsonlPath?: string;
 		artifactPaths?: ArtifactPaths;
@@ -215,7 +220,7 @@ async function runSingleAttempt(
 			childIndex: options.index ?? 0,
 		})
 		: undefined;
-	const { args, env: sharedEnv, tempDir, toolDiagnosticPath } = buildPiArgs({
+	const { args, env: sharedEnv, tempDir, toolDiagnosticPath, capabilityAudit } = buildPiArgs({
 		baseArgs: ["--mode", "json", "-p"],
 		task,
 		sessionEnabled: shared.sessionEnabled,
@@ -249,12 +254,44 @@ async function runSingleAttempt(
 		allowZeroToolBudget: options.allowZeroToolBudget,
 		childWatchdog,
 		waitToolEnabled: options.waitToolEnabled,
+		capabilityCeiling: options.capabilityCeiling,
 	});
 
+	const effectiveSystemPrompt = appendTurnBudgetSystemPrompt(shared.systemPrompt, options.turnBudget);
+	const toolPlan = resolvePiLaunchToolPlan({
+		tools: agent.tools,
+		extensions: agent.extensions,
+		subagentOnlyExtensions: agent.subagentOnlyExtensions,
+		mcpDirectTools: agent.mcpDirectTools,
+		cwd: options.cwd ?? runtimeCwd,
+		requireReadTool: Boolean(shared.resolvedSkillNames?.length),
+		structuredOutput: Boolean(options.structuredOutput),
+		capabilityCeiling: options.capabilityCeiling,
+		inheritedCapabilityCeiling: decodeSubagentCapabilityCeiling(process.env[SUBAGENT_CAPABILITY_CEILING_ENV]),
+	});
+	const launchContractDigest = launchBindingDigest({
+		definitionDigest: agentDefinitionDigest(agent),
+		task: shared.originalTask ?? task,
+		...(modelArg ? { model: modelArg } : {}),
+		modelCandidates: shared.modelCandidates,
+		...(resolvedThinking ? { thinking: resolvedThinking } : {}),
+		systemPrompt: effectiveSystemPrompt,
+		systemPromptMode: agent.systemPromptMode,
+		inheritProjectContext: agent.inheritProjectContext,
+		inheritSkills: agent.inheritSkills,
+		skills: shared.resolvedSkillNames ?? [],
+		tools: toolPlan.effectiveToolAllowlist,
+		extensions: toolPlan.extensionArgs,
+		mcpDirectTools: toolPlan.effectiveMcpTools,
+		...(options.outputPath ? { outputPath: options.outputPath } : {}),
+		outputMode: options.outputMode ?? "inline",
+		...(options.structuredOutput ? { structuredOutputSchema: options.structuredOutput.schema } : {}),
+	});
 	const result: SingleResult = withRunContext({
 		agent: agent.name,
 		task: shared.originalTask ?? task,
 		...(options.agentContract ? { agentContract: options.agentContract } : {}),
+		launchContractDigest,
 		exitCode: 0,
 		messages: [],
 		usage: emptyUsage(),
@@ -266,6 +303,8 @@ async function runSingleAttempt(
 		skillsWarning: shared.skillsWarning,
 		...(options.turnBudget ? { turnBudget: initialTurnBudgetState(options.turnBudget) } : {}),
 		...(options.toolBudget ? { toolBudget: initialToolBudgetState(options.toolBudget) } : {}),
+		...(options.capabilityCeiling ? { capabilityCeiling: options.capabilityCeiling } : {}),
+		...(capabilityAudit ? { capabilityAudit } : {}),
 	}, options.context);
 	const startTime = Date.now();
 	if (options.structuredOutput) {
@@ -1041,22 +1080,21 @@ async function runSingleAttempt(
 		result.exitCode = 1;
 	}
 	if (result.exitCode === 0 && !result.error) {
-		const errInfo = detectSubagentError(result.messages ?? []);
-		if (errInfo.hasError) {
-			result.exitCode = errInfo.exitCode ?? 1;
-			result.error = errInfo.details
-				? `${errInfo.errorType} failed (exit ${errInfo.exitCode}): ${errInfo.details}`
-				: `${errInfo.errorType} failed with exit code ${errInfo.exitCode}`;
-		}
-	}
-	if (result.exitCode === 0 && !result.error) {
-		const finalText = getFinalOutput(result.messages ?? []);
+		const messages = result.messages ?? [];
+		const finalText = getFinalOutput(messages);
 		const missingStructuredOutput = options.structuredOutput
 			? !existsSync(options.structuredOutput.outputPath)
 			: false;
-		if (!finalText?.trim() && (!options.structuredOutput || missingStructuredOutput)) {
+		const errInfo = detectSubagentError(messages);
+		const missingOutput = !finalText?.trim() && (!options.structuredOutput || missingStructuredOutput);
+		if (missingOutput && (!errInfo.hasError || hasEmptyTerminalAssistantResponse(messages))) {
 			result.exitCode = 1;
 			result.error = "Subagent produced no output (possible model cold-start or empty response).";
+		} else if (errInfo.hasError) {
+			result.exitCode = errInfo.exitCode ?? 1;
+			result.error = errInfo.details
+				? `${errInfo.errorType} failed (exit ${errInfo.exitCode}): ${errInfo.details}`
+				: `${errInfo.errorType} failed with exit code ${errInfo.exitCode}`;
 		}
 	}
 	if (options.structuredOutput && result.exitCode === 0 && !result.error) {
@@ -1194,6 +1232,10 @@ export async function runSync(
 	task: string,
 	options: RunSyncOptions,
 ): Promise<SingleResult> {
+	options = {
+		...options,
+		capabilityCeiling: options.capabilityCeiling ?? resolveCurrentSubagentCapabilityCeiling(options.parentSessionId),
+	};
 	const agent = agents.find((a) => a.name === agentName);
 	if (!agent) {
 		return withRunContext({
@@ -1317,8 +1359,11 @@ export async function runSync(
 			toolCount: target.progressSummary?.toolCount,
 			error: target.error,
 			agentContract: target.agentContract,
+			launchContractDigest: target.launchContractDigest,
 			execution: target.execution,
 			acceptance: target.acceptance,
+			capabilityCeiling: target.capabilityCeiling,
+			capabilityAudit: target.capabilityAudit,
 			review: target.review,
 			effects: target.effects,
 			...(transcriptWriter ? { transcriptPath: artifactPathsResult.transcriptPath } : {}),
@@ -1383,6 +1428,7 @@ export async function runSync(
 			artifactPaths: artifactPathsResult,
 			transcriptWriter,
 			attemptNotes,
+			modelCandidates: candidates.map((candidate) => applyThinkingSuffix(candidate, options.thinkingOverride ?? agent.thinking, options.thinkingOverride !== undefined)),
 			outputSnapshot,
 			originalTask: task,
 		});