pi-subagents 0.70.1 → 0.72.0

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

@@ -12,10 +12,6 @@ import * as path from "node:path";
 import { performance } from "node:perf_hooks";
 import { pathToFileURL } from "node:url";
 import { installRunnerHttpDispatcher } from "./runner-http-dispatcher.js";
-const isRunnerEntrypoint = Boolean(process.argv[1] && import.meta.url === pathToFileURL(path.resolve(process.argv[1])).href);
-// Detached Node runners skip Pi's CLI dispatcher setup; install the runner's own.
-if (isRunnerEntrypoint)
-    installRunnerHttpDispatcher({ agentDir: getAgentDir(), cwd: process.cwd() });
 import { writeAtomicJson } from "../../shared/atomic-json.js";
 import { writeAsyncResultFile, writePendingAsyncResultFile } from "./result-files.js";
 import { createFileCoalescer } from "../../shared/file-coalescer.js";
@@ -41,7 +37,7 @@ import { SUBAGENT_CHILD_ENV } from "../shared/child-runtime-config.js";
 import { deriveChildSessionName } from "../../shared/child-session-name.js";
 import { alignForkedSessionCwd } from "../../shared/fork-session-cwd.js";
 import { outputEntryFromAsyncResult, resolveOutputReferences } from "../shared/chain-outputs.js";
-import { clearStructuredOutputCaptures, createStructuredOutputFileCapture, createStructuredOutputRuntime, MISSING_STRUCTURED_OUTPUT_CALL_ERROR, readStructuredOutput, readStructuredOutputAcceptanceReport } from "../shared/structured-output.js";
+import { clearStructuredOutputCaptures, createStructuredOutputFileCapture, createStructuredOutputRuntime, formatStructuredOutputRejectionError, MISSING_STRUCTURED_OUTPUT_CALL_ERROR, readStructuredOutput, readStructuredOutputAcceptanceReport } from "../shared/structured-output.js";
 import { formatMidToolExitError, isOrdinaryToolForMidToolExit, isUnexplainedProcessSignal } from "../shared/process-signal.js";
 import { formatChildToolDiagnostic } from "../shared/tool-availability.js";
 import { buildTimeoutRecoverySummary, collectTrackedMutationEvidence, snapshotTrackedMutations } from "../shared/mutation-evidence.js";
@@ -49,7 +45,9 @@ import { collectDynamicResults, DynamicFanoutError, materializeDynamicParallelSt
 import { claimRunFanoutBatch, getRunFanoutBudgetSnapshot } from "../shared/run-fanout-budget.js";
 import { nestedSummaryFromAsyncStatus, projectNestedEvents, resolveNestedAsyncDir, writeNestedEvent } from "../shared/nested-events.js";
 import { isContextOverflow } from "../shared/model-resolution.js";
-import { markProcessTerminalCandidateLeaseRelease, processTerminalPath, writeProcessTerminalCandidate } from "./process-terminal.js";
+import { processTerminalPath, writeProcessTerminalCandidate } from "./process-terminal.js";
+import { persistRunnerStartupFailure } from "./runner-startup-failure.js";
+import { currentPidNamespaceScope } from "./pid-namespace.js";
 import { createSteeringStatus, recordSteeringRequest, steeringStatus, terminalSteeringNoticeState, unconsumedSteerReason, updateSteeringTarget } from "./steering.js";
 import { PROMPT_REDACTED, detectSubagentError, extractTextFromContent, extractToolArgsPreview, formatEmptyTerminalAssistantResponseError, getAgentDir, hasEmptyTerminalAssistantResponse, readStatus } from "../../shared/utils.js";
 import { planAbortRecovery } from "../shared/abort-recovery.js";
@@ -71,7 +69,6 @@ import { toolTimeoutCallKey } from "../shared/tool-timeout.js";
 import { usageBudgetExceededMessage, usageBudgetState } from "../shared/usage-budget.js";
 import { formatParallelHandoffError, formatParallelHandoffReference, parallelHandoffPath, writeParallelHandoffGroup, writeWorktreeSetupHandoff } from "../shared/parallel-handoff.js";
 import { resolveWatchdogConfig } from "../../watchdog/settings.js";
-import { acquireSessionLease } from "../shared/session-lease.js";
 import { buildExternalCliPrompt, runExternalCli } from "../shared/external-cli-runner.js";
 import { resolveClaudeCodeLaunch } from "../shared/claude-code-adapter.js";
 import { resolveCodexExecLaunch } from "../shared/codex-exec-adapter.js";
@@ -814,9 +811,9 @@ export async function runSingleStepInner(step, ctx) {
             const message = error instanceof Error ? error.message : String(error);
             return omitUndefinedProperties({ agent: step.agent, output: message, error: message, exitCode: 1, context: step.context, thinkingCeiling: step.thinkingCeiling });
         }
+        const attemptModel = omitUndefinedProperties({ model: candidate, thinking: resolveEffectiveThinking(candidate, step.thinking) });
         ctx.onAttemptStart?.(omitUndefinedProperties({
-            model: candidate,
-            thinking: resolveEffectiveThinking(candidate, step.thinking),
+            ...attemptModel,
             contextLimit: findModelInfo(candidate, step.modelVerificationRegistry)?.contextWindow,
         }));
         const outputSnapshot = captureSingleOutputSnapshot(step.outputPath);
@@ -938,6 +935,7 @@ export async function runSingleStepInner(step, ctx) {
             timeoutMessage: ctx.timeoutMessage,
             stopMessage: ctx.stopMessage,
             onChildEvent: ctx.onChildEvent,
+            onContextWindow: (contextLimit) => ctx.onAttemptStart?.({ ...attemptModel, contextLimit }),
             transcriptWriter,
             toolTimeoutMs: ctx.toolTimeoutMs,
             runDeadlineAt: ctx.deadlineAt,
@@ -970,18 +968,24 @@ export async function runSingleStepInner(step, ctx) {
         let structuredOutput;
         let structuredError;
         let validatedStructuredOutput = false;
-        if (terminalDiagnosticsEligible && effectiveStructuredOutput && run.exitCode === 0 && !run.error && !toolAvailabilityError && !midToolExitError) {
-            if (!run.structuredOutputToolInvoked) {
+        if (effectiveStructuredOutput) {
+            const otherwiseSuccessful = terminalDiagnosticsEligible && run.exitCode === 0 && !run.error && !toolAvailabilityError && !midToolExitError;
+            if (!run.structuredOutputToolInvoked && otherwiseSuccessful) {
                 structuredError = MISSING_STRUCTURED_OUTPUT_CALL_ERROR;
             }
-            else {
+            else if (run.structuredOutputToolInvoked) {
                 const structured = await readStructuredOutput({
                     schema: effectiveStructuredOutput.schema,
                     schemaPath: effectiveStructuredOutput.schemaPath,
                     outputPath: effectiveStructuredOutput.outputPath,
                 });
-                if (structured.error)
-                    structuredError = structured.error;
+                if (structured.error) {
+                    if (otherwiseSuccessful) {
+                        structuredError = structured.error === MISSING_STRUCTURED_OUTPUT_CALL_ERROR
+                            ? formatStructuredOutputRejectionError(run.messages)
+                            : structured.error;
+                    }
+                }
                 else {
                     structuredOutput = structured.value;
                     const acceptanceReport = readStructuredOutputAcceptanceReport(effectiveStructuredOutput);
@@ -1055,7 +1059,7 @@ export async function runSingleStepInner(step, ctx) {
             afterCompactionSettlement: run.afterCompactionSettlement === true,
         } : undefined;
         const fileMutationEffect = missingRequiredOutputAfterMutation ? { status: "observed", attempted: true, evidence: mutationEvidence } : undefined;
-        finalResult = { ...run, exitCode: effectiveExitCode, model: candidate ?? run.model, error, structuredOutput, runtimeAcknowledgedExtensions, ...(step.agentContract ? { agentContract: step.agentContract } : {}), ...(fileMutationEffect || settlementDiagnostic ? { effects: { ...(fileMutationEffect ? { fileMutation: fileMutationEffect } : {}), ...(settlementDiagnostic ? { settlementDiagnostic } : {}) } } : {}) };
+        finalResult = { ...run, exitCode: effectiveExitCode, model: candidate ?? run.model, error, structuredOutput, structuredOutputFailed: structuredError ? true : undefined, runtimeAcknowledgedExtensions, ...(step.agentContract ? { agentContract: step.agentContract } : {}), ...(fileMutationEffect || settlementDiagnostic ? { effects: { ...(fileMutationEffect ? { fileMutation: fileMutationEffect } : {}), ...(settlementDiagnostic ? { settlementDiagnostic } : {}) } } : {}) };
         if (run.stopped || run.timedOut || ctx.timeoutSignal?.aborted || ctx.stopSignal?.aborted || ctx.skipAcceptance?.())
             break singleLaunch;
         if (effectiveExitCode === 0 && !error)
@@ -1262,9 +1266,10 @@ export async function runSingleStepInner(step, ctx) {
         toolBudget,
         toolBudgetBlocked: toolBudgetBlocked || undefined,
         ...(finalResult?.effects ? { effects: finalResult.effects } : {}),
-        structuredOutput: timedOutAfterAcceptance || stoppedAfterAcceptance ? undefined : finalResult?.structuredOutput,
-        structuredOutputPath: timedOutAfterAcceptance || stoppedAfterAcceptance ? undefined : effectiveStructuredOutput?.outputPath,
-        structuredOutputSchemaPath: timedOutAfterAcceptance || stoppedAfterAcceptance ? undefined : effectiveStructuredOutput?.schemaPath,
+        structuredOutput: finalResult?.structuredOutput,
+        structuredOutputFailed: finalResult?.structuredOutputFailed,
+        structuredOutputPath: effectiveStructuredOutput?.outputPath,
+        structuredOutputSchemaPath: effectiveStructuredOutput?.schemaPath,
         acceptance: effectiveAcceptance,
         watchdog: finalResult?.watchdog,
         ...(capabilityAudit ? { capabilityCeiling: capabilityAudit.ceiling, capabilityAudit } : {}),
@@ -1670,6 +1675,7 @@ export async function runSubagent(config, childSessions) {
         ...(config.toolBudget ? { toolBudget: initialToolBudgetState(config.toolBudget) } : {}),
         ...(config.usageBudget ? { usageBudget: usageBudgetState(config.usageBudget, undefined) } : {}),
         pid: process.pid,
+        pidNamespaceScope: currentPidNamespaceScope(),
         cwd,
         currentStep: 0,
         chainStepCount: steps.length,
@@ -3594,6 +3600,7 @@ export async function runSubagent(config, childSessions) {
                     review: pr.review,
                     timeoutRecovery: pr.timeoutRecovery,
                     structuredOutput: pr.structuredOutput,
+                    structuredOutputFailed: pr.structuredOutputFailed,
                     structuredOutputPath: pr.structuredOutputPath,
                     structuredOutputSchemaPath: pr.structuredOutputSchemaPath,
                     acceptance: pr.acceptance,
@@ -4043,6 +4050,7 @@ export async function runSubagent(config, childSessions) {
                         review: pr.review,
                         timeoutRecovery: pr.timeoutRecovery,
                         structuredOutput: pr.structuredOutput,
+                        structuredOutputFailed: pr.structuredOutputFailed,
                         structuredOutputPath: pr.structuredOutputPath,
                         structuredOutputSchemaPath: pr.structuredOutputSchemaPath,
                         acceptance: pr.acceptance,
@@ -4348,6 +4356,7 @@ export async function runSubagent(config, childSessions) {
                 review: singleResult.review,
                 timeoutRecovery: singleResult.timeoutRecovery,
                 structuredOutput: singleResult.structuredOutput,
+                structuredOutputFailed: singleResult.structuredOutputFailed,
                 structuredOutputPath: singleResult.structuredOutputPath,
                 structuredOutputSchemaPath: singleResult.structuredOutputSchemaPath,
                 acceptance: singleResult.acceptance,
@@ -4758,6 +4767,7 @@ export async function runSubagent(config, childSessions) {
                 review: r.review,
                 effects: r.effects,
                 structuredOutput: r.structuredOutput,
+                structuredOutputFailed: r.structuredOutputFailed,
                 structuredOutputPath: r.structuredOutputPath,
                 structuredOutputSchemaPath: r.structuredOutputSchemaPath,
                 acceptance: r.acceptance,
@@ -4875,178 +4885,46 @@ export async function runSubagent(config, childSessions) {
         }
     }
 }
-async function waitForStartupControl(controlPath, token, action, timeoutMs = 30_000) {
-    const deadline = Date.now() + timeoutMs;
-    while (Date.now() <= deadline) {
-        if (fs.existsSync(controlPath)) {
-            let payload;
-            try {
-                payload = JSON.parse(fs.readFileSync(controlPath, "utf-8"));
-            }
-            catch (error) {
-                throw new Error(`Failed to read runner startup control '${controlPath}': ${error instanceof Error ? error.message : String(error)}`);
-            }
-            if (payload.token !== token)
-                throw new Error("Runner startup control token does not match.");
-            if (payload.action === action)
-                return;
-            if (payload.action !== "ack" && payload.action !== "confirm" && payload.action !== "proceed")
-                throw new Error("Runner startup control action is invalid.");
-        }
-        await new Promise((resolve) => setTimeout(resolve, 20));
-    }
-    throw new Error(`Timed out after ${timeoutMs}ms waiting for runner startup control '${action}'.`);
-}
-// Both hosts enter here: startup authorization, revival leases and disposal stay shared.
-export async function runConfiguredSubagent(config, options) {
-    let lease;
-    let startupCommitted = config.revivalLease === undefined && config.launchBarrierToken === undefined;
-    const startupPath = path.join(config.asyncDir, "runner-startup.json");
-    const startupAckPath = path.join(config.asyncDir, "runner-startup-ack.json");
-    const startupConfirmPath = path.join(config.asyncDir, "runner-startup-confirm.json");
-    const startupProceedPath = path.join(config.asyncDir, "runner-startup-proceed.json");
-    const releaseOnExit = () => {
-        try {
-            lease?.release();
-        }
-        catch {
-            // Exit cleanup is best effort; a dead-owner lease is reclaimed on the next revival.
-        }
-    };
-    process.once("exit", releaseOnExit);
+/** Heavy execution entry loaded by the bootstrap only after startup commits. */
+export async function runConfiguredSubagentExecution(config, options) {
+    let childSessions;
     try {
-        if (config.launchBarrierToken) {
-            await waitForStartupControl(startupProceedPath, config.launchBarrierToken, "proceed");
-            startupCommitted = true;
-            try {
-                fs.rmSync(startupProceedPath, { force: true });
-            }
-            catch {
-                // Startup control cleanup is best effort after the parent commits the run.
-            }
-        }
-        else if (config.revivalLease) {
-            lease = acquireSessionLease(config.revivalLease);
-            config.revivalLeaseToken = lease.owner.token;
-            writeAtomicJson(startupPath, { state: "ready", token: lease.owner.token, pid: process.pid, owner: lease.owner });
-            await waitForStartupControl(startupAckPath, lease.owner.token, "ack");
-            writeAtomicJson(startupPath, { state: "acknowledged", token: lease.owner.token, pid: process.pid });
-            await waitForStartupControl(startupConfirmPath, lease.owner.token, "confirm");
-            writeAtomicJson(startupPath, { state: "confirmed", token: lease.owner.token, pid: process.pid });
-            await waitForStartupControl(startupProceedPath, lease.owner.token, "proceed");
-            startupCommitted = true;
-            for (const controlPath of [startupAckPath, startupConfirmPath, startupProceedPath]) {
-                try {
-                    fs.rmSync(controlPath, { force: true });
-                }
-                catch {
-                    // Startup control cleanup is best effort after the parent commits the run.
-                }
-            }
-        }
-        const childSessions = await loadRunnerChildSessionFactory(config, options);
-        try {
-            await runSubagent(config, childSessions);
-        }
-        finally {
-            try {
-                await childSessions.dispose();
-            }
-            catch (error) {
-                console.error("Failed to dispose runner child sessions:", error);
-            }
-        }
+        // Detached Node runners do not receive Pi's CLI dispatcher setup. Binary
+        // hosts install the same dispatcher before entering the shared bootstrap.
+        if (!options?.loadPiCodingAgent)
+            installRunnerHttpDispatcher({ agentDir: getAgentDir(), cwd: process.cwd() });
+        childSessions = await loadRunnerChildSessionFactory(config, options);
     }
     catch (error) {
-        if (!startupCommitted) {
-            try {
-                writeAtomicJson(startupPath, { state: "error", pid: process.pid, error: error instanceof Error ? error.message : String(error) });
-            }
-            catch {
-                // The parent will time out and terminate this runner if the handshake cannot be written.
-            }
+        try {
+            persistRunnerStartupFailure({
+                asyncDir: config.asyncDir,
+                runId: config.id,
+                runnerProcessInstanceId: config.runnerProcessInstanceId ?? "unknown-runner-instance",
+                message: `Subagent runner startup failed: ${error instanceof Error ? error.message : String(error)}`,
+                ...(config.sessionId ? { sessionId: config.sessionId } : {}),
+                ...(config.completionOwnerId ? { completionOwnerId: config.completionOwnerId } : {}),
+                candidate: {
+                    ...(config.revivalLease?.sessionFile ? { sessionFile: config.revivalLease.sessionFile } : {}),
+                    ...(config.revivalLeaseToken ? { revivalLeaseToken: config.revivalLeaseToken } : {}),
+                },
+            });
+        }
+        catch (persistenceError) {
+            console.error("Failed to persist runner setup failure:", persistenceError);
         }
         throw error;
     }
-    finally {
-        process.off("exit", releaseOnExit);
-        if (lease) {
-            let acknowledged = false;
-            try {
-                acknowledged = lease.release();
-            }
-            catch (error) {
-                console.error("Failed to release session revival lease:", error);
-            }
-            try {
-                markProcessTerminalCandidateLeaseRelease(config.asyncDir, lease.owner.token, acknowledged);
-            }
-            catch (error) {
-                console.error("Failed to record session revival lease release:", error);
-            }
-        }
+    try {
+        await runSubagent(config, childSessions);
     }
-}
-function startConfiguredSubagent(config) {
-    // Child sessions and the extensions loaded into them may leave handles
-    // behind even after shutdown; the run is fully persisted by now, so exit
-    // explicitly instead of waiting for the event loop to drain.
-    runConfiguredSubagent(config).then(() => process.exit(0), (runErr) => {
-        console.error("Subagent runner error:", runErr);
-        process.exit(1);
-    });
-}
-function monitorTestParent() {
-    const parentPid = Number(process.env.PI_SUBAGENTS_TEST_PARENT_PID);
-    if (!Number.isSafeInteger(parentPid) || parentPid <= 0 || parentPid === process.pid)
-        return;
-    const check = () => {
+    finally {
         try {
-            process.kill(parentPid, 0);
-        }
-        catch {
-            process.exit(1);
+            await childSessions.dispose();
         }
-    };
-    check();
-    setInterval(check, 250).unref();
-}
-if (isRunnerEntrypoint) {
-    monitorTestParent();
-    const configArg = process.argv[2];
-    if (configArg) {
-        try {
-            const configJson = fs.readFileSync(configArg, "utf-8");
-            const config = JSON.parse(configJson);
-            try {
-                fs.unlinkSync(configArg);
-            }
-            catch {
-                // Temp config cleanup is best effort.
-            }
-            startConfiguredSubagent(config);
-        }
-        catch (err) {
-            console.error("Subagent runner error:", err);
-            process.exit(1);
+        catch (error) {
+            console.error("Failed to dispose runner child sessions:", error);
         }
     }
-    else {
-        let input = "";
-        process.stdin.setEncoding("utf-8");
-        process.stdin.on("data", (chunk) => {
-            input += chunk;
-        });
-        process.stdin.on("end", () => {
-            try {
-                const config = JSON.parse(input);
-                startConfiguredSubagent(config);
-            }
-            catch (err) {
-                console.error("Subagent runner error:", err);
-                process.exit(1);
-            }
-        });
-    }
 }
 //# sourceMappingURL=subagent-runner.js.map