@opengsd/gsd-pi 1.2.0-dev.5457a158 → 1.2.0-dev.822c9439

package/src/resources/extensions/gsd/auto-verification.ts

@@ -15,7 +15,7 @@
 
 import type { ExtensionContext, ExtensionAPI } from "@gsd/pi-coding-agent";
 import { existsSync, mkdirSync, readFileSync, readdirSync, writeFileSync } from "node:fs";
-import { gsdProjectionRoot, resolveSliceFile, resolveSlicePath, resolveMilestoneFile } from "./paths.js";
+import { gsdProjectionRoot, resolveSliceFile, resolveSlicePath } from "./paths.js";
 import { resolveMilestoneValidationVerdict } from "./milestone-validation-verdict.js";
 import { resolveCanonicalMilestoneRoot } from "./worktree-manager.js";
 import { parseUnitId } from "./unit-id.js";
@@ -25,8 +25,6 @@ import { loadEffectiveGSDPreferences } from "./preferences.js";
 import type { GSDPreferences } from "./preferences-types.js";
 import { isClosedStatus } from "./status-guards.js";
 import { loadFile } from "./files.js";
-import { parseRoadmap } from "./parsers-legacy.js";
-import { isMilestoneComplete } from "./state.js";
 import {
   runVerificationGate,
   runVerificationGateForTargets,
@@ -52,6 +50,7 @@ import { getSlice } from "./gsd-db.js";
 import { getLedger } from "./metrics.js";
 import { getUnitCostSpikeAction, resolveUnitCostSpikeMultiplier } from "./auto-budget.js";
 import { formatPostUnitStatusCard } from "./auto-status-message.js";
+import { detectWebApp } from "./web-app-uat.js";
 
 export interface VerificationContext {
   s: AutoSession;
@@ -373,29 +372,13 @@ async function runValidateMilestonePostCheck(
  * DB-backed projects are authoritative (#4094 peer review); falls back to
  * roadmap parsing only when the DB is unavailable.
  */
-async function countIncompleteSlices(basePath: string, milestoneId: string): Promise<number> {
-  if (isDbAvailable()) {
-    const slices = getMilestoneSlices(milestoneId);
-    if (slices.length === 0) {
-      // No DB rows — treat as "unknown", do not pause.
-      return 1;
-    }
-    return slices.filter((slice) => !isClosedStatus(slice.status)).length;
-  }
-
-  // Filesystem fallback: parse the roadmap markdown.
-  try {
-    const roadmapFile = resolveMilestoneFile(basePath, milestoneId, "ROADMAP");
-    if (!roadmapFile) return 1;
-    const roadmapContent = await loadFile(roadmapFile);
-    if (!roadmapContent) return 1;
-    const roadmap = parseRoadmap(roadmapContent);
-    if (roadmap.slices.length === 0) return 1;
-    return isMilestoneComplete(roadmap) ? 0 : 1;
-  } catch {
-    // Parsing failures should not cause false-positive pauses.
-    return 1;
-  }
+async function countIncompleteSlices(_basePath: string, milestoneId: string): Promise<number> {
+  // DB-authoritative (ADR-017): no markdown fallback. DB unavailable or no
+  // rows means "unknown" — do not pause.
+  if (!isDbAvailable()) return 1;
+  const slices = getMilestoneSlices(milestoneId);
+  if (slices.length === 0) return 1;
+  return slices.filter((slice) => !isClosedStatus(slice.status)).length;
 }
 
 /**
@@ -805,17 +788,32 @@ export async function runPostUnitVerification(
       s.verificationRetryFailureHashes.delete(retryKey);
       s.pendingVerificationRetry = null;
       return "continue";
+    } else if (
+      verdict.reason === "no-host-checks" &&
+      taskAlreadyComplete &&
+      detectWebApp(s.basePath) &&
+      !result.runtimeErrors?.some((e) => e.blocking)
+    ) {
+      s.verificationRetryCount.delete(retryKey);
+      s.verificationRetryFailureHashes.delete(retryKey);
+      s.pendingVerificationRetry = null;
+      ctx.ui.notify(
+        "No task-level host verification command was found for a completed browser-facing task; continuing so slice UAT can verify the UI with browser tools.",
+        "warning",
+      );
+      return "continue";
     } else if (verdict.reason === "no-host-checks") {
       s.verificationRetryCount.delete(retryKey);
       s.verificationRetryFailureHashes.delete(retryKey);
       s.pendingVerificationRetry = null;
+      const pauseMessage = `Verification failed: ${verdict.failureContext}`;
       ctx.ui.notify(
-        "Verification gate FAILED — no runnable host-owned verification checks were discovered. Pausing for human review.",
+        `Verification gate FAILED — ${verdict.failureContext}`,
         "error",
       );
       process.stderr.write(`verification-gate: ${verdict.failureContext}\n`);
       await pauseAuto(ctx, pi, {
-        message: "Verification failed: no runnable host-owned verification checks were discovered.",
+        message: pauseMessage,
         category: "unknown",
       });
       return "pause";

package/src/resources/extensions/gsd/auto-worktree.ts

@@ -972,6 +972,14 @@ function _resolveIntegrationBranchForReuse(
   }
 }
 
+function safeCwd(fallback: string): string {
+  try {
+    return process.cwd();
+  } catch {
+    return fallback;
+  }
+}
+
 /**
  * When reusing an existing milestone branch, fast-forward it onto the
  * integration branch when that's safe (branch is a strict ancestor of
@@ -1155,7 +1163,7 @@ export function teardownAutoWorktree(
 
   const branch = autoWorktreeBranch(milestoneId);
   const { preserveBranch = false, preserveWorktree = false } = opts;
-  const previousCwd = process.cwd();
+  const previousCwd = safeCwd(originalBasePath);
 
   // Wrap the entire teardown body in a single try/finally so activeWorkspace
   // is ALWAYS cleared — even if process.chdir throws (e.g. originalBasePath
@@ -2229,6 +2237,11 @@ export function mergeMilestoneToMain(
       process.chdir(originalBasePath_);
     } catch (err) {
       logWarning("worktree", `chdir to project root after merge failed: ${err instanceof Error ? err.message : String(err)}`);
+      debugLog("mergeMilestoneToMain", {
+        phase: "post-merge-chdir-failed",
+        target: originalBasePath_,
+        error: err instanceof Error ? err.message : String(err),
+      });
     }
   };
 

package/src/resources/extensions/gsd/auto.ts

@@ -924,6 +924,14 @@ export function setCurrentDispatchedModelId(model: { provider: string; id: strin
   s.currentDispatchedModelId = model ? `${model.provider}/${model.id}` : null;
 }
 
+/**
+ * Update the active unit model after runtime recovery switches models mid-unit.
+ * The next session restore path reads this field before dispatching again.
+ */
+export function setCurrentUnitModelForRecovery(model: any | null): void {
+  s.currentUnitModel = model;
+}
+
 // Tool tracking — delegates to auto-tool-tracking.ts
 export function markToolStart(toolCallId: string, toolName?: string): void {
   _markToolStart(toolCallId, s.active, toolName);
@@ -1000,6 +1008,41 @@ export function stopAutoRemote(projectRoot: string): {
   }
 }
 
+/**
+ * Force-stop a remote auto-mode session before stealing its lock.
+ * The normal stop path stays SIGTERM-only so cooperative sessions can clean up;
+ * this path is only for the explicit "Force start" action.
+ */
+export function forceStopAutoRemote(projectRoot: string): {
+  found: boolean;
+  pid?: number;
+  error?: string;
+} {
+  const lock = readCrashLock(projectRoot);
+  if (!lock) return { found: false };
+
+  if (lock.pid === process.pid) {
+    clearLock(projectRoot);
+    return { found: false };
+  }
+
+  if (!isLockProcessAlive(lock)) {
+    clearLock(projectRoot);
+    return { found: false };
+  }
+
+  try {
+    process.kill(lock.pid, "SIGTERM");
+    if (isLockProcessAlive(lock)) {
+      process.kill(lock.pid, "SIGKILL");
+    }
+    clearLock(projectRoot);
+    return { found: true, pid: lock.pid };
+  } catch (err) {
+    return { found: false, error: (err as Error).message };
+  }
+}
+
 /**
  * Check if a remote auto-mode session is running (from a different process).
  * Reads the crash lock, checks PID liveness, and returns session details.
@@ -2400,7 +2443,7 @@ export async function startAuto(
     const pid = freshStartAssessment.lock?.pid;
     ctx.ui.notify(
       pid
-        ? `Another auto-mode session (PID ${pid}) appears to be running.\nStop it with \`kill ${pid}\` before starting a new session.`
+        ? `Another auto-mode session (PID ${pid}) appears to be running.\nRun \`/gsd stop\` for graceful shutdown, or choose "Force start" from \`/gsd auto\` to terminate it.`
         : "Another auto-mode session appears to be running.",
       "error",
     );

package/src/resources/extensions/gsd/blocked-models.ts

@@ -23,6 +23,15 @@ interface BlockedModelsFile {
   blocked: BlockedModelEntry[];
 }
 
+interface TemporaryBlockedModelEntry {
+  provider: string;
+  id: string;
+  reason: string;
+  blockedUntil: number;
+}
+
+const temporaryBlockedModels = new Map<string, TemporaryBlockedModelEntry>();
+
 function blockedModelsPath(basePath: string): string {
   return join(gsdRoot(basePath), "runtime", "blocked-models.json");
 }
@@ -31,6 +40,10 @@ function modelKey(provider: string, id: string): string {
   return `${provider.toLowerCase()}/${id.toLowerCase()}`;
 }
 
+function temporaryModelKey(basePath: string, provider: string, id: string): string {
+  return `${basePath}:${modelKey(provider, id)}`;
+}
+
 function readFileSafe(path: string): BlockedModelsFile {
   if (!existsSync(path)) return { version: 1, blocked: [] };
   try {
@@ -66,6 +79,42 @@ export function isModelBlocked(
   );
 }
 
+export function blockModelUntil(
+  basePath: string,
+  provider: string,
+  id: string,
+  blockedUntil: number,
+  reason: string,
+): void {
+  const key = temporaryModelKey(basePath, provider, id);
+  if (blockedUntil <= Date.now()) {
+    temporaryBlockedModels.delete(key);
+    return;
+  }
+  temporaryBlockedModels.set(key, { provider, id, reason, blockedUntil });
+}
+
+export function isModelTemporarilyUnavailable(
+  basePath: string,
+  provider: string | undefined,
+  id: string | undefined,
+  now = Date.now(),
+): boolean {
+  if (!provider || !id) return false;
+  const key = temporaryModelKey(basePath, provider, id);
+  const entry = temporaryBlockedModels.get(key);
+  if (!entry) return false;
+  if (entry.blockedUntil <= now) {
+    temporaryBlockedModels.delete(key);
+    return false;
+  }
+  return true;
+}
+
+export function clearTemporaryModelBlocksForTest(): void {
+  temporaryBlockedModels.clear();
+}
+
 export function blockModel(
   basePath: string,
   provider: string,

package/src/resources/extensions/gsd/bootstrap/agent-end-recovery.ts

@@ -19,6 +19,7 @@ import {
   isAutoCompletionStopInProgress,
   pauseAuto,
   setCurrentDispatchedModelId,
+  setCurrentUnitModelForRecovery,
 } from "../auto.js";
 import { getNextFallbackModel, resolveModelWithFallbacksForUnit } from "../preferences.js";
 import { pauseAutoForProviderError } from "../provider-error-pause.js";
@@ -41,7 +42,7 @@ import {
   isTransient,
   type ErrorClass,
 } from "../error-classifier.js";
-import { blockModel, isModelBlocked } from "../blocked-models.js";
+import { blockModel, blockModelUntil, isModelBlocked, isModelTemporarilyUnavailable } from "../blocked-models.js";
 import { getProjectGSDPreferencesPath } from "../preferences.js";
 import { resolveProviderErrorGuidance } from "../provider-error-guidance.js";
 import { formatGuidance } from "../guidance.js";
@@ -143,9 +144,14 @@ async function tryProviderModelFallback(params: ProviderModelFallbackParams): Pr
       const nextModelId = getNextFallbackModel(cursorModelId, modelConfig);
       if (!nextModelId) break;
       const candidate = resolveModelId(nextModelId, availableModels, rejectedProvider);
-      if (candidate && !isModelBlocked(basePath, candidate.provider, candidate.id)) {
+      if (
+        candidate &&
+        !isModelBlocked(basePath, candidate.provider, candidate.id) &&
+        !isModelTemporarilyUnavailable(basePath, candidate.provider, candidate.id)
+      ) {
         const ok = await pi.setModel(candidate, { persist: false });
         if (ok) {
+          setCurrentUnitModelForRecovery(candidate);
           setCurrentDispatchedModelId({ provider: candidate.provider, id: candidate.id });
           switchedNotify(`${candidate.provider}/${candidate.id}`);
           pi.sendMessage(
@@ -163,7 +169,8 @@ async function tryProviderModelFallback(params: ProviderModelFallbackParams): Pr
   if (
     sessionModel &&
     !(sessionModel.provider === rejectedProvider && sessionModel.id === rejectedId) &&
-    !isModelBlocked(basePath, sessionModel.provider, sessionModel.id)
+    !isModelBlocked(basePath, sessionModel.provider, sessionModel.id) &&
+    !isModelTemporarilyUnavailable(basePath, sessionModel.provider, sessionModel.id)
   ) {
     const startModel = availableModels.find(
       (m) => m.provider === sessionModel.provider && m.id === sessionModel.id,
@@ -171,6 +178,7 @@ async function tryProviderModelFallback(params: ProviderModelFallbackParams): Pr
     if (startModel) {
       const ok = await pi.setModel(startModel, { persist: false });
       if (ok) {
+        setCurrentUnitModelForRecovery(startModel);
         setCurrentDispatchedModelId({ provider: startModel.provider, id: startModel.id });
         switchedNotify(`${startModel.provider}/${startModel.id}`);
         pi.sendMessage(
@@ -676,6 +684,16 @@ export async function handleAgentEnd(
       if (currentProvider === "openai-codex" || currentProvider === "google-gemini-cli") {
         cls.retryAfterMs = Math.min(cls.retryAfterMs, 30_000);
       }
+      const dash = getAutoDashboardData();
+      if (dash.basePath && ctx.model?.provider && ctx.model?.id) {
+        blockModelUntil(
+          dash.basePath,
+          ctx.model.provider,
+          ctx.model.id,
+          Date.now() + cls.retryAfterMs,
+          rawErrorMsg || displayMsg || "rate limit",
+        );
+      }
     }
 
     // ── 2. Decide & Act ──────────────────────────────────────────────────
@@ -721,9 +739,14 @@ export async function handleAgentEnd(
             retryState.networkRetryCount = 0;
             retryState.currentRetryModelId = undefined;
             const modelToSet = resolveModelId(nextModelId, availableModels, ctx.model?.provider);
-            if (modelToSet) {
+            const modelUnavailable = dash.basePath && modelToSet
+              ? isModelBlocked(dash.basePath, modelToSet.provider, modelToSet.id) ||
+                isModelTemporarilyUnavailable(dash.basePath, modelToSet.provider, modelToSet.id)
+              : false;
+            if (modelToSet && !modelUnavailable) {
               const ok = await pi.setModel(modelToSet, { persist: false });
               if (ok) {
+                setCurrentUnitModelForRecovery(modelToSet);
                 setCurrentDispatchedModelId({ provider: modelToSet.provider, id: modelToSet.id });
                 ctx.ui.notify(`Model error${errorDetail}. Switched to fallback: ${nextModelId} and resuming.`, "warning");
                 pi.sendMessage({ customType: "gsd-auto-timeout-recovery", content: "Continue execution.", display: false }, { triggerTurn: true });
@@ -737,11 +760,17 @@ export async function handleAgentEnd(
       // Try restoring session model
       const sessionModel = getAutoModeStartModel();
       if (sessionModel) {
-        if (ctx.model?.id !== sessionModel.id || ctx.model?.provider !== sessionModel.provider) {
+        const dash = getAutoDashboardData();
+        const sessionModelUnavailable = dash.basePath
+          ? isModelBlocked(dash.basePath, sessionModel.provider, sessionModel.id) ||
+            isModelTemporarilyUnavailable(dash.basePath, sessionModel.provider, sessionModel.id)
+          : false;
+        if (!sessionModelUnavailable && (ctx.model?.id !== sessionModel.id || ctx.model?.provider !== sessionModel.provider)) {
           const startModel = ctx.modelRegistry.getAvailable().find((m) => m.provider === sessionModel.provider && m.id === sessionModel.id);
           if (startModel) {
             const ok = await pi.setModel(startModel, { persist: false });
             if (ok) {
+              setCurrentUnitModelForRecovery(startModel);
               setCurrentDispatchedModelId({ provider: startModel.provider, id: startModel.id });
               retryState.networkRetryCount = 0;
               retryState.currentRetryModelId = undefined;

package/src/resources/extensions/gsd/bootstrap/dynamic-tools.ts

@@ -7,7 +7,6 @@ import { join } from "node:path";
 import type { ExtensionAPI } from "@gsd/pi-coding-agent";
 import { createBashTool, createEditTool, createReadTool, createWriteTool } from "@gsd/pi-coding-agent";
 
-import { DEFAULT_BASH_TIMEOUT_SECS } from "../constants.js";
 import { logWarning } from "../workflow-logger.js";
 import { openWorkflowDatabase } from "../db-workspace.js";
 import { getAutoWorktreePath } from "../auto-worktree.js";
@@ -87,8 +86,30 @@ export function registerDynamicTools(pi: ExtensionAPI): void {
   const baseBash = createBashTool(fallbackRoot, {
     spawnHook: (ctx) => ctx,
   });
+  // The auto-mode stalled-tool watchdog only exists in GSD/auto-mode, so the
+  // watchdog verbiage is injected here (the GSD-registered tool) rather than in
+  // core bash.ts, which is reused by non-GSD embeddings that have no watchdog.
+  const WATCHDOG_DETAIL =
+    "Genuine hangs are caught by the auto-mode stalled-tool watchdog (stalled: 5m / idle: 10m / soft: 20m / hard: 30m).";
+  const gsdBashDescription = `${(baseBash as any).description} ${WATCHDOG_DETAIL}`;
+  const gsdBashParameters = (() => {
+    const params: any = (baseBash as any).parameters;
+    if (!params?.properties?.timeout) return params;
+    return {
+      ...params,
+      properties: {
+        ...params.properties,
+        timeout: {
+          ...params.properties.timeout,
+          description: `${params.properties.timeout.description} ${WATCHDOG_DETAIL}`,
+        },
+      },
+    };
+  })();
   const dynamicBash = {
     ...baseBash,
+    description: gsdBashDescription,
+    parameters: gsdBashParameters,
     execute: async (
       toolCallId: string,
       params: { command: string; timeout?: number },
@@ -100,11 +121,7 @@ export function registerDynamicTools(pi: ExtensionAPI): void {
       const fresh = createBashTool(basePath, {
         spawnHook: (spawnCtx) => ({ ...spawnCtx, cwd: basePath }),
       });
-      const paramsWithTimeout = {
-        ...params,
-        timeout: params.timeout ?? DEFAULT_BASH_TIMEOUT_SECS,
-      };
-      return (fresh as any).execute(toolCallId, paramsWithTimeout, signal, onUpdate, ctx);
+      return (fresh as any).execute(toolCallId, params, signal, onUpdate, ctx);
     },
   };
   pi.registerTool(dynamicBash as any);

package/src/resources/extensions/gsd/bootstrap/exec-tools.ts

@@ -137,8 +137,8 @@ export function registerExecTools(pi: ExtensionAPI): void {
     parameters: Type.Object({
       query: Type.Optional(Type.String({ description: "Substring matched against id and purpose (case-insensitive)." })),
       runtime: Type.Optional(
-        Type.Union([Type.Literal("bash"), Type.Literal("node"), Type.Literal("python")], {
-          description: "Restrict to one runtime.",
+        Type.String({
+          description: "Restrict to one runtime: bash, node, or python.",
         }),
       ),
       failing_only: Type.Optional(Type.Boolean({ description: "Only non-zero exit codes and timeouts." })),

package/src/resources/extensions/gsd/bootstrap/register-extension.ts

@@ -17,6 +17,7 @@ import { registerHooks } from "./register-hooks.js";
 import { registerShortcuts } from "./register-shortcuts.js";
 import { writeCrashLog } from "./crash-log.js";
 import { logWarning } from "../workflow-logger.js";
+import { UNIT_TOOL_CONTRACTS } from "../unit-tool-contracts.js";
 // Static import so cmux event listeners are registered synchronously during
 // extension bootstrap. Prior implementation used `void import().then()` which
 // queued listener registration as a microtask — any CMUX_CHANNELS emit fired
@@ -37,6 +38,12 @@ const EPIPE_STORM_WINDOW_MS = 10_000;
 let epipeCount = 0;
 let epipeWindowStart = 0;
 
+export const CRITICAL_GSD_WORKFLOW_TOOL_NAMES = [...new Set(
+  Object.values(UNIT_TOOL_CONTRACTS)
+    .flatMap((contract) => contract.requiredWorkflowTools)
+    .filter((toolName) => toolName.startsWith("gsd_")),
+)].sort();
+
 /** Write to stderr without ever re-throwing — stderr can EPIPE too, which would
  *  re-enter this handler and re-loop. */
 function safeStderr(msg: string): void {
@@ -133,6 +140,21 @@ export function installEpipeGuard(): void {
   }
 }
 
+function assertCriticalGsdWorkflowToolsRegistered(pi: ExtensionAPI): void {
+  if (typeof pi.getAllTools !== "function") return;
+
+  const registered = new Set(pi.getAllTools().map((tool) => tool.name));
+  const missing = CRITICAL_GSD_WORKFLOW_TOOL_NAMES.filter((toolName) => !registered.has(toolName));
+  if (missing.length === 0) return;
+
+  const message = [
+    `Critical GSD workflow tool registration failed; missing required tool(s): ${missing.join(", ")}.`,
+    "Check earlier bootstrap warnings for the registration slot that failed.",
+  ].join(" ");
+  logWarning("bootstrap", message);
+  throw new Error(message);
+}
+
 export function registerGsdExtension(pi: ExtensionAPI): void {
   // Note: registerGSDCommand is called by index.ts before this function,
   // so we intentionally skip it here to avoid double-registration.
@@ -214,4 +236,6 @@ export function registerGsdExtension(pi: ExtensionAPI): void {
       );
     }
   }
+
+  assertCriticalGsdWorkflowToolsRegistered(pi);
 }