gsd-pi 2.76.0-dev.fe143342a → 2.77.0-dev.1d17f366c

package/src/resources/extensions/gsd/auto/turn-epoch.ts

@@ -0,0 +1,108 @@
+/**
+ * auto/turn-epoch.ts — Turn generation counter + AsyncLocalStorage-backed
+ * capture for stale-turn write dropping.
+ *
+ * Problem: when auto-timeout-recovery synthetically resolves a timed-out
+ * unit so the loop can advance, the original LLM turn keeps running in the
+ * background. Its subsequent writes (journal events, audit events, tool
+ * calls that flow through closeout) then race the replacement unit's
+ * writes. DB-level guards (complete-task/complete-slice) block double
+ * state transitions, but journal/audit/closeout side-effects still fire
+ * with fresh identifiers and pollute forensics.
+ *
+ * Containment: every time we decide a turn is done (timeout recovery,
+ * explicit cancellation), bump a module-level generation counter.
+ * Turn-aware call sites wrap their body in `runWithTurnGeneration`, which
+ * captures the generation into AsyncLocalStorage. Write sites deep in the
+ * stack call `isStaleWrite` — if the captured generation is older than
+ * current, the turn has been superseded and the write is dropped.
+ *
+ * Failure mode: if AsyncLocalStorage context is lost across some exotic
+ * async boundary (e.g. a native-side worker callback), the write site sees
+ * `no-store` and falls through to current behavior — the write proceeds
+ * normally. That is a safe default; the correctness regression is only
+ * "noisier forensics under rare boundary loss," not duplicated state.
+ */
+
+import { AsyncLocalStorage } from "node:async_hooks";
+
+import { debugLog } from "../debug-logger.js";
+
+let _currentGeneration = 0;
+
+const turnContext = new AsyncLocalStorage<{ capturedGen: number }>();
+
+/** Current turn generation. Mutated only by bumpTurnGeneration. */
+export function getCurrentTurnGeneration(): number {
+  return _currentGeneration;
+}
+
+/**
+ * Bump the turn generation and return the new value. Every caller should
+ * pass a short `reason` string so forensics can reconstruct why a given
+ * turn was marked stale.
+ */
+export function bumpTurnGeneration(reason: string): number {
+  _currentGeneration += 1;
+  debugLog("turnEpoch.bump", { reason, newGeneration: _currentGeneration });
+  return _currentGeneration;
+}
+
+/**
+ * Run fn() with `capturedGen` attached to AsyncLocalStorage so that any
+ * write site reached from within fn() can check for staleness without
+ * parameter threading.
+ */
+export function runWithTurnGeneration<T>(capturedGen: number, fn: () => T): T {
+  return turnContext.run({ capturedGen }, fn);
+}
+
+/**
+ * True when the current async context was started at a turn generation
+ * older than the current one — meaning the turn has been superseded by
+ * recovery/cancellation since it began.
+ *
+ * Returns false when there is no captured generation (e.g. the write is
+ * happening outside any wrapped turn). That is the safe default: writes
+ * proceed as they did before this epoch was introduced.
+ */
+export function isStaleWrite(component?: string): boolean {
+  const store = turnContext.getStore();
+  if (!store) return false;
+  const captured = store.capturedGen;
+  const current = _currentGeneration;
+  if (captured < current) {
+    debugLog("turnEpoch.stale", {
+      component: component ?? "unknown",
+      captured,
+      current,
+    });
+    return true;
+  }
+  return false;
+}
+
+/**
+ * Snapshot of both the captured turn generation and the current one.
+ * Used by closeoutUnit to persist an orphan-marker entry instead of
+ * silently skipping the full closeout on a stale turn.
+ */
+export function describeTurnEpoch(): {
+  captured: number | null;
+  current: number;
+  stale: boolean;
+} {
+  const store = turnContext.getStore();
+  const captured = store?.capturedGen ?? null;
+  const current = _currentGeneration;
+  return {
+    captured,
+    current,
+    stale: captured !== null && captured < current,
+  };
+}
+
+/** Test helper — resets module state so tests start from a known baseline. */
+export function _resetTurnEpoch(): void {
+  _currentGeneration = 0;
+}

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

@@ -9,7 +9,7 @@ import type { ExtensionAPI, ExtensionContext } from "@gsd/pi-coding-agent";
 import type { AutoSession } from "./session.js";
 import type { GSDPreferences } from "../preferences.js";
 import type { GSDState } from "../types.js";
-import type { CmuxLogLevel } from "../../cmux/index.js";
+import type { CmuxLogLevel } from "../../shared/cmux-events.js";
 import type { LoopDeps } from "./loop-deps.js";
 
 /**

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

@@ -34,6 +34,7 @@ import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
 import { logWarning, logError } from "./workflow-logger.js";
 import { join } from "node:path";
 import { hasImplementationArtifacts } from "./auto-recovery.js";
+import { classifyMilestoneSummaryContent } from "./milestone-summary-classifier.js";
 import {
   buildDiscussMilestonePrompt,
   buildResearchMilestonePrompt,
@@ -58,6 +59,7 @@ import {
 import { resolveModelWithFallbacksForUnit } from "./preferences-models.js";
 import { resolveUokFlags } from "./uok/flags.js";
 import { selectReactiveDispatchBatch } from "./uok/execution-graph.js";
+import { EXECUTION_ENTRY_PHASES, hasFinalizedMilestoneContext } from "./uok/plan-v2.js";
 
 // ─── Types ────────────────────────────────────────────────────────────────
 
@@ -242,6 +244,38 @@ export const DISPATCH_RULES: DispatchRule[] = [
       };
     },
   },
+  {
+    // #4671 — Recovery path for execution-entry phases with missing CONTEXT.md.
+    //
+    // Once `deriveStateFromDb` returns an execution-entry phase (executing /
+    // summarizing / validating-milestone / completing-milestone), the
+    // pre-planning guard at `pre-planning (no context) → discuss-milestone`
+    // no longer fires. The plan-v2 gate correctly detects the missing context
+    // but can only block — it cannot redispatch. Without this rule the
+    // milestone is stuck until `/gsd doctor heal` repairs it (and heal
+    // historically missed this check too).
+    //
+    // Fire BEFORE the execution-entry phase rules so we redispatch to
+    // `discuss-milestone` instead of hitting the plan-v2 gate.
+    name: "execution-entry phase (no context) → discuss-milestone",
+    match: async ({ state, mid, midTitle, basePath, structuredQuestionsAvailable }) => {
+      if (!EXECUTION_ENTRY_PHASES.has(state.phase)) return null;
+      // Align with the plan-v2 gate's lookup semantics: whitespace-only counts
+      // as missing, and an auto worktree may fall back to GSD_PROJECT_ROOT.
+      if (hasFinalizedMilestoneContext(basePath, mid)) return null;
+      return {
+        action: "dispatch",
+        unitType: "discuss-milestone",
+        unitId: mid,
+        prompt: await buildDiscussMilestonePrompt(
+          mid,
+          midTitle,
+          basePath,
+          structuredQuestionsAvailable,
+        ),
+      };
+    },
+  },
   {
     name: "summarizing → complete-slice",
     match: async ({ state, mid, midTitle, basePath }) => {
@@ -421,6 +455,26 @@ export const DISPATCH_RULES: DispatchRule[] = [
       };
     },
   },
+  {
+    name: "planning (require_slice_discussion) → pause for discussion (#3454)",
+    match: async ({ state, mid, basePath, prefs }) => {
+      if (state.phase !== "planning") return null;
+      if (!prefs?.phases?.require_slice_discussion) return null;
+      if (!state.activeSlice) return null;
+      // Only pause if the slice has no context file yet (discussion not done).
+      // resolveSliceFile returns null when the file does not exist on disk,
+      // but cachedReaddir could return a stale hit — verify with existsSync
+      // so the guard is defence-in-depth and the contract is explicit at the
+      // call site.
+      const sliceContextFile = resolveSliceFile(basePath, mid, state.activeSlice.id, "CONTEXT");
+      if (sliceContextFile && existsSync(sliceContextFile)) return null; // discussion already done, proceed
+      return {
+        action: "stop" as const,
+        reason: `Slice ${state.activeSlice.id} requires discussion before planning (require_slice_discussion is enabled). Run /gsd discuss to discuss this slice, then /gsd auto to resume.`,
+        level: "info" as const,
+      };
+    },
+  },
   {
     // Keep this rule before the single-slice research rule so the multi-slice
     // path wins whenever 2+ slices are ready.
@@ -568,15 +622,28 @@ export const DISPATCH_RULES: DispatchRule[] = [
   },
   {
     name: "planning → plan-slice",
-    match: async ({ state, mid, midTitle, basePath, sessionContextWindow, modelRegistry }) => {
+    match: async ({ state, mid, midTitle, basePath, sessionContextWindow, modelRegistry, session }) => {
       if (state.phase !== "planning") return null;
       if (!state.activeSlice) return missingSliceStop(mid, state.phase);
       const sid = state.activeSlice!.id;
       const sTitle = state.activeSlice!.title;
+      // #4551: Consume any persisted pre-exec failure for this slice so the
+      // re-dispatched prompt includes the exact blocked references. Clear the
+      // field immediately after reading to prevent stale context leaking into
+      // a later, unrelated plan-slice run.
+      const unitId = `${mid}/${sid}`;
+      let priorPreExecFailure: { blockingFindings: string[]; verdictExcerpt: string } | undefined;
+      if (session?.lastPreExecFailure?.unitId === unitId) {
+        priorPreExecFailure = {
+          blockingFindings: session.lastPreExecFailure.blockingFindings,
+          verdictExcerpt: session.lastPreExecFailure.verdictExcerpt,
+        };
+        session.lastPreExecFailure = null;
+      }
       return {
         action: "dispatch",
         unitType: "plan-slice",
-        unitId: `${mid}/${sid}`,
+        unitId,
         prompt: await buildPlanSlicePrompt(
           mid,
           midTitle,
@@ -584,7 +651,7 @@ export const DISPATCH_RULES: DispatchRule[] = [
           sTitle,
           basePath,
           undefined,
-          { sessionContextWindow, modelRegistry },
+          { sessionContextWindow, modelRegistry, priorPreExecFailure },
         ),
       };
     },
@@ -870,21 +937,13 @@ export const DISPATCH_RULES: DispatchRule[] = [
         }
       }
 
-      // Reconciliation guard (#4324): when the SUMMARY file already exists
-      // on disk but the DB says the milestone is not complete, the DB is
-      // out of sync (e.g. journal reset, partial merge, crash recovery).
-      // Reconcile the DB status directly instead of re-dispatching the
-      // tool, which would overwrite the richer on-disk SUMMARY with a
-      // thinner regenerated version — causing silent data loss.
       const existingSummary = resolveMilestoneFile(basePath, mid, "SUMMARY");
-      if (existingSummary && isDbAvailable()) {
-        try {
-          updateMilestoneStatus(mid, "complete", new Date().toISOString());
-          logWarning("dispatch", `Milestone ${mid} has SUMMARY on disk but DB status was not complete — reconciled DB to complete (#4324)`);
-        } catch (err) {
-          logWarning("dispatch", `Failed to reconcile milestone ${mid} status: ${err instanceof Error ? err.message : String(err)}`);
+      let summaryOutcome: "success" | "failure" | "unknown" = "unknown";
+      if (existingSummary) {
+        const summaryContent = await loadFile(existingSummary);
+        if (summaryContent) {
+          summaryOutcome = classifyMilestoneSummaryContent(summaryContent);
         }
-        return { action: "skip" };
       }
 
       // Safety guard (#2675): block completion when VALIDATION verdict is
@@ -969,6 +1028,48 @@ export const DISPATCH_RULES: DispatchRule[] = [
         logWarning("dispatch", `verification class check failed: ${err instanceof Error ? err.message : String(err)}`);
       }
 
+      // Disk/DB mismatch handling (#4658): SUMMARY presence alone is not enough.
+      // Apply post-gate policy:
+      // - success summary: reconcile DB and skip re-dispatch
+      // - failure summary: pause/fail-closed
+      // - unknown summary: pause/fail-closed
+      if (existingSummary) {
+        const milestone = isDbAvailable() ? getMilestone(mid) : null;
+        const status = milestone?.status ?? (isDbAvailable() ? "missing" : "unavailable");
+
+        if (summaryOutcome === "success") {
+          if (!isDbAvailable()) {
+            logWarning("dispatch", `Milestone ${mid} SUMMARY indicates completion while DB is unavailable — skipping duplicate complete-milestone dispatch`);
+            return { action: "skip" };
+          }
+          try {
+            updateMilestoneStatus(mid, "complete", new Date().toISOString());
+            logWarning("dispatch", `Milestone ${mid} SUMMARY indicates completion while DB status was "${status}" — reconciled DB to complete (#4658)`);
+            return { action: "skip" };
+          } catch (err) {
+            return {
+              action: "stop",
+              level: "warning",
+              reason: `Milestone ${mid} SUMMARY indicates completion but DB reconciliation failed (${err instanceof Error ? err.message : String(err)}). Auto-mode paused for manual review.`,
+            };
+          }
+        }
+
+        if (summaryOutcome === "failure") {
+          return {
+            action: "stop",
+            level: "warning",
+            reason: `Milestone ${mid} has a failure-path SUMMARY while DB status is "${status}". Auto-mode will not promote completion from failure artifacts. Re-run complete-milestone only after blockers are resolved and verification passes.`,
+          };
+        }
+
+        return {
+          action: "stop",
+          level: "warning",
+          reason: `Milestone ${mid} has an ambiguous SUMMARY while DB status is "${status}". Auto-mode paused instead of promoting completion from file presence alone.`,
+        };
+      }
+
       return {
         action: "dispatch",
         unitType: "complete-milestone",

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

@@ -9,7 +9,7 @@
 
 export { autoLoop, runUokKernelLoop, runLegacyAutoLoop } from "./auto/loop.js";
 export { isInfrastructureError, INFRA_ERROR_CODES } from "./auto/infra-errors.js";
-export { resolveAgentEnd, resolveAgentEndCancelled, isSessionSwitchInFlight, _resetPendingResolve, _setActiveSession } from "./auto/resolve.js";
+export { resolveAgentEnd, resolveAgentEndCancelled, isSessionSwitchInFlight, _hasPendingResolveForTest, _resetPendingResolve, _setActiveSession } from "./auto/resolve.js";
 export { detectStuck } from "./auto/detect-stuck.js";
 export { runUnit } from "./auto/run-unit.js";
 export type { LoopDeps } from "./auto/loop-deps.js";

package/src/resources/extensions/gsd/auto-model-selection.ts

@@ -395,7 +395,7 @@ export async function selectAndApplyModel(
         // ADR-005: Adjust active tool set for the selected model's provider capabilities.
         // Hard-filter incompatible tools, then let extensions override via adjust_tool_set hook.
         const activeToolNames = pi.getActiveTools();
-        const { toolNames: compatibleTools, removedTools } = adjustToolSet(activeToolNames, model.api);
+        const { toolNames: compatibleTools, removedTools } = adjustToolSet(activeToolNames, model.api, model.provider);
         let finalToolNames = compatibleTools;
 
         // Fire adjust_tool_set hook — extensions can override the filtered tool set

package/src/resources/extensions/gsd/auto-post-unit.ts

@@ -55,7 +55,7 @@ import { hasPendingCaptures, loadPendingCaptures, revertExecutorResolvedCaptures
 import { debugLog } from "./debug-logger.js";
 import { runSafely } from "./auto-utils.js";
 import type { AutoSession, SidecarItem } from "./auto/session.js";
-import { getEvidence } from "./safety/evidence-collector.js";
+import { getEvidence, clearEvidenceFromDisk } from "./safety/evidence-collector.js";
 import { validateFileChanges } from "./safety/file-change-validator.js";
 // crossReferenceEvidence available for future use when verification_evidence is stored in DB
 // import { crossReferenceEvidence, type ClaimedEvidence } from "./safety/evidence-cross-ref.js";
@@ -70,9 +70,24 @@ import { ensureCodebaseMapFresh } from "./codebase-generator.js";
 import { resolveUokFlags } from "./uok/flags.js";
 import { UokGateRunner } from "./uok/gate-runner.js";
 import { writeTurnGitTransaction } from "./uok/gitops.js";
+import { isClosedStatus } from "./status-guards.js";
+import { detectAbandonMilestone } from "./abandon-detect.js";
 
 /** Maximum verification retry attempts before escalating to blocker placeholder (#2653). */
 const MAX_VERIFICATION_RETRIES = 3;
+const COMPLETE_MILESTONE_DB_SETTLE_MS = 1500;
+const COMPLETE_MILESTONE_DB_SETTLE_POLL_MS = 100;
+
+async function waitForMilestoneDbClose(mid: string): Promise<boolean> {
+  const deadline = Date.now() + COMPLETE_MILESTONE_DB_SETTLE_MS;
+  while (Date.now() < deadline) {
+    if (!isDbAvailable()) return false;
+    const milestone = getMilestone(mid);
+    if (milestone && isClosedStatus(milestone.status)) return true;
+    await new Promise((resolve) => setTimeout(resolve, COMPLETE_MILESTONE_DB_SETTLE_POLL_MS));
+  }
+  return false;
+}
 
 
 /** Enqueue a sidecar item (hook, triage, or quick-task) for the main loop to
@@ -107,7 +122,6 @@ import {
   updateProgressWidget as _updateProgressWidget,
   updateSliceProgressCache,
   unitVerb,
-  hideFooter,
   describeNextUnit,
 } from "./auto-dashboard.js";
 import { existsSync, unlinkSync } from "node:fs";
@@ -556,6 +570,35 @@ export async function postUnitPreVerification(pctx: PostUnitContext, opts?: PreV
     // Rewrite-docs completion
     if (s.currentUnit.type === "rewrite-docs") {
       await runSafely("postUnit", "rewrite-docs-resolve", async () => {
+        // Detect abandon/descope overrides BEFORE resolving them (#3490).
+        // If an override is about abandoning the milestone, park it so the
+        // state engine skips it. Without this, rewrite-docs only edits
+        // markdown but the DB still has the milestone as active.
+        try {
+          const { loadActiveOverrides } = await import("./files.js");
+          const overrides = await loadActiveOverrides(s.basePath);
+          const decision = detectAbandonMilestone(overrides, s.currentMilestoneId);
+          if (decision.shouldPark && s.currentMilestoneId) {
+            const { parkMilestone } = await import("./milestone-actions.js");
+            const parked = parkMilestone(s.basePath, s.currentMilestoneId, decision.reason);
+            if (parked) {
+              ctx.ui.notify(`Milestone ${s.currentMilestoneId} parked: "${decision.reason}"`, "info");
+            } else {
+              // Park refused: milestone directory missing, milestone already
+              // completed (SUMMARY present), or PARKED.md already exists.
+              // resolveAllOverrides below will still consume the override —
+              // surface this loudly so the user notices state drift rather
+              // than silently losing the abandon directive.
+              const msg = `Abandon detected for ${s.currentMilestoneId} but park refused (milestone is completed, already parked, or missing). Override will be resolved anyway — verify state is correct.`;
+              logError("engine", msg);
+              ctx.ui.notify(msg, "warning");
+            }
+          }
+        } catch (err) {
+          logError("engine", `abandon-detect failed: ${(err as Error).message}`);
+          ctx.ui.notify(`Abandon detection failed — check logs. Overrides will still be resolved.`, "warning");
+        }
+
         await resolveAllOverrides(s.basePath);
         // Reset both disk and in-memory counters. Disk counter is authoritative
         // (survives restarts); in-memory is kept in sync for the current session.
@@ -654,7 +697,7 @@ export async function postUnitPreVerification(pctx: PostUnitContext, opts?: PreV
             if (taskRow) {
               const expectedOutput = taskRow.expected_output ?? [];
               const plannedFiles = taskRow.files ?? [];
-              const audit = validateFileChanges(s.basePath, expectedOutput, plannedFiles);
+              const audit = validateFileChanges(s.basePath, expectedOutput, plannedFiles, safetyConfig.file_change_allowlist);
               if (audit && audit.violations.length > 0) {
                 const warnings = audit.violations.filter(v => v.severity === "warning");
                 for (const v of warnings) {
@@ -712,6 +755,16 @@ export async function postUnitPreVerification(pctx: PostUnitContext, opts?: PreV
             debugLog("postUnit", { phase: "safety-content-validation", error: String(e) });
           }
         }
+
+        // Clear persisted evidence file now that post-unit processing is complete
+        // (Bug #4385 — prevents stale evidence from affecting retries of same unit ID).
+        if (safetyConfig.evidence_collection && s.currentUnit.type === "execute-task" && sMid && sSid && sTid) {
+          try {
+            clearEvidenceFromDisk(s.basePath, sMid, sSid, sTid);
+          } catch (e) {
+            debugLog("postUnit", { phase: "safety-evidence-clear", error: String(e) });
+          }
+        }
       }
     } catch (e) {
       debugLog("postUnit", { phase: "safety-harness", error: String(e) });
@@ -731,6 +784,25 @@ export async function postUnitPreVerification(pctx: PostUnitContext, opts?: PreV
 
       // If verification failed, attempt to regenerate missing projection files
       // from DB data before giving up (e.g. research-slice produces PLAN from engine).
+      if (!triggerArtifactVerified) {
+        if (s.currentUnit.type === "complete-milestone") {
+          try {
+            const { milestone: mid } = parseUnitId(s.currentUnit.id);
+            if (mid) {
+              const settled = await waitForMilestoneDbClose(mid);
+              if (settled) {
+                triggerArtifactVerified = verifyExpectedArtifact(s.currentUnit.type, s.currentUnit.id, s.basePath);
+                if (triggerArtifactVerified) {
+                  invalidateAllCaches();
+                }
+              }
+            }
+          } catch (e) {
+            debugLog("postUnit", { phase: "artifact-verify-settle-db", error: String(e) });
+          }
+        }
+      }
+
       if (!triggerArtifactVerified) {
         try {
           const { milestone: mid, slice: sid } = parseUnitId(s.currentUnit.id);
@@ -1134,6 +1206,15 @@ export async function postUnitPostVerification(pctx: PostUnitContext): Promise<"
             `Pre-execution checks failed: ${blockingCount} blocking issue${blockingCount === 1 ? "" : "s"} found\n${details}${suffix}${evidenceNote}`,
             "error",
           );
+          // Persist failure context so the next plan-slice re-dispatch can inject
+          // it into the prompt and break the infinite loop (#4551).
+          s.lastPreExecFailure = {
+            unitId: currentUnit.id,
+            blockingFindings: blockingChecks.map(
+              c => `[${c.category}] ${c.target}: ${c.message}`,
+            ),
+            verdictExcerpt: `status=${result.status}; ${blockingCount} blocking issue${blockingCount === 1 ? "" : "s"} detected`,
+          };
           preExecPauseNeeded = true;
         } else if (result.status === "warn") {
           ctx.ui.notify(
@@ -1142,6 +1223,14 @@ export async function postUnitPostVerification(pctx: PostUnitContext): Promise<"
           );
           // Strict mode: treat warnings as blocking
           if (prefs?.enhanced_verification_strict === true) {
+            const warnChecks = result.checks.filter(c => !c.passed);
+            s.lastPreExecFailure = {
+              unitId: currentUnit.id,
+              blockingFindings: warnChecks.map(
+                c => `[${c.category}] ${c.target}: ${c.message}`,
+              ),
+              verdictExcerpt: `status=${result.status} (strict mode); ${warnChecks.length} warning${warnChecks.length === 1 ? "" : "s"} treated as blocking`,
+            };
             preExecPauseNeeded = true;
           }
         }

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

@@ -1380,7 +1380,18 @@ async function renderSlicePrompt(options: {
 
 export async function buildPlanSlicePrompt(
   mid: string, _midTitle: string, sid: string, sTitle: string, base: string, level?: InlineLevel,
-  options?: { softScopeHint?: string; sessionContextWindow?: number; modelRegistry?: MinimalModelRegistry },
+  options?: {
+    softScopeHint?: string;
+    sessionContextWindow?: number;
+    modelRegistry?: MinimalModelRegistry;
+    /** Failure context from a prior pre-exec gate run (#4551). When present, a
+     *  "Fix these specific issues" section is appended so the LLM addresses the
+     *  exact problems instead of producing an identical plan that fails again. */
+    priorPreExecFailure?: {
+      blockingFindings: string[];
+      verdictExcerpt: string;
+    };
+  },
 ): Promise<string> {
   const prependBlocks: string[] = [];
   // ADR-011: when the refining-phase dispatch rule gracefully downgrades to
@@ -1393,6 +1404,22 @@ export async function buildPlanSlicePrompt(
       `This scope was captured during an earlier progressive-planning pass that was later disabled. Treat it as context only — you may plan beyond it if the work genuinely requires more scope. Do NOT treat this as a hard boundary.`,
     );
   }
+  // #4551: inject pre-exec failure context so the re-dispatched plan-slice
+  // addresses the exact blocked references rather than reproducing the same plan.
+  if (options?.priorPreExecFailure) {
+    const { blockingFindings, verdictExcerpt } = options.priorPreExecFailure;
+    const findingsList = blockingFindings.length > 0
+      ? blockingFindings.map(f => `- ${f}`).join("\n")
+      : "- (no specific findings recorded)";
+    prependBlocks.push(
+      `## Fix these specific issues from the prior pre-exec check\n\n` +
+      `The previous plan-slice attempt was blocked by pre-execution validation.\n` +
+      `Gate verdict: ${verdictExcerpt}\n\n` +
+      `Blocked references that must be resolved in this plan:\n${findingsList}\n\n` +
+      `Revise the plan so that every reference listed above is satisfied before execution begins. ` +
+      `Do not reproduce the same file paths, package names, or task ordering that caused these failures.`,
+    );
+  }
   return renderSlicePrompt({
     mid, sid, sTitle, base,
     level: level ?? resolveInlineLevel(),

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

@@ -13,10 +13,11 @@ import { appendEvent } from "./workflow-events.js";
 import { atomicWriteSync } from "./atomic-write.js";
 import { clearParseCache } from "./files.js";
 import { parseRoadmap as parseLegacyRoadmap, parsePlan as parseLegacyPlan } from "./parsers-legacy.js";
-import { isDbAvailable, getTask, getSlice, getSliceTasks, getPendingGates, updateTaskStatus, updateSliceStatus } from "./gsd-db.js";
+import { isDbAvailable, getTask, getSlice, getSliceTasks, getPendingGates, updateTaskStatus, updateSliceStatus, insertSlice, getMilestone } from "./gsd-db.js";
 import { isValidationTerminal } from "./state.js";
 import { getErrorMessage } from "./error-utils.js";
 import { logWarning, logError } from "./workflow-logger.js";
+import { isClosedStatus } from "./status-guards.js";
 import {
   nativeConflictFiles,
   nativeCommit,
@@ -50,9 +51,14 @@ import {
   resolveExpectedArtifactPath,
   diagnoseExpectedArtifact,
 } from "./auto-artifact-paths.js";
+import { classifyMilestoneSummaryContent } from "./milestone-summary-classifier.js";
 
 // Re-export so existing consumers of auto-recovery.ts keep working.
 export { resolveExpectedArtifactPath, diagnoseExpectedArtifact };
+export {
+  classifyMilestoneSummaryContent,
+  type MilestoneSummaryOutcome,
+} from "./milestone-summary-classifier.js";
 
 // ─── Artifact Resolution & Verification ───────────────────────────────────────
 
@@ -268,11 +274,26 @@ export function verifyExpectedArtifact(
   // RESEARCH file. Without this, resolveExpectedArtifactPath returns null and
   // the retry/escalation machinery silently re-dispatches forever.
   //
+  // #4068: Also treat a PARALLEL-BLOCKER placeholder as a terminal completion
+  // so that timeout-recovery can write the blocker, have verifyExpectedArtifact
+  // return true, and let the dispatch loop advance past this unit.  Without
+  // this, the blocker is written but verification still returns false, the unit
+  // is never cleared from unitDispatchCount, and on the next iteration the
+  // dispatch rule (which correctly skips parallel-research when PARALLEL-BLOCKER
+  // exists) returns null — leaving the loop stuck re-deriving indefinitely.
+  //
   // NOTE: this predicate mirrors the dispatch rule at
   // auto-dispatch.ts parallel-research-slices — keep the two in sync.
   if (unitType === "research-slice" && unitId.endsWith("/parallel-research")) {
     const { milestone: mid } = parseUnitId(unitId);
     if (!mid) return false;
+
+    // #4068: PARALLEL-BLOCKER written by timeout-recovery is a terminal state.
+    const blockerPath = resolveExpectedArtifactPath(unitType, unitId, base);
+    if (blockerPath && existsSync(blockerPath)) {
+      return true;
+    }
+
     const roadmapFile = resolveMilestoneFile(base, mid, "ROADMAP");
     if (!roadmapFile || !existsSync(roadmapFile)) {
       logWarning("recovery", `verify-fail ${unitType} ${unitId}: roadmap missing`);
@@ -462,6 +483,14 @@ export function verifyExpectedArtifact(
   // A milestone with only .gsd/ plan files and zero implementation code is
   // not genuinely complete — the LLM wrote plan files but skipped actual work.
   if (unitType === "complete-milestone") {
+    const summaryOutcome = classifyMilestoneSummaryContent(readFileSync(absPath, "utf-8"));
+    if (summaryOutcome === "failure") return false;
+    const { milestone: mid } = parseUnitId(unitId);
+    if (mid && isDbAvailable()) {
+      const dbMilestone = getMilestone(mid);
+      if (!dbMilestone) return false;
+      if (!isClosedStatus(dbMilestone.status) && summaryOutcome !== "success") return false;
+    }
     if (hasImplementationArtifacts(base) === "absent") return false;
   }
 
@@ -532,6 +561,16 @@ export function writeBlockerPlaceholder(
       try { updateSliceStatus(mid, sid, "complete", ts); } catch (e) { logWarning("recovery", `updateSliceStatus failed during context exhaustion: ${e instanceof Error ? e.message : String(e)}`); }
       try { appendEvent(base, { cmd: "complete-slice", params: { milestoneId: mid, sliceId: sid }, ts, actor: "system", trigger_reason: "blocker-placeholder-recovery" }); } catch (e) { logWarning("recovery", `appendEvent failed for slice recovery: ${e instanceof Error ? e.message : String(e)}`); }
     }
+    // Insert a placeholder complete slice so deriveState sees activeMilestoneSlices.length > 0
+    // and exits the pre-planning phase. Without this, activeMilestoneSlices stays empty
+    // after the blocker ROADMAP.md is written, causing deriveState to return phase:'pre-planning'
+    // indefinitely and re-dispatching plan-milestone in an infinite loop (#4378).
+    if (unitType === "plan-milestone" && mid) {
+      try {
+        insertSlice({ id: "S00-blocker", milestoneId: mid, title: "Blocker placeholder — planning failed", status: "complete", sequence: 0 });
+      } catch (e) { logWarning("recovery", `insertSlice placeholder failed for plan-milestone recovery: ${e instanceof Error ? e.message : String(e)}`); }
+      try { appendEvent(base, { cmd: "plan-milestone", params: { milestoneId: mid }, ts, actor: "system", trigger_reason: "blocker-placeholder-recovery" }); } catch (e) { logWarning("recovery", `appendEvent failed for plan-milestone recovery: ${e instanceof Error ? e.message : String(e)}`); }
+    }
   }
 
   return diagnoseExpectedArtifact(unitType, unitId, base);