gsd-pi 2.80.0-dev.2848d9009 → 2.80.0-dev.3065f7bfa

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

@@ -11,7 +11,12 @@ import type { ExtensionAPI, ExtensionContext } from "@gsd/pi-coding-agent";
 import type { AutoSession } from "./session.js";
 import { NEW_SESSION_TIMEOUT_MS } from "./session.js";
 import type { UnitResult } from "./types.js";
-import { _clearCurrentResolve, _setCurrentResolve, _setSessionSwitchInFlight } from "./resolve.js";
+import {
+  _clearCurrentResolve,
+  _consumePendingSwitchCancellation,
+  _setCurrentResolve,
+  _setSessionSwitchInFlight,
+} from "./resolve.js";
 import {
   getCurrentTurnGeneration,
   runWithTurnGeneration,
@@ -43,44 +48,22 @@ export async function runUnit(
 ): Promise<UnitResult> {
   debugLog("runUnit", { phase: "start", unitType, unitId });
 
-  // Ensure cwd matches basePath BEFORE newSession() captures it. The new
-  // session reads process.cwd() during construction to anchor its tool
-  // runtime and system prompt; if cwd has drifted (async_bash, background
-  // jobs, prior unit cleanup), the session would otherwise be rooted to
-  // the wrong directory. Must be synchronous — no awaits between chdir
-  // and newSession (#1389, #4762 follow-up).
-  try {
-    if (process.cwd() !== s.basePath) {
-      process.chdir(s.basePath);
-    }
-  } catch (e) {
-    const msg = `Failed to chdir to basePath before newSession (basePath: ${s.basePath}): ${String(e)}`;
-    logWarning("engine", msg, { basePath: s.basePath, error: String(e) });
-    return {
-      status: "cancelled",
-      errorContext: {
-        message: msg,
-        category: "session-failed",
-        isTransient: true,
-      },
-    };
-  }
-
   // ── Session creation with timeout ──
   debugLog("runUnit", { phase: "session-create", unitType, unitId });
 
   let sessionResult: { cancelled: boolean };
   let sessionTimeoutHandle: ReturnType<typeof setTimeout> | undefined;
   const mySessionSwitchGeneration = ++sessionSwitchGeneration;
-  // #3731: Cancellation controller for newSession(). When the session-creation
-  // timeout fires, we abort this controller so that the still-in-flight
-  // newSession() discards itself after await this.abort() completes, preventing
-  // it from capturing the (now-root) process.cwd() and rebuilding the tool
-  // runtime with the wrong cwd.
+  // #3731: Cancellation controller for newSession(). When session creation
+  // times out, abort before a late session switch can rebuild the tool runtime
+  // against a stale workspace root.
   const sessionAbortController = new AbortController();
   _setSessionSwitchInFlight(true);
   try {
-    const sessionPromise = s.cmdCtx!.newSession({ abortSignal: sessionAbortController.signal }).finally(() => {
+    const sessionPromise = s.cmdCtx!.newSession({
+      abortSignal: sessionAbortController.signal,
+      workspaceRoot: s.basePath,
+    }).finally(() => {
       if (sessionSwitchGeneration === mySessionSwitchGeneration) {
         _setSessionSwitchInFlight(false);
       }
@@ -97,6 +80,7 @@ export async function runUnit(
     sessionResult = await Promise.race([sessionPromise, timeoutPromise]);
   } catch (sessionErr) {
     if (sessionTimeoutHandle) clearTimeout(sessionTimeoutHandle);
+    _consumePendingSwitchCancellation();
     const msg =
       sessionErr instanceof Error ? sessionErr.message : String(sessionErr);
     debugLog("runUnit", {
@@ -110,17 +94,20 @@ export async function runUnit(
   if (sessionTimeoutHandle) clearTimeout(sessionTimeoutHandle);
 
   if (sessionResult.cancelled) {
+    _consumePendingSwitchCancellation();
     debugLog("runUnit-session-timeout", { unitType, unitId });
     return { status: "cancelled", errorContext: { message: "Session creation timed out", category: "timeout", isTransient: true } };
   }
 
   if (!s.active) {
+    _consumePendingSwitchCancellation();
     return { status: "cancelled" };
   }
 
   if (s.currentUnitModel && typeof pi.setModel === "function") {
     const restored = await pi.setModel(s.currentUnitModel, { persist: false });
     if (!restored) {
+      _consumePendingSwitchCancellation();
       const message =
         `Failed to restore configured model ${s.currentUnitModel.provider}/${s.currentUnitModel.id} after session creation`;
       ctx.ui.notify(
@@ -145,6 +132,14 @@ export async function runUnit(
   const unitPromise = new Promise<UnitResult>((resolve) => {
     _setCurrentResolve(resolve);
   });
+  const pendingSwitchCancellation = _consumePendingSwitchCancellation();
+  if (pendingSwitchCancellation) {
+    _clearCurrentResolve();
+    return {
+      status: "cancelled",
+      ...(pendingSwitchCancellation.errorContext ? { errorContext: pendingSwitchCancellation.errorContext } : {}),
+    };
+  }
 
   // ── Provider request-readiness pre-check (#4555) ──
   // Verify the provider can accept requests before dispatching. If the token

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

@@ -178,6 +178,13 @@ export class AutoSession {
    * stale context bleeding into unrelated slices.
    */
   lastPreExecFailure: PreExecFailure | null = null;
+  /**
+   * Tracks how many consecutive times each slice unit has failed pre-execution
+   * checks. Keyed by unitId (e.g. "M001/S01"). Used to break the infinite
+   * plan-slice → pre-exec fail → re-dispatch loop when the planner cannot fix
+   * the issues after MAX_PRE_EXEC_RETRIES re-attempts.
+   */
+  readonly preExecRetryCount: Map<string, number> = new Map();
 
   // ── Tool invocation errors (#2883) ──────────────────────────────────
   /** Set when a GSD tool execution ends with isError due to malformed/truncated
@@ -343,6 +350,7 @@ export class AutoSession {
     this.rewriteAttemptCount = 0;
     this.consecutiveCompleteBootstraps = 0;
     this.lastPreExecFailure = null;
+    this.preExecRetryCount.clear();
     this.lastToolInvocationError = null;
     this.lastUnitAgentEndMessages = null;
     this.lastGitActionFailure = null;

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

@@ -31,7 +31,6 @@ import { loadEffectiveGSDPreferences } from "./preferences.js";
 import type { MinimalModelRegistry } from "./context-budget.js";
 import { pauseAuto } from "./auto.js";
 import { resolveCanonicalMilestoneRoot } from "./worktree-manager.js";
-import { logWarning } from "./workflow-logger.js";
 import {
   getWorkflowTransportSupportError,
   getRequiredWorkflowToolsForAutoUnit,
@@ -290,38 +289,13 @@ export async function dispatchDirectPhase(
 
   ctx.ui.notify(`Dispatching ${unitType} for ${unitId}...`, "info");
 
-  const originalCwd = process.cwd();
-
-  try {
-    // Ensure cwd matches dispatchBase BEFORE newSession() captures it. Synchronous —
-    // no awaits between chdir and newSession.
-    try {
-      if (process.cwd() !== dispatchBase) {
-        process.chdir(dispatchBase);
-      }
-    } catch (err) {
-      const msg = `Failed to chdir before direct-dispatch newSession (basePath: ${dispatchBase}): ${err instanceof Error ? err.message : String(err)}`;
-      logWarning("engine", msg, { file: "auto-direct-dispatch.ts", basePath: dispatchBase, error: err instanceof Error ? err.message : String(err) });
-      ctx.ui.notify(`${msg}. Cancelling dispatch to avoid running in the wrong directory.`, "error");
-      return;
-    }
-
-    const result = await ctx.newSession();
-    if (result.cancelled) {
-      ctx.ui.notify("Session creation cancelled.", "warning");
-      return;
-    }
-    pi.sendMessage(
-      { customType: "gsd-dispatch", content: prompt, display: false },
-      { triggerTurn: true },
-    );
-  } finally {
-    try {
-      if (process.cwd() !== originalCwd) {
-        process.chdir(originalCwd);
-      }
-    } catch (err) {
-      logWarning("engine", `Failed to restore cwd after direct dispatch: ${err instanceof Error ? err.message : String(err)}`, { file: "auto-direct-dispatch.ts", basePath: originalCwd });
-    }
+  const result = await ctx.newSession({ workspaceRoot: dispatchBase });
+  if (result.cancelled) {
+    ctx.ui.notify("Session creation cancelled.", "warning");
+    return;
   }
+  pi.sendMessage(
+    { customType: "gsd-dispatch", content: prompt, display: false },
+    { triggerTurn: true },
+  );
 }

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

@@ -938,6 +938,22 @@ export const DISPATCH_RULES: DispatchRule[] = [
       const unitId = `${mid}/${sid}`;
       let priorPreExecFailure: { blockingFindings: string[]; verdictExcerpt: string } | undefined;
       if (session?.lastPreExecFailure?.unitId === unitId) {
+        // Circuit breaker: stop re-dispatching after 2 failed retries. The
+        // planner has had multiple attempts with injected failure context and
+        // still cannot produce a valid plan — human review is required.
+        const MAX_PRE_EXEC_RETRIES = 2;
+        const retryCount = session.preExecRetryCount?.get(unitId) ?? 0;
+        if (retryCount >= MAX_PRE_EXEC_RETRIES) {
+          const findings = session.lastPreExecFailure.blockingFindings.join("; ");
+          session.lastPreExecFailure = null;
+          session.preExecRetryCount?.delete(unitId);
+          return {
+            action: "stop",
+            reason: `Pre-execution checks failed ${retryCount} times for ${unitId} — manual intervention required. Blocking findings: ${findings}. Fix the plan manually, then run /gsd auto to resume.`,
+            level: "error",
+            matchedRule: "planning → plan-slice",
+          };
+        }
         priorPreExecFailure = {
           blockingFindings: session.lastPreExecFailure.blockingFindings,
           verdictExcerpt: session.lastPreExecFailure.verdictExcerpt,

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

@@ -1320,8 +1320,11 @@ export async function postUnitPostVerification(pctx: PostUnitContext): Promise<"
 
         const strictMode = prefs?.enhanced_verification_strict === true;
 
-        // Run pre-execution checks
-        const result: PreExecutionResult = await runPreExecutionChecks(tasks, s.basePath);
+        // Run pre-execution checks against the canonical project root. In
+        // worktree isolation, s.basePath can point at a metadata-only worktree,
+        // while source files remain under the project root.
+        const preExecutionBasePath = s.canonicalProjectRoot;
+        const result: PreExecutionResult = await runPreExecutionChecks(tasks, preExecutionBasePath);
 
         // Log summary to stderr in existing verification output format
         const emoji = result.status === "pass" ? "✅" : result.status === "warn" ? "⚠️" : "❌";
@@ -1338,12 +1341,12 @@ export async function postUnitPostVerification(pctx: PostUnitContext): Promise<"
         }
 
         // Write evidence JSON to slice artifacts directory
-        const slicePath = resolveSlicePath(s.basePath, mid, sid);
+        const slicePath = resolveSlicePath(preExecutionBasePath, mid, sid);
         const evidenceFileName = `${sid}-PRE-EXEC-VERIFY.json`;
         let evidencePath = join(".gsd", "milestones", mid, "slices", sid, evidenceFileName);
         if (slicePath) {
           writePreExecutionEvidence(result, slicePath, mid, sid);
-          evidencePath = relative(s.basePath, join(slicePath, evidenceFileName)) || evidenceFileName;
+          evidencePath = relative(preExecutionBasePath, join(slicePath, evidenceFileName)) || evidenceFileName;
         }
 
         if (uokFlags.gates) {
@@ -1398,6 +1401,9 @@ export async function postUnitPostVerification(pctx: PostUnitContext): Promise<"
             ),
             verdictExcerpt: `status=${result.status}; ${blockingCount} blocking issue${blockingCount === 1 ? "" : "s"} detected`,
           };
+          // Track consecutive pre-exec failures per slice for loop detection.
+          const retryKey = currentUnit.id;
+          s.preExecRetryCount.set(retryKey, (s.preExecRetryCount.get(retryKey) ?? 0) + 1);
           preExecPauseNeeded = true;
         } else if (result.status === "warn") {
           ctx.ui.notify(
@@ -1414,10 +1420,18 @@ export async function postUnitPostVerification(pctx: PostUnitContext): Promise<"
               ),
               verdictExcerpt: `status=${result.status} (strict mode); ${warnChecks.length} warning${warnChecks.length === 1 ? "" : "s"} treated as blocking`,
             };
+            const retryKey = currentUnit.id;
+            s.preExecRetryCount.set(retryKey, (s.preExecRetryCount.get(retryKey) ?? 0) + 1);
             preExecPauseNeeded = true;
           }
         }
 
+        // Reset the retry counter when checks pass — a successful re-plan
+        // should not carry over a stale failure count into future slices.
+        if (result.status === "pass") {
+          s.preExecRetryCount.delete(currentUnit.id);
+        }
+
         debugLog("postUnitPostVerification", {
           phase: "pre-execution-checks",
           status: result.status,

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

@@ -35,6 +35,7 @@ import {
 import { formatDecisionsCompact, formatRequirementsCompact } from "./structured-data-formatter.js";
 import { readPhaseAnchor, formatAnchorForPrompt } from "./phase-anchor.js";
 import { composeContextModeInstructions, composeInlinedContext, type ArtifactResolver, type ContextModeRenderMode } from "./unit-context-composer.js";
+import { readCompactionSnapshot } from "./compaction-snapshot.js";
 import { logWarning } from "./workflow-logger.js";
 import { inlineGraphSubgraph } from "./graph-context.js";
 import { buildExtractionStepsBlock } from "./commands-extract-learnings.js";
@@ -44,12 +45,12 @@ import { classifyProject, type ProjectClassification } from "./detection.js";
 // ─── Preamble Cap ─────────────────────────────────────────────────────────────
 
 /**
- * Historical static ceiling for the preamble cap. Kept as an upper bound even
+ * Static ceiling for the preamble cap. Kept as an upper bound even
  * after context-window-aware sizing so large-window users don't suddenly see
- * 10× looser caps than before. Small-window users get a tighter cap derived
+ * 10× looser caps than needed. Small-window users get a tighter cap derived
  * from their configured executor window.
  */
-const MAX_PREAMBLE_CHARS = 30_000;
+const MAX_PREAMBLE_CHARS = 20_000;
 
 /**
  * Resolve prompt budgets from the configured executor context window.
@@ -184,8 +185,8 @@ function formatCloseoutReviewInstructions(validationContent: string | null, vali
 }
 
 function capPreamble(preamble: string): string {
-  // Cap inlined context at min(historical 30K ceiling, scaled inline budget).
-  // The ceiling preserves pre-fix behavior for large-window users; the scaled
+  // Cap inlined context at min(static ceiling, scaled inline budget).
+  // The ceiling bounds repeated auto prompt payloads; the scaled
   // budget tightens the cap for small-window users whose true safe limit is
   // below 30K. `computeBudgets` allocates 40% of total chars to inline context.
   const budget = Math.min(MAX_PREAMBLE_CHARS, resolvePromptBudgets().inlineContextBudgetChars);
@@ -205,13 +206,28 @@ function renderContextModeForPrompt(
   });
 }
 
+function renderContextModeBlockForPrompt(
+  unitType: string,
+  base: string,
+  renderMode: ContextModeRenderMode = "standalone",
+): string {
+  const contextMode = renderContextModeForPrompt(unitType, base, renderMode);
+  if (!contextMode) return "";
+  if (renderMode === "nested") return contextMode;
+
+  const snapshot = readCompactionSnapshot(base);
+  if (!snapshot?.trim()) return contextMode;
+
+  return `${contextMode}\n\n## Context Snapshot\nSource: \`.gsd/last-snapshot.md\`\n\n${snapshot.trimEnd()}`;
+}
+
 function prependContextModeToBlock(
   unitType: string,
   base: string,
   block: string,
   renderMode: ContextModeRenderMode = "standalone",
 ): string {
-  const contextMode = renderContextModeForPrompt(unitType, base, renderMode);
+  const contextMode = renderContextModeBlockForPrompt(unitType, base, renderMode);
   if (!contextMode) return block;
   if (!block.trim()) return contextMode;
   return `${contextMode}\n\n${block}`;
@@ -620,10 +636,83 @@ export async function buildSliceSummaryExcerpt(
     return lines.join("\n");
   } catch {
     // Defensive — any parse failure falls back to full inline.
-    return `### ${sid} Summary\nSource: \`${relPath}\`\n\n${content.trim()}`;
+    return `### ${sid} Summary\nSource: \`${relPath}\`\n\n${capMalformedSummary(content, relPath)}`;
   }
 }
 
+export async function buildTaskSummaryExcerpt(
+  absPath: string | null, relPath: string, tid: string, options?: { blocker?: boolean },
+): Promise<string> {
+  const label = options?.blocker ? "Blocker Task Summary" : "Task Summary";
+  const header = `### ${label}: ${tid} (excerpt)\nSource: \`${relPath}\``;
+  const content = absPath ? await loadFile(absPath) : null;
+  if (!content) {
+    return `${header}\n\n_(not found — file does not exist yet)_`;
+  }
+
+  try {
+    const s = parseSummary(content);
+    if (!s.frontmatter.id) {
+      return `### ${label}: ${tid}\nSource: \`${relPath}\`\n\n${capMalformedSummary(content, relPath)}`;
+    }
+
+    const lines: string[] = [header, ""];
+    if (s.title) lines.push(`**Title:** ${s.title}`);
+    if (s.oneLiner) lines.push(`**One-liner:** ${s.oneLiner}`);
+    if (s.frontmatter.verification_result) {
+      lines.push(`**Verification:** \`${s.frontmatter.verification_result}\``);
+    }
+    lines.push(`**Blocker discovered:** ${s.frontmatter.blocker_discovered ? "yes — read full summary if blocker details are insufficient" : "no"}`);
+    if (s.frontmatter.provides.length > 0) lines.push(`**Provides:** ${s.frontmatter.provides.slice(0, 4).join("; ")}`);
+    if (s.frontmatter.key_decisions.length > 0) lines.push(`**Key decisions:** ${s.frontmatter.key_decisions.slice(0, 4).join("; ")}`);
+    if (s.frontmatter.patterns_established.length > 0) lines.push(`**Patterns established:** ${s.frontmatter.patterns_established.slice(0, 4).join("; ")}`);
+    if (s.frontmatter.key_files.length > 0) {
+      const files = s.frontmatter.key_files.slice(0, 6);
+      const more = s.frontmatter.key_files.length > files.length ? ` (+${s.frontmatter.key_files.length - files.length} more)` : "";
+      lines.push(`**Key files:** ${files.join(", ")}${more}`);
+    }
+
+    const SECTION_CAP_CHARS = 500;
+    const capSection = (body: string): string => {
+      const trimmed = body.trim();
+      if (trimmed.length <= SECTION_CAP_CHARS) return trimmed;
+      return `${trimmed.slice(0, SECTION_CAP_CHARS)}\n… (truncated — see full \`${relPath}\`)`;
+    };
+
+    const verification = extractMarkdownSection(content, "Verification");
+    const diagnostics = extractMarkdownSection(content, "Diagnostics");
+    const knownIssues = extractMarkdownSection(content, "Known Issues");
+
+    if (verification && verification.trim()) {
+      lines.push("", "#### Verification", capSection(verification));
+    }
+    if (diagnostics && diagnostics.trim()) {
+      lines.push("", "#### Diagnostics", capSection(diagnostics));
+    }
+    if (s.deviations && s.deviations.trim()) {
+      lines.push("", "#### Deviations", capSection(s.deviations));
+    }
+    if (knownIssues && knownIssues.trim()) {
+      lines.push("", "#### Known issues", capSection(knownIssues));
+    }
+
+    lines.push(
+      "",
+      `> **On-demand:** read \`${relPath}\` only when this excerpt is absent/truncated or you need fuller blocker, implementation, or file-change evidence.`,
+    );
+    return lines.join("\n");
+  } catch {
+    return `### ${label}: ${tid}\nSource: \`${relPath}\`\n\n${capMalformedSummary(content, relPath)}`;
+  }
+}
+
+function capMalformedSummary(content: string, relPath: string): string {
+  const trimmed = content.trim();
+  const limit = 1_500;
+  if (trimmed.length <= limit) return trimmed;
+  return `${trimmed.slice(0, limit).trimEnd()}\n\n[Truncated malformed summary — read \`${relPath}\` for full details.]`;
+}
+
 /**
  * Load and inline dependency slice summaries (full content, not just paths).
  */
@@ -898,7 +987,7 @@ export async function inlineKnowledgeScoped(
  * plan-milestone, complete-slice, complete-milestone, validate-milestone,
  * reassess-roadmap) previously injected the full KNOWLEDGE.md (~226KB for a
  * real project) on every invocation. This helper scopes by caller-supplied
- * keywords and caps the payload at `maxChars` (default 30,000 chars).
+ * keywords and caps the payload at `maxChars` (default 12,000 chars).
  *
  * Returns null when no KNOWLEDGE.md exists or no entries match any keyword.
  */
@@ -907,7 +996,7 @@ export async function inlineKnowledgeBudgeted(
   keywords: string[],
   options?: { maxChars?: number },
 ): Promise<string | null> {
-  const DEFAULT_MAX_CHARS = 30_000;
+  const DEFAULT_MAX_CHARS = 12_000;
   const HARD_MAX_CHARS = 100_000;
   const raw = Number(options?.maxChars ?? DEFAULT_MAX_CHARS);
   const maxChars = Number.isFinite(raw)
@@ -2069,8 +2158,17 @@ export async function buildPlanSlicePrompt(
       `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.`,
+      `**How to fix each type of issue:**\n` +
+      `- **"[file] X doesn't exist and isn't created by prior or same-task outputs"**: ` +
+      `Either (a) add an earlier task that creates X on disk before the task that needs it, ` +
+      `or (b) if this task IS the one that creates X, move X from inputs to expected_output. ` +
+      `Do NOT put X in a task's expected_output if that task only reads or verifies X — only tasks that actually write X to disk should list it in expected_output.\n` +
+      `- **"[file] X: Task T_early reads X but it's created by task T_late (sequence violation)"**: ` +
+      `Either (a) reorder tasks so T_late (the creator) runs before T_early (the reader), ` +
+      `or (b) if T_late doesn't actually create X (it only reads/tests it), remove X from T_late's expected_output entirely.\n` +
+      `- **"[package] P not found on npm"**: Either remove the npm install for P, or use the correct package name.\n\n` +
+      `Every file listed in a task's inputs must either exist on disk already or appear in an earlier task's expected_output. ` +
+      `A task's expected_output must only list files it actually writes to disk.`,
     );
   }
   return renderSlicePrompt({
@@ -2353,10 +2451,9 @@ export async function buildCompleteSlicePrompt(
         const blocks: string[] = [];
         for (const file of summaryFiles) {
           const absPath = join(tDir, file);
-          const content = await loadFile(absPath);
-          if (!content) continue;
           const relPath = `${sRel}/tasks/${file}`;
-          blocks.push(`### Task Summary: ${file.replace(/-SUMMARY\.md$/i, "")}\nSource: \`${relPath}\`\n\n${content.trim()}`);
+          const taskId = file.replace(/-SUMMARY\.md$/i, "");
+          blocks.push(await buildTaskSummaryExcerpt(absPath, relPath, taskId));
         }
         return blocks.length > 0 ? blocks.join("\n\n---\n\n") : null;
       }
@@ -2733,7 +2830,7 @@ export async function buildReplanSlicePrompt(
       const relPath = `${sRel}/tasks/${file}`;
       if (summary.frontmatter.blocker_discovered) {
         blockerTaskId = summary.frontmatter.id || file.replace(/-SUMMARY\.md$/i, "");
-        inlined.push(`### Blocker Task Summary: ${blockerTaskId}\nSource: \`${relPath}\`\n\n${content.trim()}`);
+        inlined.push(await buildTaskSummaryExcerpt(absPath, relPath, blockerTaskId, { blocker: true }));
       }
     }
   }

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

@@ -67,6 +67,60 @@ export {
 
 // ─── Artifact Resolution & Verification ───────────────────────────────────────
 
+export type ArtifactRecoveryDbRefreshResult =
+  | { ok: true }
+  | { ok: false; fatal: boolean; message: string; reason: string };
+
+export function refreshRecoveryDbForArtifact(
+  unitType: string,
+  unitId: string,
+): ArtifactRecoveryDbRefreshResult {
+  if (unitType !== "plan-slice" && unitType !== "execute-task") return { ok: true };
+  if (!isDbAvailable()) return { ok: true };
+
+  if (!refreshOpenDatabaseFromDisk()) {
+    return {
+      ok: false,
+      fatal: unitType === "execute-task",
+      reason: `${unitType}-db-refresh-failed`,
+      message: `Stuck recovery found ${unitType} ${unitId} artifacts, but the DB refresh failed.`,
+    };
+  }
+
+  if (unitType !== "execute-task") return { ok: true };
+
+  const { milestone: mid, slice: sid, task: tid } = parseUnitId(unitId);
+  if (!mid || !sid || !tid) {
+    return {
+      ok: false,
+      fatal: true,
+      reason: "execute-task-invalid-unit-id",
+      message: `Stuck recovery found execute-task ${unitId} artifacts, but the unit id could not be parsed for DB verification.`,
+    };
+  }
+
+  const task = getTask(mid, sid, tid);
+  if (!task) {
+    return {
+      ok: false,
+      fatal: true,
+      reason: "execute-task-artifact-db-missing",
+      message: `Stuck recovery found execute-task ${unitId} artifacts, but no matching DB task row exists after refresh.`,
+    };
+  }
+
+  if (!isClosedStatus(task.status)) {
+    return {
+      ok: false,
+      fatal: true,
+      reason: "execute-task-artifact-db-mismatch",
+      message: `Stuck recovery found execute-task ${unitId} artifacts, but the DB task status is still '${task.status}' after refresh.`,
+    };
+  }
+
+  return { ok: true };
+}
+
 function hasCapturedWorkflowPrefs(base: string): boolean {
   const prefsPath = resolveExpectedArtifactPath("workflow-preferences", "WORKFLOW-PREFS", base);
   if (!prefsPath || !existsSync(prefsPath)) return false;