gsd-pi 2.52.0 → 2.53.0-dev.a67436f

package/src/resources/extensions/gsd/tests/parallel-orchestrator-zombie-cleanup.test.ts

@@ -0,0 +1,277 @@
+/**
+ * Regression tests for zombie worker cleanup (#2736).
+ *
+ * Verifies that:
+ * 1. refreshWorkerStatuses() deactivates the orchestrator when all workers
+ *    are in terminal states (error/stopped).
+ * 2. restoreRuntimeState() (via getWorkerStatuses) returns empty when the
+ *    cached state has only dead workers.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { mkdirSync, writeFileSync, rmSync, existsSync } from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+import { randomUUID } from "node:crypto";
+
+import {
+  persistState,
+  resetOrchestrator,
+  refreshWorkerStatuses,
+  isParallelActive,
+  getOrchestratorState,
+  getWorkerStatuses,
+  type PersistedState,
+} from "../parallel-orchestrator.ts";
+
+function makeTmpBase(): string {
+  const base = join(tmpdir(), `gsd-test-zombie-${randomUUID()}`);
+  mkdirSync(join(base, ".gsd", "parallel"), { recursive: true });
+  return base;
+}
+
+function cleanup(base: string): void {
+  try {
+    rmSync(base, { recursive: true, force: true });
+  } catch { /* non-fatal */ }
+}
+
+/** Write a fake orchestrator.json to simulate persisted state. */
+function writePersistedState(basePath: string, data: PersistedState): void {
+  const dest = join(basePath, ".gsd", "orchestrator.json");
+  writeFileSync(dest, JSON.stringify(data, null, 2), "utf-8");
+}
+
+/** Write a fake session status file to .gsd/parallel/<milestoneId>.status.json */
+function writeSessionStatusFile(
+  basePath: string,
+  milestoneId: string,
+  state: "running" | "paused" | "stopped" | "error",
+  pid: number,
+): void {
+  const dest = join(basePath, ".gsd", "parallel", `${milestoneId}.status.json`);
+  writeFileSync(
+    dest,
+    JSON.stringify({
+      milestoneId,
+      pid,
+      state,
+      currentUnit: null,
+      completedUnits: 0,
+      cost: 0.5,
+      lastHeartbeat: Date.now(),
+      startedAt: Date.now() - 60_000,
+      worktreePath: join(basePath, "worktrees", milestoneId),
+    }),
+    "utf-8",
+  );
+}
+
+// Use a PID that is guaranteed dead — PID 1 is init/launchd and won't be
+// killable by this process, but 2147483647 is unlikely to exist.
+const DEAD_PID = 2147483647;
+
+// ─── refreshWorkerStatuses: deactivates when all workers dead ──────────
+
+test("#2736: refreshWorkerStatuses deactivates orchestrator when all workers are error/stopped", (t) => {
+  const base = makeTmpBase();
+  t.after(() => {
+    resetOrchestrator();
+    cleanup(base);
+  });
+
+  // Seed persisted state with two workers using current PID (alive) so
+  // restoreState() accepts them, then immediately mark them as error via
+  // session status files so refreshWorkerStatuses sees terminal states.
+  const persisted: PersistedState = {
+    active: true,
+    workers: [
+      {
+        milestoneId: "M001",
+        title: "Milestone 1",
+        pid: process.pid, // alive PID so restoreState accepts it
+        worktreePath: join(base, "worktrees", "M001"),
+        startedAt: Date.now() - 60_000,
+        state: "running",
+        cost: 1.0,
+      },
+      {
+        milestoneId: "M002",
+        title: "Milestone 2",
+        pid: process.pid,
+        worktreePath: join(base, "worktrees", "M002"),
+        startedAt: Date.now() - 60_000,
+        state: "running",
+        cost: 0.5,
+      },
+    ],
+    totalCost: 1.5,
+    startedAt: Date.now() - 60_000,
+    configSnapshot: { max_workers: 3 },
+  };
+  writePersistedState(base, persisted);
+
+  // First, restore the state into memory via getWorkerStatuses (triggers restoreIfNeeded)
+  const workers = getWorkerStatuses(base);
+  assert.equal(workers.length, 2, "should have 2 workers after restore");
+  assert.ok(isParallelActive(), "orchestrator should be active after restore");
+
+  // Now write session status files marking both workers as error
+  writeSessionStatusFile(base, "M001", "error", process.pid);
+  writeSessionStatusFile(base, "M002", "error", process.pid);
+
+  // Refresh — should detect all-dead and deactivate
+  refreshWorkerStatuses(base);
+
+  assert.equal(isParallelActive(), false, "orchestrator should be inactive after all workers died");
+  assert.equal(getOrchestratorState(), null, "state should be null after cleanup");
+});
+
+test("#2736: refreshWorkerStatuses keeps orchestrator active when some workers are still running", (t) => {
+  const base = makeTmpBase();
+  t.after(() => {
+    resetOrchestrator();
+    cleanup(base);
+  });
+
+  const persisted: PersistedState = {
+    active: true,
+    workers: [
+      {
+        milestoneId: "M001",
+        title: "Milestone 1",
+        pid: process.pid,
+        worktreePath: join(base, "worktrees", "M001"),
+        startedAt: Date.now() - 60_000,
+        state: "running",
+        cost: 1.0,
+      },
+      {
+        milestoneId: "M002",
+        title: "Milestone 2",
+        pid: process.pid,
+        worktreePath: join(base, "worktrees", "M002"),
+        startedAt: Date.now() - 60_000,
+        state: "running",
+        cost: 0.5,
+      },
+    ],
+    totalCost: 1.5,
+    startedAt: Date.now() - 60_000,
+    configSnapshot: { max_workers: 3 },
+  };
+  writePersistedState(base, persisted);
+
+  // Restore state
+  getWorkerStatuses(base);
+
+  // Mark M001 as error but keep M002 running
+  writeSessionStatusFile(base, "M001", "error", process.pid);
+  writeSessionStatusFile(base, "M002", "running", process.pid);
+
+  refreshWorkerStatuses(base);
+
+  assert.ok(isParallelActive(), "orchestrator should remain active with a running worker");
+  assert.ok(getOrchestratorState() !== null, "state should still exist");
+});
+
+// ─── restoreRuntimeState: returns false when cached state has only dead workers ─
+
+test("#2736: getWorkerStatuses returns empty when all cached workers are in error state", (t) => {
+  const base = makeTmpBase();
+  t.after(() => {
+    resetOrchestrator();
+    cleanup(base);
+  });
+
+  // First, set up active state with live workers
+  const persisted: PersistedState = {
+    active: true,
+    workers: [
+      {
+        milestoneId: "M001",
+        title: "Milestone 1",
+        pid: process.pid,
+        worktreePath: join(base, "worktrees", "M001"),
+        startedAt: Date.now() - 60_000,
+        state: "running",
+        cost: 0.5,
+      },
+    ],
+    totalCost: 0.5,
+    startedAt: Date.now() - 60_000,
+    configSnapshot: { max_workers: 3 },
+  };
+  writePersistedState(base, persisted);
+
+  // Restore into memory
+  getWorkerStatuses(base);
+  assert.ok(isParallelActive(), "should be active initially");
+
+  // Simulate all workers dying: write error status then refresh to update
+  writeSessionStatusFile(base, "M001", "error", process.pid);
+  refreshWorkerStatuses(base);
+
+  // State should now be cleared
+  assert.equal(getOrchestratorState(), null, "state should be null after all workers error");
+
+  // Reset and try again — getWorkerStatuses with restoreIfNeeded should
+  // find no live workers on disk (orchestrator.json was cleaned up)
+  const workers = getWorkerStatuses(base);
+  assert.equal(workers.length, 0, "should return empty when no live workers exist");
+});
+
+test("#2736: restoreRuntimeState clears stale state when all workers are stopped", (t) => {
+  const base = makeTmpBase();
+  t.after(() => {
+    resetOrchestrator();
+    cleanup(base);
+  });
+
+  // Set up and restore state
+  const persisted: PersistedState = {
+    active: true,
+    workers: [
+      {
+        milestoneId: "M001",
+        title: "Milestone 1",
+        pid: process.pid,
+        worktreePath: join(base, "worktrees", "M001"),
+        startedAt: Date.now() - 60_000,
+        state: "running",
+        cost: 0.3,
+      },
+      {
+        milestoneId: "M002",
+        title: "Milestone 2",
+        pid: process.pid,
+        worktreePath: join(base, "worktrees", "M002"),
+        startedAt: Date.now() - 60_000,
+        state: "running",
+        cost: 0.7,
+      },
+    ],
+    totalCost: 1.0,
+    startedAt: Date.now() - 60_000,
+    configSnapshot: { max_workers: 3 },
+  };
+  writePersistedState(base, persisted);
+
+  // Restore into memory
+  getWorkerStatuses(base);
+  assert.ok(isParallelActive(), "should be active initially");
+
+  // Mark all as stopped via session status, then refresh
+  writeSessionStatusFile(base, "M001", "stopped", process.pid);
+  writeSessionStatusFile(base, "M002", "stopped", process.pid);
+  refreshWorkerStatuses(base);
+
+  // Orchestrator should be deactivated and state cleaned
+  assert.equal(isParallelActive(), false, "should be inactive after all workers stopped");
+  assert.equal(getOrchestratorState(), null, "state should be null");
+
+  // Verify the state file was removed
+  const stateFile = join(base, ".gsd", "orchestrator.json");
+  assert.equal(existsSync(stateFile), false, "orchestrator.json should be removed");
+});

package/src/resources/extensions/gsd/tests/phases-merge-error-stops-auto.test.ts

@@ -0,0 +1,103 @@
+/**
+ * phases-merge-error-stops-auto.test.ts — Regression test for #2766.
+ *
+ * When mergeAndExit throws a non-MergeConflictError, the auto loop must
+ * stop instead of continuing with unmerged work. This test verifies that
+ * all catch blocks in auto/phases.ts that handle mergeAndExit errors
+ * call stopAuto and return { action: "break" } for non-conflict errors.
+ */
+
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
+import { createTestContext } from "./test-helpers.ts";
+
+const { assertTrue, report } = createTestContext();
+
+const phasesPath = join(import.meta.dirname, "..", "auto", "phases.ts");
+const phasesSrc = readFileSync(phasesPath, "utf-8");
+
+console.log("\n=== #2766: Non-MergeConflictError stops auto mode ===");
+
+// ── Test 1: phases.ts calls logError for non-conflict merge errors ──────
+
+assertTrue(
+  phasesPath.length > 0 && phasesPath.endsWith("phases.ts"),
+  "phases.ts file exists and is readable",
+);
+
+// Count all mergeAndExit catch blocks by finding "} catch (mergeErr)" patterns
+const mergeErrCatches = [...phasesPath.matchAll(/\} catch \(mergeErr\)/g)];
+// Use the source itself for matching
+const mergeErrCatchCount = [...phasesSrc.matchAll(/\} catch \(mergeErr\)/g)].length;
+assertTrue(
+  mergeErrCatchCount >= 3,
+  `all mergeAndExit call sites have catch (mergeErr) blocks (found ${mergeErrCatchCount}, expected >= 3)`,
+);
+
+// ── Test 2: Every mergeErr catch block handles non-MergeConflictError ───
+
+// Find each catch block and verify it has the non-conflict error handling pattern
+const catchPattern = /\} catch \(mergeErr\) \{/g;
+let match;
+let blocksWithNonConflictHandling = 0;
+let blocksTotal = 0;
+
+while ((match = catchPattern.exec(phasesSrc)) !== null) {
+  blocksTotal++;
+  // Look at the ~800 chars after the catch to find both the MergeConflictError
+  // instanceof check AND the non-conflict handling
+  const afterCatch = phasesSrc.slice(match.index, match.index + 1200);
+
+  const hasInstanceofCheck = afterCatch.includes("instanceof MergeConflictError");
+  const hasNonConflictStop = afterCatch.includes('reason: "merge-failed"');
+  const hasStopAuto = afterCatch.includes("stopAuto");
+  const hasLogError = afterCatch.includes("logError");
+
+  if (hasInstanceofCheck && hasNonConflictStop && hasStopAuto && hasLogError) {
+    blocksWithNonConflictHandling++;
+  }
+}
+
+assertTrue(
+  blocksWithNonConflictHandling === blocksTotal && blocksTotal >= 3,
+  `all ${blocksTotal} mergeAndExit catch blocks stop auto on non-conflict errors (${blocksWithNonConflictHandling}/${blocksTotal})`,
+);
+
+// ── Test 3: Non-conflict handler returns break (does not continue) ──────
+
+// Verify the pattern: after the MergeConflictError instanceof block,
+// the non-conflict path returns { action: "break", reason: "merge-failed" }
+const mergeFailedReasons = [...phasesSrc.matchAll(/reason: "merge-failed"/g)].length;
+assertTrue(
+  mergeFailedReasons >= 3,
+  `all catch blocks return reason: "merge-failed" (found ${mergeFailedReasons}, expected >= 3)`,
+);
+
+// ── Test 4: Non-conflict handler notifies user ──────────────────────────
+
+// Each non-conflict block should call ctx.ui.notify with error severity
+const notifyErrorPattern = /Merge failed:.*Resolve and run \/gsd auto to resume/g;
+const notifyCount = [...phasesSrc.matchAll(notifyErrorPattern)].length;
+assertTrue(
+  notifyCount >= 3,
+  `all catch blocks notify user about merge failure (found ${notifyCount}, expected >= 3)`,
+);
+
+// ── Test 5: logError replaces logWarning for non-conflict merge errors ──
+
+// The old code used logWarning — verify logError is used instead
+const logWarningMergePattern = /logWarning\(.*Milestone merge failed with non-conflict error/g;
+const logWarningCount = [...phasesSrc.matchAll(logWarningMergePattern)].length;
+assertTrue(
+  logWarningCount === 0,
+  "logWarning is no longer used for non-conflict merge errors (replaced by logError)",
+);
+
+const logErrorMergePattern = /logError\(.*Milestone merge failed with non-conflict error/g;
+const logErrorCount = [...phasesSrc.matchAll(logErrorMergePattern)].length;
+assertTrue(
+  logErrorCount >= 3,
+  `logError is used for non-conflict merge errors (found ${logErrorCount}, expected >= 3)`,
+);
+
+report();

package/src/resources/extensions/gsd/tests/preferences.test.ts

@@ -59,7 +59,7 @@ test("solo mode applies correct defaults", () => {
   const result = applyModeDefaults("solo", { mode: "solo" });
   assert.equal(result.git?.auto_push, true);
   assert.equal(result.git?.push_branches, false);
-  assert.equal(result.git?.pre_merge_check, false);
+  assert.equal(result.git?.pre_merge_check, "auto");
   assert.equal(result.git?.merge_strategy, "squash");
   assert.equal(result.git?.isolation, "none");
   assert.equal(result.unique_milestone_ids, false);

package/src/resources/extensions/gsd/tests/rate-limit-model-fallback.test.ts

@@ -0,0 +1,90 @@
+/**
+ * rate-limit-model-fallback.test.ts — Regression test for #2770.
+ *
+ * Rate-limit errors enter the model fallback path before falling through
+ * to pause. This verifies the structural contract in agent-end-recovery.ts.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import { join, dirname } from "node:path";
+import { fileURLToPath } from "node:url";
+
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const RECOVERY_PATH = join(__dirname, "..", "bootstrap", "agent-end-recovery.ts");
+
+function getRecoverySource(): string {
+  return readFileSync(RECOVERY_PATH, "utf-8");
+}
+
+// ── Rate-limit errors attempt model fallback (#2770) ─────────────────────────
+
+test("rate-limit errors enter the model fallback branch alongside other transient errors", () => {
+  const src = getRecoverySource();
+
+  // The condition that gates model fallback must include rate-limit.
+  // Match the if-condition that contains both "rate-limit" and fallback-related kinds.
+  const fallbackConditionRe = /if\s*\([^)]*cls\.kind\s*===\s*"rate-limit"[^)]*cls\.kind\s*===\s*"network"/;
+  const fallbackConditionReAlt = /if\s*\([^)]*cls\.kind\s*===\s*"network"[^)]*cls\.kind\s*===\s*"rate-limit"/;
+
+  assert.ok(
+    fallbackConditionRe.test(src) || fallbackConditionReAlt.test(src),
+    'rate-limit must appear in the same if-condition as network/server for model fallback (#2770)',
+  );
+});
+
+test("rate-limit errors are NOT short-circuited to pause before model fallback", () => {
+  const src = getRecoverySource();
+
+  // The old code had a dedicated rate-limit early-return block before the fallback block.
+  // Verify it no longer exists.
+  const earlyRateLimitPause = /if\s*\(\s*cls\.kind\s*===\s*"rate-limit"\s*\)\s*\{[^}]*pauseTransientWithBackoff/;
+  assert.ok(
+    !earlyRateLimitPause.test(src),
+    'rate-limit must NOT have a dedicated early pause before the model fallback path (#2770)',
+  );
+});
+
+test("rate-limit errors fall through to pause if no fallback model is available", () => {
+  const src = getRecoverySource();
+
+  // After the fallback block, the transient fallback pause must still fire for rate-limit.
+  // The isTransient check covers rate-limit (verified by error-classifier tests).
+  // Verify pauseTransientWithBackoff is called with isRateLimit derived from cls.kind.
+  assert.ok(
+    src.includes('cls.kind === "rate-limit"'),
+    'agent-end-recovery.ts must reference cls.kind === "rate-limit" for fallback and pause paths (#2770)',
+  );
+
+  // The transient fallback pause must pass the isRateLimit flag correctly.
+  const pauseCallRe = /pauseTransientWithBackoff\([^)]*cls\.kind\s*===\s*"rate-limit"/;
+  assert.ok(
+    pauseCallRe.test(src),
+    'pauseTransientWithBackoff must receive isRateLimit based on cls.kind === "rate-limit" (#2770)',
+  );
+});
+
+test("other transient errors (server, connection, stream) still attempt model fallback", () => {
+  const src = getRecoverySource();
+
+  // All transient kinds must appear in the fallback condition.
+  for (const kind of ["server", "connection", "stream"]) {
+    assert.ok(
+      src.includes(`cls.kind === "${kind}"`),
+      `model fallback condition must include cls.kind === "${kind}"`,
+    );
+  }
+});
+
+test("permanent errors still bypass model fallback and pause indefinitely", () => {
+  const src = getRecoverySource();
+
+  // The permanent/unknown error handler must exist and call pauseAutoForProviderError
+  // with isTransient: false.
+  const permanentPauseRe = /pauseAutoForProviderError[\s\S]{0,300}isTransient:\s*false/;
+  assert.ok(
+    permanentPauseRe.test(src),
+    'permanent errors must pause with isTransient: false (no auto-resume)',
+  );
+});

package/src/resources/extensions/gsd/tests/session-lock-transient-read.test.ts

@@ -95,13 +95,13 @@ async function main(): Promise<void> {
       writeFileSync(lockFile, JSON.stringify(lockData, null, 2));
 
       // Simulate transient unavailability: move file away, spawn a child process
-      // to restore it after 100ms. The child runs outside our event loop so it
-      // fires even during busy-wait retries.
+      // to restore it shortly after. The child runs outside our event loop so it
+      // fires even during busy-wait retries. Give the test extra retry budget so
+      // it stays stable under full-suite CPU contention.
       renameSync(lockFile, tmpFile);
-      spawn('bash', ['-c', `sleep 0.1 && mv "${tmpFile}" "${lockFile}"`], { stdio: 'ignore', detached: true }).unref();
+      spawn('bash', ['-c', `sleep 0.05 && mv "${tmpFile}" "${lockFile}"`], { stdio: 'ignore', detached: true }).unref();
 
-      // With retries (3 attempts, 200ms delay), it should recover on 2nd or 3rd attempt
-      const result = readExistingLockDataWithRetry(lockFile, { maxAttempts: 3, delayMs: 200 });
+      const result = readExistingLockDataWithRetry(lockFile, { maxAttempts: 8, delayMs: 400 });
       assertTrue(result !== null, 'data recovered after transient unavailability');
       if (result) {
         assertEq(result.pid, process.pid, 'correct PID after recovery');
@@ -131,11 +131,12 @@ async function main(): Promise<void> {
       writeFileSync(lockFile, JSON.stringify(lockData, null, 2));
 
       // Remove read permission to simulate NFS/CIFS latency, then spawn a child
-      // to restore permissions after 100ms (runs outside our event loop).
+      // to restore permissions shortly after (runs outside our event loop).
+      // Use the same wider retry window as the rename case for full-suite stability.
       chmodSync(lockFile, 0o000);
-      spawn('bash', ['-c', `sleep 0.1 && chmod 644 "${lockFile}"`], { stdio: 'ignore', detached: true }).unref();
+      spawn('bash', ['-c', `sleep 0.05 && chmod 644 "${lockFile}"`], { stdio: 'ignore', detached: true }).unref();
 
-      const result = readExistingLockDataWithRetry(lockFile, { maxAttempts: 3, delayMs: 200 });
+      const result = readExistingLockDataWithRetry(lockFile, { maxAttempts: 8, delayMs: 400 });
       assertTrue(result !== null, 'data recovered after transient permission error');
       if (result) {
         assertEq(result.pid, process.pid, 'correct PID after permission recovery');

package/src/resources/extensions/gsd/tests/stash-pop-gsd-conflict.test.ts

@@ -0,0 +1,125 @@
+/**
+ * stash-pop-gsd-conflict.test.ts — Regression test for #2766.
+ *
+ * When a squash merge stash-pops and hits conflicts on .gsd/ state files,
+ * the UU entries block every subsequent merge. This test verifies that
+ * mergeMilestoneToMain auto-resolves .gsd/ conflicts by accepting HEAD
+ * and drops the stash, leaving the repo in a clean state.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync, mkdirSync, writeFileSync, rmSync, existsSync, readFileSync, realpathSync } from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+import { execSync } from "node:child_process";
+
+import { createAutoWorktree, mergeMilestoneToMain } from "../auto-worktree.ts";
+
+function run(cmd: string, cwd: string): string {
+  return execSync(cmd, { cwd, stdio: ["ignore", "pipe", "pipe"], encoding: "utf-8" }).trim();
+}
+
+function createTempRepo(): string {
+  const dir = realpathSync(mkdtempSync(join(tmpdir(), "wt-stashpop-test-")));
+  run("git init", dir);
+  run("git config user.email test@test.com", dir);
+  run("git config user.name Test", dir);
+  writeFileSync(join(dir, "README.md"), "# test\n");
+  mkdirSync(join(dir, ".gsd"), { recursive: true });
+  writeFileSync(join(dir, ".gsd", "STATE.md"), "version: 1\n");
+  run("git add .", dir);
+  run("git commit -m init", dir);
+  run("git branch -M main", dir);
+  return dir;
+}
+
+function makeRoadmap(milestoneId: string, title: string, slices: Array<{ id: string; title: string }>): string {
+  const sliceLines = slices.map(s => `- [x] **${s.id}: ${s.title}**`).join("\n");
+  return `# ${milestoneId}: ${title}\n\n## Slices\n${sliceLines}\n`;
+}
+
+test("#2766: stash pop conflict on .gsd/ files is auto-resolved", () => {
+  const repo = createTempRepo();
+  try {
+    const wtPath = createAutoWorktree(repo, "M300");
+
+    // Add a slice with real code on the milestone branch
+    const normalizedPath = wtPath.replaceAll("\\", "/");
+    const worktreeName = normalizedPath.split("/").pop() || "M300";
+    const sliceBranch = `slice/${worktreeName}/S01`;
+    run(`git checkout -b "${sliceBranch}"`, wtPath);
+    writeFileSync(join(wtPath, "feature.ts"), "export const feature = true;\n");
+
+    // Modify .gsd/STATE.md on the milestone branch (diverges from main)
+    writeFileSync(join(wtPath, ".gsd", "STATE.md"), "version: 2-milestone\n");
+    run("git add .", wtPath);
+    run('git commit -m "add feature and update state"', wtPath);
+    run("git checkout milestone/M300", wtPath);
+    run(`git merge --no-ff "${sliceBranch}" -m "merge S01: feature"`, wtPath);
+
+    // Dirty .gsd/STATE.md in the main repo (stash will conflict on pop)
+    writeFileSync(join(repo, ".gsd", "STATE.md"), "version: 2-main-dirty\n");
+
+    const roadmap = makeRoadmap("M300", "Stash pop conflict test", [
+      { id: "S01", title: "Feature" },
+    ]);
+
+    // mergeMilestoneToMain should succeed — .gsd/ conflict auto-resolved
+    const result = mergeMilestoneToMain(repo, "M300", roadmap);
+    assert.ok(
+      result.commitMessage.includes("GSD-Milestone: M300"),
+      "merge succeeds despite stash pop conflict on .gsd/ file",
+    );
+    assert.ok(existsSync(join(repo, "feature.ts")), "milestone code merged to main");
+
+    // Verify repo is clean (no UU entries blocking future merges)
+    const status = run("git status --porcelain", repo);
+    assert.ok(
+      !status.includes("UU "),
+      "no unmerged (UU) entries remain after stash pop conflict resolution",
+    );
+
+    // Stash should be dropped (no remaining stash entries)
+    let stashList = "";
+    try { stashList = run("git stash list", repo); } catch { /* empty stash */ }
+    assert.strictEqual(stashList, "", "stash is empty after .gsd/ conflict auto-resolution");
+  } finally {
+    try { rmSync(repo, { recursive: true, force: true, maxRetries: 3, retryDelay: 100 }); } catch { /* cleanup best-effort */ }
+  }
+});
+
+test("#2766: stash pop conflict on non-.gsd files preserves stash for manual resolution", () => {
+  const repo = createTempRepo();
+  try {
+    const wtPath = createAutoWorktree(repo, "M301");
+
+    // Add a slice that modifies a file also dirty on main
+    const normalizedPath = wtPath.replaceAll("\\", "/");
+    const worktreeName = normalizedPath.split("/").pop() || "M301";
+    const sliceBranch = `slice/${worktreeName}/S01`;
+    run(`git checkout -b "${sliceBranch}"`, wtPath);
+    writeFileSync(join(wtPath, "README.md"), "# milestone version\n");
+    run("git add .", wtPath);
+    run('git commit -m "update readme"', wtPath);
+    run("git checkout milestone/M301", wtPath);
+    run(`git merge --no-ff "${sliceBranch}" -m "merge S01: readme"`, wtPath);
+
+    // Dirty README.md in the main repo — this will conflict on stash pop
+    // and is NOT a .gsd/ file, so it should be left for manual resolution
+    writeFileSync(join(repo, "README.md"), "# locally modified\n");
+
+    const roadmap = makeRoadmap("M301", "Non-gsd stash conflict", [
+      { id: "S01", title: "Readme update" },
+    ]);
+
+    // The merge itself should still succeed (stash pop conflict is non-fatal)
+    const result = mergeMilestoneToMain(repo, "M301", roadmap);
+    assert.ok(
+      result.commitMessage.includes("GSD-Milestone: M301"),
+      "merge succeeds even with non-.gsd stash pop conflict",
+    );
+  } finally {
+    try { rmSync(repo, { recursive: true, force: true, maxRetries: 3, retryDelay: 100 }); } catch { /* cleanup best-effort */ }
+  }
+});

package/src/resources/extensions/gsd/tests/status-guards.test.ts

@@ -0,0 +1,30 @@
+// GSD — status-guards unit tests
+
+import test from 'node:test';
+import assert from 'node:assert/strict';
+
+import { isClosedStatus } from '../status-guards.ts';
+
+test('isClosedStatus: "complete" returns true', () => {
+  assert.equal(isClosedStatus('complete'), true);
+});
+
+test('isClosedStatus: "done" returns true', () => {
+  assert.equal(isClosedStatus('done'), true);
+});
+
+test('isClosedStatus: "pending" returns false', () => {
+  assert.equal(isClosedStatus('pending'), false);
+});
+
+test('isClosedStatus: "in_progress" returns false', () => {
+  assert.equal(isClosedStatus('in_progress'), false);
+});
+
+test('isClosedStatus: "active" returns false', () => {
+  assert.equal(isClosedStatus('active'), false);
+});
+
+test('isClosedStatus: "" (empty string) returns false', () => {
+  assert.equal(isClosedStatus(''), false);
+});