gsd-pi 2.62.1 → 2.63.0-dev.351157b

package/src/resources/extensions/gsd/tests/discuss-tool-scoping.test.ts

@@ -0,0 +1,130 @@
+/**
+ * discuss-tool-scoping.test.ts — Tests for #2949.
+ *
+ * xAI/Grok returns "Grammar is too complex" (400) when the combined tool
+ * schemas exceed the provider's grammar limit. The GSD discuss flow only
+ * needs a small subset of tools (summary_save, decision_save, etc.), but
+ * was sending ALL ~30+ tools to the provider.
+ *
+ * These tests verify:
+ *   1. DISCUSS_TOOLS_ALLOWLIST is exported and contains only the tools
+ *      needed during discuss flows (no heavy planning/execution/completion tools).
+ *   2. Heavy execution tools are NOT in the allowlist.
+ *   3. The allowlist includes the tools actually referenced by discuss prompts.
+ *   4. dispatchWorkflow scopes tools when unitType is a discuss variant.
+ */
+
+import { describe, 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";
+
+import { DISCUSS_TOOLS_ALLOWLIST } from "../constants.ts";
+
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const promptsDir = join(__dirname, "..", "prompts");
+const guidedFlowPath = join(__dirname, "..", "guided-flow.ts");
+
+// ─── Heavy tools that should NOT be in discuss scope ─────────────────────────
+
+/** Tools that are only needed during planning, execution, or completion phases */
+const HEAVY_TOOLS = [
+  "gsd_plan_slice",
+  "gsd_slice_plan",
+  "gsd_plan_task",
+  "gsd_task_plan",
+  "gsd_task_complete",
+  "gsd_complete_task",
+  "gsd_slice_complete",
+  "gsd_complete_slice",
+  "gsd_complete_milestone",
+  "gsd_milestone_complete",
+  "gsd_validate_milestone",
+  "gsd_milestone_validate",
+  "gsd_replan_slice",
+  "gsd_slice_replan",
+  "gsd_reassess_roadmap",
+  "gsd_roadmap_reassess",
+  "gsd_save_gate_result",
+];
+
+// ─── Tools that discuss prompts reference ────────────────────────────────────
+
+/** Tools explicitly called by discuss prompt templates */
+const DISCUSS_REQUIRED_TOOLS = [
+  "gsd_summary_save",          // guided-discuss-slice.md, guided-discuss-milestone.md, discuss.md
+  "gsd_decision_save",         // discuss.md output phase
+  "gsd_plan_milestone",        // discuss.md output phase (single + multi milestone)
+  "gsd_milestone_generate_id", // discuss.md multi-milestone Phase 1
+  "gsd_requirement_update",    // used during discuss for requirement updates
+];
+
+// ─── Tests ───────────────────────────────────────────────────────────────────
+
+describe("discuss tool scoping (#2949)", () => {
+  test("DISCUSS_TOOLS_ALLOWLIST is exported and non-empty", () => {
+    assert.ok(Array.isArray(DISCUSS_TOOLS_ALLOWLIST), "should be an array");
+    assert.ok(DISCUSS_TOOLS_ALLOWLIST.length > 0, "should not be empty");
+  });
+
+  test("DISCUSS_TOOLS_ALLOWLIST excludes heavy execution/completion tools", () => {
+    for (const heavy of HEAVY_TOOLS) {
+      assert.ok(
+        !DISCUSS_TOOLS_ALLOWLIST.includes(heavy),
+        `allowlist should NOT include heavy tool "${heavy}"`,
+      );
+    }
+  });
+
+  test("DISCUSS_TOOLS_ALLOWLIST includes tools referenced by discuss prompts", () => {
+    for (const required of DISCUSS_REQUIRED_TOOLS) {
+      assert.ok(
+        DISCUSS_TOOLS_ALLOWLIST.includes(required),
+        `allowlist should include "${required}" (used by discuss prompts)`,
+      );
+    }
+  });
+
+  test("DISCUSS_TOOLS_ALLOWLIST is significantly smaller than full tool set", () => {
+    // Full set is 27 DB tools + dynamic + journal = 33+
+    // Discuss set should be roughly 10 GSD tools (5 canonical + 5 aliases)
+    assert.ok(
+      DISCUSS_TOOLS_ALLOWLIST.length <= 12,
+      `allowlist should have at most 12 GSD tools, got ${DISCUSS_TOOLS_ALLOWLIST.length}`,
+    );
+  });
+
+  test("guided-discuss-slice.md references gsd_summary_save", () => {
+    const prompt = readFileSync(join(promptsDir, "guided-discuss-slice.md"), "utf-8");
+    assert.ok(
+      prompt.includes("gsd_summary_save"),
+      "guided-discuss-slice.md should reference gsd_summary_save",
+    );
+  });
+
+  test("discuss.md references gsd_plan_milestone and gsd_decision_save", () => {
+    const prompt = readFileSync(join(promptsDir, "discuss.md"), "utf-8");
+    assert.ok(
+      prompt.includes("gsd_plan_milestone"),
+      "discuss.md should reference gsd_plan_milestone",
+    );
+    assert.ok(
+      prompt.includes("gsd_decision_save"),
+      "discuss.md should reference gsd_decision_save",
+    );
+  });
+
+  test("dispatchWorkflow source code scopes tools for discuss unit types", () => {
+    const source = readFileSync(guidedFlowPath, "utf-8");
+    // Verify that dispatchWorkflow references the allowlist for tool scoping
+    assert.ok(
+      source.includes("DISCUSS_TOOLS_ALLOWLIST"),
+      "guided-flow.ts should reference DISCUSS_TOOLS_ALLOWLIST for tool scoping",
+    );
+    assert.ok(
+      source.includes("setActiveTools"),
+      "guided-flow.ts should call setActiveTools to scope tools during discuss",
+    );
+  });
+});

package/src/resources/extensions/gsd/tests/doctor-fix-flag.test.ts

@@ -0,0 +1,92 @@
+/**
+ * Regression test for #1919: --fix flag not stripped before positional parse.
+ *
+ * parseDoctorArgs("--fix") must:
+ *   1. Set fixFlag = true
+ *   2. Not leak "--fix" into requestedScope
+ *   3. Keep mode as "doctor" (the flag is not a positional subcommand)
+ */
+
+import { parseDoctorArgs } from "../commands-handlers.js";
+import { createTestContext } from "./test-helpers.ts";
+
+const { assertEq, assertTrue, report } = createTestContext();
+
+async function main(): Promise<void> {
+  // ── 1. Bare --fix flag ──────────────────────────────────────────────────────
+  console.log("\n=== bare --fix flag (#1919) ===");
+  {
+    const r = parseDoctorArgs("--fix");
+    assertTrue(r.fixFlag, "--fix sets fixFlag to true");
+    assertEq(r.mode, "doctor", "--fix does not change mode from doctor");
+    assertEq(r.requestedScope, undefined, "--fix is stripped and does not become requestedScope");
+  }
+
+  // ── 2. --fix with a scope ──────────────────────────────────────────────────
+  console.log("\n=== --fix with scope ===");
+  {
+    const r = parseDoctorArgs("--fix M001/S01");
+    assertTrue(r.fixFlag, "--fix M001/S01 sets fixFlag to true");
+    assertEq(r.mode, "doctor", "--fix M001/S01 keeps mode as doctor");
+    assertEq(r.requestedScope, "M001/S01", "scope is M001/S01 after stripping --fix");
+  }
+
+  // ── 3. Positional fix still works ──────────────────────────────────────────
+  console.log("\n=== positional fix subcommand ===");
+  {
+    const r = parseDoctorArgs("fix");
+    assertEq(r.fixFlag, false, "positional fix does not set fixFlag");
+    assertEq(r.mode, "fix", "positional fix sets mode to fix");
+    assertEq(r.requestedScope, undefined, "no scope with bare positional fix");
+  }
+
+  // ── 4. Positional fix with scope ───────────────────────────────────────────
+  console.log("\n=== positional fix with scope ===");
+  {
+    const r = parseDoctorArgs("fix M001");
+    assertEq(r.mode, "fix", "fix M001 sets mode to fix");
+    assertEq(r.requestedScope, "M001", "fix M001 parses scope as M001");
+  }
+
+  // ── 5. --fix combined with other flags ─────────────────────────────────────
+  console.log("\n=== --fix combined with --dry-run ===");
+  {
+    const r = parseDoctorArgs("--fix --dry-run");
+    assertTrue(r.fixFlag, "--fix --dry-run sets fixFlag");
+    assertTrue(r.dryRun, "--fix --dry-run sets dryRun");
+    assertEq(r.requestedScope, undefined, "no scope leaked from combined flags");
+  }
+
+  // ── 6. --fix combined with --json ──────────────────────────────────────────
+  console.log("\n=== --fix with --json ===");
+  {
+    const r = parseDoctorArgs("--fix --json");
+    assertTrue(r.fixFlag, "--fix --json sets fixFlag");
+    assertTrue(r.jsonMode, "--fix --json sets jsonMode");
+    assertEq(r.requestedScope, undefined, "no scope leaked from --fix --json");
+  }
+
+  // ── 7. Empty args (baseline) ───────────────────────────────────────────────
+  console.log("\n=== empty args baseline ===");
+  {
+    const r = parseDoctorArgs("");
+    assertEq(r.fixFlag, false, "empty args: fixFlag false");
+    assertEq(r.mode, "doctor", "empty args: mode is doctor");
+    assertEq(r.requestedScope, undefined, "empty args: no scope");
+  }
+
+  // ── 8. heal and audit modes unaffected ─────────────────────────────────────
+  console.log("\n=== heal and audit modes ===");
+  {
+    const rh = parseDoctorArgs("heal M001/S01");
+    assertEq(rh.mode, "heal", "heal mode parsed correctly");
+    assertEq(rh.requestedScope, "M001/S01", "heal scope parsed correctly");
+
+    const ra = parseDoctorArgs("audit");
+    assertEq(ra.mode, "audit", "audit mode parsed correctly");
+  }
+
+  report();
+}
+
+main();

package/src/resources/extensions/gsd/tests/finalize-timeout-guard.test.ts

@@ -0,0 +1,116 @@
+/**
+ * Regression test for #2344: Auto-loop hangs after plan-slice completes
+ * because postUnitPostVerification() never resolves.
+ *
+ * When postUnitPostVerification() hangs (e.g., due to a module import
+ * deadlock or SQLite transaction hang), the auto-loop blocks forever
+ * with no error message, no notification, and no recovery.
+ *
+ * The fix adds a timeout guard around postUnitPostVerification() in
+ * runFinalize(). If it doesn't resolve within the timeout, the function
+ * force-returns "continue" and logs an error, allowing the loop to
+ * proceed to the next iteration.
+ *
+ * This test verifies the timeout utility used by the fix, since the
+ * full runFinalize function has too many transitive dependencies for
+ * isolated unit testing.
+ */
+
+import { createTestContext } from "./test-helpers.ts";
+import {
+  withTimeout,
+  FINALIZE_POST_TIMEOUT_MS,
+} from "../auto/finalize-timeout.ts";
+
+const { assertTrue, assertEq, report } = createTestContext();
+
+// ═══ Test: withTimeout resolves when inner promise resolves promptly ══════════
+
+{
+  console.log("\n=== #2344: withTimeout passes through when promise resolves ===");
+
+  const result = await withTimeout(
+    Promise.resolve("ok"),
+    1000,
+    "test-timeout",
+  );
+  assertEq(result.value, "ok", "should return inner value");
+  assertEq(result.timedOut, false, "should not be timed out");
+}
+
+// ═══ Test: withTimeout returns fallback when inner promise hangs ══════════════
+
+{
+  console.log("\n=== #2344: withTimeout returns fallback on hang ===");
+
+  const startTime = Date.now();
+  const result = await withTimeout(
+    new Promise<string>(() => {
+      // Never resolves
+    }),
+    100, // short timeout for testing
+    "test-timeout",
+  );
+  const elapsed = Date.now() - startTime;
+
+  assertEq(result.timedOut, true, "should report timeout");
+  assertEq(result.value, undefined, "value should be undefined on timeout");
+  assertTrue(elapsed >= 90, `should wait at least 90ms (took ${elapsed}ms)`);
+  assertTrue(elapsed < 500, `should not wait too long (took ${elapsed}ms)`);
+}
+
+// ═══ Test: withTimeout handles rejection gracefully ═══════════════════════════
+
+{
+  console.log("\n=== #2344: withTimeout propagates rejection ===");
+
+  let caught = false;
+  try {
+    await withTimeout(
+      Promise.reject(new Error("boom")),
+      1000,
+      "test-timeout",
+    );
+  } catch (err: any) {
+    caught = true;
+    assertEq(err.message, "boom", "should propagate the error");
+  }
+  assertTrue(caught, "rejection should propagate");
+}
+
+// ═══ Test: FINALIZE_POST_TIMEOUT_MS is defined and reasonable ═════════════════
+
+{
+  console.log("\n=== #2344: timeout constant is defined and reasonable ===");
+
+  assertTrue(
+    typeof FINALIZE_POST_TIMEOUT_MS === "number",
+    "FINALIZE_POST_TIMEOUT_MS should be a number",
+  );
+  assertTrue(
+    FINALIZE_POST_TIMEOUT_MS >= 30_000,
+    `timeout should be >= 30s (got ${FINALIZE_POST_TIMEOUT_MS}ms)`,
+  );
+  assertTrue(
+    FINALIZE_POST_TIMEOUT_MS <= 120_000,
+    `timeout should be <= 120s (got ${FINALIZE_POST_TIMEOUT_MS}ms)`,
+  );
+}
+
+// ═══ Test: withTimeout cleans up timer on success ════════════════════════════
+
+{
+  console.log("\n=== #2344: withTimeout cleans up timer on success ===");
+
+  // If the timer isn't cleaned up, this test would keep the process alive.
+  // Relying on process.exit behavior — if test completes, timers were cleaned.
+  const result = await withTimeout(
+    new Promise<string>((r) => setTimeout(() => r("delayed"), 50)),
+    5000,
+    "cleanup-test",
+  );
+  assertEq(result.value, "delayed", "should resolve with delayed value");
+  assertEq(result.timedOut, false, "should not time out");
+}
+
+report();

package/src/resources/extensions/gsd/tests/forensics-stuck-loops.test.ts

@@ -0,0 +1,103 @@
+/**
+ * Forensics detectStuckLoops tests — #1943
+ *
+ * Verifies that detectStuckLoops counts distinct dispatches (unique startedAt
+ * values per type/id) instead of raw entry count, which produces false-positive
+ * stuck-loop anomalies when idle-watchdog duplicate metrics entries exist.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import type { UnitMetrics } from "../metrics.js";
+import { detectStuckLoops, type ForensicAnomaly } from "../forensics.js";
+
+// ── Helpers ──────────────────────────────────────────────────────────────────
+
+function makeUnit(overrides: Partial<UnitMetrics> = {}): UnitMetrics {
+  return {
+    type: "execute-task",
+    id: "M001/S01/T01",
+    model: "claude-sonnet-4-20250514",
+    startedAt: 1000,
+    finishedAt: 2000,
+    tokens: { input: 1000, output: 500, cacheRead: 200, cacheWrite: 100, total: 1800 },
+    cost: 0.05,
+    toolCalls: 3,
+    assistantMessages: 2,
+    userMessages: 1,
+    ...overrides,
+  };
+}
+
+
+// ── Tests ────────────────────────────────────────────────────────────────────
+
+test("#1943 detectStuckLoops does not flag idle-watchdog duplicates as stuck loops", () => {
+  const anomalies: ForensicAnomaly[] = [];
+  const startedAt = 1774011016218;
+
+  // 20 entries with the SAME startedAt — these are idle-watchdog duplicates,
+  // not real re-dispatches. They should count as 1 dispatch.
+  const units: UnitMetrics[] = [];
+  for (let i = 0; i < 20; i++) {
+    units.push(makeUnit({
+      type: "research-slice",
+      id: "M009/S02",
+      startedAt,
+      finishedAt: startedAt + (i + 1) * 15000,
+      cost: 1.50 + i * 0.05,
+      toolCalls: 0,
+    }));
+  }
+
+  detectStuckLoops(units, anomalies);
+
+  // A single dispatch (same startedAt) should NOT trigger a stuck-loop anomaly
+  assert.equal(
+    anomalies.length, 0,
+    `expected 0 anomalies for 20 watchdog snapshots of the same dispatch, got ${anomalies.length}: ${anomalies.map(a => a.summary).join(", ")}`,
+  );
+});
+
+test("#1943 detectStuckLoops correctly flags real re-dispatches", () => {
+  const anomalies: ForensicAnomaly[] = [];
+
+  // 3 entries with DIFFERENT startedAt values — these are real re-dispatches
+  const units: UnitMetrics[] = [
+    makeUnit({ type: "execute-task", id: "M001/S01/T01", startedAt: 1000, finishedAt: 2000, cost: 0.05 }),
+    makeUnit({ type: "execute-task", id: "M001/S01/T01", startedAt: 3000, finishedAt: 4000, cost: 0.06 }),
+    makeUnit({ type: "execute-task", id: "M001/S01/T01", startedAt: 5000, finishedAt: 6000, cost: 0.07 }),
+  ];
+
+  detectStuckLoops(units, anomalies);
+
+  assert.equal(anomalies.length, 1, "3 distinct dispatches of the same unit should flag 1 anomaly");
+  assert.equal(anomalies[0].type, "stuck-loop");
+  assert.ok(anomalies[0].summary.includes("3 times"), `summary should mention 3 dispatches: ${anomalies[0].summary}`);
+});
+
+test("#1943 detectStuckLoops ignores watchdog duplicates but flags real re-dispatches in mixed data", () => {
+  const anomalies: ForensicAnomaly[] = [];
+
+  const units: UnitMetrics[] = [
+    // 5 watchdog duplicates for dispatch 1 (same startedAt = 1000)
+    ...Array.from({ length: 5 }, (_, i) =>
+      makeUnit({ type: "execute-task", id: "M001/S01/T01", startedAt: 1000, finishedAt: 1000 + (i + 1) * 15000, cost: 0.05 + i * 0.01 }),
+    ),
+    // 3 watchdog duplicates for dispatch 2 (same startedAt = 100000)
+    ...Array.from({ length: 3 }, (_, i) =>
+      makeUnit({ type: "execute-task", id: "M001/S01/T01", startedAt: 100000, finishedAt: 100000 + (i + 1) * 15000, cost: 0.08 + i * 0.01 }),
+    ),
+    // 1 entry for dispatch 3 (startedAt = 200000)
+    makeUnit({ type: "execute-task", id: "M001/S01/T01", startedAt: 200000, finishedAt: 260000, cost: 0.10 }),
+    // Different unit — only 1 dispatch, should NOT be flagged
+    makeUnit({ type: "plan-slice", id: "M001/S01", startedAt: 500, finishedAt: 1500, cost: 0.02 }),
+  ];
+
+  detectStuckLoops(units, anomalies);
+
+  // M001/S01/T01 has 3 distinct dispatches (startedAt: 1000, 100000, 200000) — should be flagged
+  // M001/S01 has 1 dispatch — should NOT be flagged
+  assert.equal(anomalies.length, 1, `expected 1 anomaly (for the 3x dispatched task), got ${anomalies.length}`);
+  assert.ok(anomalies[0].summary.includes("3 times"));
+});

package/src/resources/extensions/gsd/tests/insert-slice-no-wipe.test.ts

@@ -0,0 +1,88 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+
+import { openDatabase, closeDatabase, insertMilestone, insertSlice, getSlice } from '../gsd-db.ts';
+
+test('insertSlice with minimal args does not wipe populated fields', (t) => {
+  t.after(() => { try { closeDatabase(); } catch { /* noop */ } });
+  openDatabase(":memory:");
+
+  insertMilestone({ id: 'M001', title: 'Milestone', status: 'active' });
+
+  // First insert: full data
+  insertSlice({
+    id: 'S01',
+    milestoneId: 'M001',
+    title: 'Auth flow',
+    status: 'in-progress',
+    risk: 'high',
+    demo: 'Login page renders.',
+    sequence: 3,
+    planning: {
+      goal: 'Secure authentication',
+      successCriteria: 'All tests pass',
+      proofLevel: 'integration',
+      integrationClosure: 'Fully integrated',
+      observabilityImpact: 'Metrics available',
+    },
+  });
+
+  const before = getSlice('M001', 'S01');
+  assert.ok(before, 'slice should exist after first insert');
+  assert.equal(before.title, 'Auth flow');
+  assert.equal(before.demo, 'Login page renders.');
+  assert.equal(before.risk, 'high');
+
+  // Second insert: minimal "ensure exists" call (mirrors complete-task.ts usage)
+  insertSlice({ id: 'S01', milestoneId: 'M001' });
+
+  const after = getSlice('M001', 'S01');
+  assert.ok(after, 'slice should still exist after second insert');
+
+  // These must NOT be wiped to empty strings
+  assert.equal(after.title, 'Auth flow', 'title must survive minimal re-insert');
+  assert.equal(after.demo, 'Login page renders.', 'demo must survive minimal re-insert');
+  assert.equal(after.risk, 'high', 'risk must survive minimal re-insert');
+  assert.equal(after.sequence, 3, 'sequence must survive minimal re-insert');
+
+  // Planning fields must also survive
+  assert.equal(after.goal, 'Secure authentication', 'goal must survive minimal re-insert');
+  assert.equal(after.success_criteria, 'All tests pass', 'success_criteria must survive');
+  assert.equal(after.proof_level, 'integration', 'proof_level must survive');
+  assert.equal(after.integration_closure, 'Fully integrated', 'integration_closure must survive');
+  assert.equal(after.observability_impact, 'Metrics available', 'observability_impact must survive');
+});
+
+test('insertSlice ON CONFLICT preserves completed status', (t) => {
+  t.after(() => { try { closeDatabase(); } catch { /* noop */ } });
+  openDatabase(":memory:");
+
+  insertMilestone({ id: 'M001', title: 'Milestone', status: 'active' });
+
+  insertSlice({ id: 'S01', milestoneId: 'M001', title: 'Done slice', status: 'complete' });
+
+  // Re-insert with pending status (default) should NOT overwrite complete
+  insertSlice({ id: 'S01', milestoneId: 'M001' });
+
+  const after = getSlice('M001', 'S01');
+  assert.ok(after);
+  assert.equal(after.status, 'complete', 'completed status must not be overwritten');
+});
+
+test('insertSlice ON CONFLICT allows explicit updates to non-empty values', (t) => {
+  t.after(() => { try { closeDatabase(); } catch { /* noop */ } });
+  openDatabase(":memory:");
+
+  insertMilestone({ id: 'M001', title: 'Milestone', status: 'active' });
+
+  insertSlice({ id: 'S01', milestoneId: 'M001', title: 'Original', demo: 'Old demo', risk: 'low' });
+
+  // Explicit update with real values should overwrite
+  insertSlice({ id: 'S01', milestoneId: 'M001', title: 'Updated', demo: 'New demo', risk: 'high' });
+
+  const after = getSlice('M001', 'S01');
+  assert.ok(after);
+  assert.equal(after.title, 'Updated', 'explicit title update should apply');
+  assert.equal(after.demo, 'New demo', 'explicit demo update should apply');
+  assert.equal(after.risk, 'high', 'explicit risk update should apply');
+});

package/src/resources/extensions/gsd/tests/integration/git-service.test.ts

@@ -1246,7 +1246,7 @@ describe('git-service', async () => {
   test('nativeAddAllWithExclusions: symlinked .gsd fallback', () => {
     // When .gsd is a symlink, git rejects `:!.gsd/...` pathspecs with
     // "fatal: pathspec '...' is beyond a symbolic link". The fix falls
-    // back to plain `git add -A`, which respects .gitignore.
+    // back to `git add -u` (tracked files only), NOT `git add -A`.
     const repo = initTempRepo();
 
     // Create the real .gsd directory outside the repo, then symlink it
@@ -1258,11 +1258,18 @@ describe('git-service', async () => {
     // Symlink .gsd -> external directory
     symlinkSync(externalGsd, join(repo, ".gsd"));
 
-    // Add .gitignore so git add -A fallback skips .gsd/
+    // Add .gitignore so .gsd/ is ignored
     writeFileSync(join(repo, ".gitignore"), ".gsd\n");
 
-    // Create a real file that should be staged
+    // Create a tracked file and commit it, then modify it
     createFile(repo, "src/app.ts", "export const x = 1;");
+    run("git add -A", repo);
+    run('git commit -m "add app"', repo);
+    writeFileSync(join(repo, "src/app.ts"), "export const x = 2;");
+
+    // Create an untracked file simulating large data (NOT in .gitignore)
+    // This is the key scenario: large untracked dirs that git add -A would traverse
+    createFile(repo, "data/large-model.bin", "pretend this is 10GB");
 
     // nativeAddAllWithExclusions should NOT throw despite .gsd being a symlink
     let threw = false;
@@ -1274,9 +1281,15 @@ describe('git-service', async () => {
     }
     assert.ok(!threw, "nativeAddAllWithExclusions does not throw with symlinked .gsd");
 
-    // Verify the real file was staged
+    // Verify the tracked modified file was staged
     const staged = run("git diff --cached --name-only", repo);
-    assert.ok(staged.includes("src/app.ts"), "real file staged despite symlinked .gsd");
+    assert.ok(staged.includes("src/app.ts"), "modified tracked file staged despite symlinked .gsd");
+
+    // CRITICAL: untracked files must NOT be staged — the symlink fallback
+    // should use `git add -u` (tracked only), not `git add -A` (all files).
+    // Using `git add -A` on a repo with large untracked data dirs hangs. (#1977)
+    assert.ok(!staged.includes("data/large-model.bin"),
+      "symlink fallback must not stage untracked files (would hang on large repos)");
     assert.ok(!staged.includes(".gsd"), ".gsd content not staged");
 
     rmSync(repo, { recursive: true, force: true });
@@ -1435,13 +1448,20 @@ describe('git-service', async () => {
     run('git add .gitignore', repo);
     run('git commit -m "add gitignore"', repo);
 
+    // Pre-commit a tracked source file so git add -u can stage modifications.
+    // The symlink fallback uses git add -u (tracked files only), so the file
+    // must be tracked before the autoCommit scenario runs.
+    createFile(repo, "src/feature.ts", "export const feature = true;");
+    run('git add src/feature.ts', repo);
+    run('git commit -m "add feature"', repo);
+
     // Simulate new milestone artifacts created during execution
     writeFileSync(join(externalGsd, "milestones", "M009", "M009-SUMMARY.md"), "# M009 Summary");
     writeFileSync(join(externalGsd, "milestones", "M009", "S01-SUMMARY.md"), "# S01 Summary");
     writeFileSync(join(externalGsd, "milestones", "M009", "T01-VERIFY.json"), '{"passed":true}');
 
-    // Also create a normal source file change
-    createFile(repo, "src/feature.ts", "export const feature = true;");
+    // Modify the tracked source file — git add -u will stage this change
+    writeFileSync(join(repo, "src/feature.ts"), "export const feature = false; // updated");
 
     const svc = new GitServiceImpl(repo);
     const msg = svc.autoCommit("complete-milestone", "M009");

package/src/resources/extensions/gsd/tests/integration/idle-recovery.test.ts

@@ -357,3 +357,37 @@ test('writeBlockerPlaceholder: does NOT update DB for non-execute-task types', a
     cleanup(base);
   }
 });
+
+test('writeBlockerPlaceholder: updates DB slice status for complete-slice (#2653)', async () => {
+  const base = createFixtureBase();
+  try {
+    const { openDatabase, closeDatabase, insertMilestone, insertSlice, getSlice, isDbAvailable } =
+      await import("../../gsd-db.ts");
+
+    const dbPath = join(base, ".gsd", "gsd.db");
+    mkdirSync(join(base, ".gsd", "milestones", "M001", "slices", "S01"), { recursive: true });
+
+    openDatabase(dbPath);
+    try {
+      insertMilestone({ id: "M001", title: "Test", status: "active" });
+      insertSlice({ id: "S01", milestoneId: "M001", title: "Slice", status: "active" });
+
+      // complete-slice blocker should update slice DB status to "complete"
+      writeBlockerPlaceholder("complete-slice", "M001/S01", base, "context exhaustion recovery");
+
+      const slice = getSlice("M001", "S01");
+      assert.equal(slice?.status, "complete",
+        "writeBlockerPlaceholder must update DB slice status to 'complete' for complete-slice so dispatch guard unblocks downstream (#2653)");
+
+      // Verify the full chain works: verifyExpectedArtifact should return true
+      // (requires both UAT file and DB status = complete)
+      // Note: the placeholder writes a SUMMARY file, but complete-slice also needs UAT.
+      // The placeholder itself doesn't write UAT, so artifact verification may still fail
+      // for complete-slice — but the DB status is now correct, breaking the circular dep.
+    } finally {
+      if (isDbAvailable()) closeDatabase();
+    }
+  } finally {
+    cleanup(base);
+  }
+});