gsd-pi 2.63.0-dev.026d309 → 2.63.0-dev.351157b

package/src/resources/extensions/gsd/tests/deferred-slice-dispatch.test.ts

@@ -0,0 +1,203 @@
+/**
+ * Regression test for #2661: Auto-mode dispatches deferred slices.
+ *
+ * When a decision defers a slice, the dispatcher must skip it and advance
+ * to the next eligible slice. This tests both:
+ *   1. deriveStateFromDb skips slices with status "deferred"
+ *   2. saveDecisionToDb updates the slice status when the decision is a deferral
+ */
+
+import { describe, test } from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync, mkdirSync, rmSync, writeFileSync } from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+
+import { deriveStateFromDb, invalidateStateCache } from "../state.ts";
+import {
+  openDatabase,
+  closeDatabase,
+  isDbAvailable,
+  insertMilestone,
+  insertSlice,
+  insertTask,
+  insertArtifact,
+  updateSliceStatus,
+} from "../gsd-db.ts";
+import { isDeferredStatus } from "../status-guards.ts";
+
+// ─── Helpers ──────────────────────────────────────────────────────────────
+
+function createFixtureBase(): string {
+  const base = mkdtempSync(join(tmpdir(), "gsd-deferred-dispatch-"));
+  mkdirSync(join(base, ".gsd", "milestones"), { recursive: true });
+  return base;
+}
+
+function writeFile(base: string, relativePath: string, content: string): void {
+  const full = join(base, ".gsd", relativePath);
+  mkdirSync(join(full, ".."), { recursive: true });
+  writeFileSync(full, content);
+}
+
+function cleanup(base: string): void {
+  rmSync(base, { recursive: true, force: true });
+}
+
+// ─── Tests ────────────────────────────────────────────────────────────────
+
+describe("deferred-slice-dispatch (#2661)", () => {
+  test("isDeferredStatus returns true for 'deferred'", () => {
+    assert.ok(isDeferredStatus("deferred"), "should recognize 'deferred'");
+    assert.ok(!isDeferredStatus("active"), "should not match 'active'");
+    assert.ok(!isDeferredStatus("complete"), "should not match 'complete'");
+    assert.ok(!isDeferredStatus("pending"), "should not match 'pending'");
+  });
+
+  test("deriveStateFromDb skips deferred slice and picks next eligible", async () => {
+    const base = createFixtureBase();
+    try {
+      openDatabase(":memory:");
+      assert.ok(isDbAvailable());
+
+      // M001 with three slices: S01 complete, S02 deferred, S03 pending
+      insertMilestone({ id: "M001", title: "Test Milestone", status: "active" });
+
+      insertSlice({ id: "S01", milestoneId: "M001", title: "Done Slice", status: "complete", risk: "low", depends: [] });
+      insertSlice({ id: "S02", milestoneId: "M001", title: "Deferred Slice", status: "deferred", risk: "low", depends: [] });
+      insertSlice({ id: "S03", milestoneId: "M001", title: "Next Slice", status: "pending", risk: "low", depends: [] });
+
+      // S01 needs a SUMMARY file to count as complete for milestone-level checks
+      writeFile(base, "milestones/M001/M001-ROADMAP.md", `# M001: Test Milestone
+
+**Vision:** Test deferred slices.
+
+## Slices
+
+- [x] **S01: Done Slice** \`risk:low\` \`depends:[]\`
+  > Done.
+
+- [ ] **S02: Deferred Slice** \`risk:low\` \`depends:[]\`
+  > Deferred.
+
+- [ ] **S03: Next Slice** \`risk:low\` \`depends:[]\`
+  > Next.
+`);
+      writeFile(base, "milestones/M001/slices/S01/S01-SUMMARY.md", "# S01 Summary\nDone.");
+
+      invalidateStateCache();
+      const state = await deriveStateFromDb(base);
+
+      // The active slice must be S03, NOT S02 (which is deferred)
+      assert.equal(state.activeMilestone?.id, "M001", "active milestone is M001");
+      assert.equal(state.activeSlice?.id, "S03", "active slice should skip deferred S02 and land on S03");
+      assert.notEqual(state.activeSlice?.id, "S02", "active slice must NOT be the deferred S02");
+
+      closeDatabase();
+    } finally {
+      closeDatabase();
+      cleanup(base);
+    }
+  });
+
+  test("deriveStateFromDb does not count deferred slices as done for progress", async () => {
+    const base = createFixtureBase();
+    try {
+      openDatabase(":memory:");
+
+      insertMilestone({ id: "M001", title: "Test", status: "active" });
+      insertSlice({ id: "S01", milestoneId: "M001", title: "Complete", status: "complete", risk: "low", depends: [] });
+      insertSlice({ id: "S02", milestoneId: "M001", title: "Deferred", status: "deferred", risk: "low", depends: [] });
+      insertSlice({ id: "S03", milestoneId: "M001", title: "Pending", status: "pending", risk: "low", depends: [] });
+
+      writeFile(base, "milestones/M001/M001-ROADMAP.md", `# M001
+## Slices
+- [x] **S01: Complete** \`risk:low\` \`depends:[]\`
+- [ ] **S02: Deferred** \`risk:low\` \`depends:[]\`
+- [ ] **S03: Pending** \`risk:low\` \`depends:[]\`
+`);
+      writeFile(base, "milestones/M001/slices/S01/S01-SUMMARY.md", "# Done");
+
+      invalidateStateCache();
+      const state = await deriveStateFromDb(base);
+
+      // Deferred slices should not count as "done" in progress
+      // Only S01 (complete) counts as done
+      assert.equal(state.progress?.slices?.done, 1, "only 1 slice (S01) should be done");
+      // Total should still be 3 (deferred slices are still part of the milestone)
+      assert.equal(state.progress?.slices?.total, 3, "all 3 slices counted in total");
+
+      closeDatabase();
+    } finally {
+      closeDatabase();
+      cleanup(base);
+    }
+  });
+
+  test("all slices deferred results in blocked state", async () => {
+    const base = createFixtureBase();
+    try {
+      openDatabase(":memory:");
+
+      insertMilestone({ id: "M001", title: "Test", status: "active" });
+      insertSlice({ id: "S01", milestoneId: "M001", title: "Deferred A", status: "deferred", risk: "low", depends: [] });
+      insertSlice({ id: "S02", milestoneId: "M001", title: "Deferred B", status: "deferred", risk: "low", depends: [] });
+
+      writeFile(base, "milestones/M001/M001-ROADMAP.md", `# M001
+## Slices
+- [ ] **S01: Deferred A** \`risk:low\` \`depends:[]\`
+- [ ] **S02: Deferred B** \`risk:low\` \`depends:[]\`
+`);
+
+      invalidateStateCache();
+      const state = await deriveStateFromDb(base);
+
+      // No eligible slice — should be blocked
+      assert.equal(state.activeSlice, null, "no active slice when all deferred");
+      assert.equal(state.phase, "blocked", "phase should be blocked when all slices deferred");
+
+      closeDatabase();
+    } finally {
+      closeDatabase();
+      cleanup(base);
+    }
+  });
+
+  test("saveDecisionToDb marks slice as deferred when decision is a deferral", async () => {
+    const base = createFixtureBase();
+    try {
+      openDatabase(":memory:");
+
+      insertMilestone({ id: "M001", title: "Test", status: "active" });
+      insertSlice({ id: "S03", milestoneId: "M001", title: "Target Slice", status: "active", risk: "low", depends: [] });
+
+      writeFile(base, "milestones/M001/M001-ROADMAP.md", `# M001
+## Slices
+- [ ] **S03: Target Slice** \`risk:low\` \`depends:[]\`
+`);
+
+      const { saveDecisionToDb } = await import("../db-writer.ts");
+      const { getSlice } = await import("../gsd-db.ts");
+
+      // Save a deferral decision that references M001/S03
+      await saveDecisionToDb(
+        {
+          scope: "deferral",
+          decision: "Defer S03 to focus on higher priority work",
+          choice: "defer M001/S03",
+          rationale: "Not ready yet",
+        },
+        base,
+      );
+
+      // The slice status should now be "deferred"
+      const slice = getSlice("M001", "S03");
+      assert.equal(slice?.status, "deferred", "slice status should be updated to 'deferred' after deferral decision");
+
+      closeDatabase();
+    } finally {
+      closeDatabase();
+      cleanup(base);
+    }
+  });
+});

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"));
+});