gsd-pi 2.77.0 → 2.78.0-dev.aeeb2ca00

package/src/resources/extensions/gsd/tests/test-helpers.ts

@@ -59,3 +59,156 @@ export function createTestContext() {
 
   return { assertEq, assertTrue, assertMatch, assertNoMatch, report };
 }
+
+// ─── Source-inspection helpers (DEPRECATED — do not add new callers) ──────────
+//
+// `extractSourceRegion` was introduced in #4773 / #4774 as a more-durable
+// replacement for hand-rolled `src.slice(idx, idx + N)` patterns. Issue
+// #4784 surfaced that the helper is the wrong target: its use case —
+// testing that an identifier or phrase exists in production source text
+// — is the source-grep antipattern that causes false coverage in the
+// first place. Tests should exercise behaviour (call the function,
+// assert on its return value or observable effects) rather than check
+// whether specific identifier strings remain in source.
+//
+// CONTRIBUTING.md forbids new callers of this helper. Existing callers
+// are being migrated one file at a time; the helper is slated for
+// removal once the last caller is converted.
+//
+// Do not add new callers. If you are tempted to, file an issue
+// describing the behavioural invariant you want to test and we will
+// find a runtime assertion instead.
+
+/**
+ * @deprecated Use a real behaviour test against the function under
+ *   inspection instead. See #4784 and CONTRIBUTING.md > "Test behaviour,
+ *   not source shape". No new callers.
+ *
+ * Extract a region of source between a start anchor and either an explicit
+ * end anchor or, if none is given, a set of reasonable structural
+ * terminators (next `private `/`export `/`function `/`class `/`interface `/
+ * `//` section separator). Falls back to end-of-source if none match.
+ *
+ * @param src        The source text.
+ * @param startAnchor Literal substring that marks the start of the region.
+ *                   Typically a function name, a section header comment, or
+ *                   a distinctive statement.
+ * @param endAnchor  Optional. Either a literal substring that marks the end,
+ *                   or `{ fromIdx: number }` to find the *next* occurrence
+ *                   of `startAnchor` at or after `fromIdx` (useful when the
+ *                   anchor appears multiple times and a positional search
+ *                   is required). When omitted, structural terminators are
+ *                   used.
+ *
+ * Returns the extracted region (including the start anchor), or an empty
+ * string if `startAnchor` is not found.
+ */
+export function extractSourceRegion(
+  src: string,
+  startAnchor: string,
+  endAnchor?: string | { fromIdx: number },
+): string {
+  const fromIdx = typeof endAnchor === "object" && endAnchor !== null
+    ? endAnchor.fromIdx
+    : 0;
+  const endLiteral = typeof endAnchor === "string" ? endAnchor : undefined;
+
+  const startIdx = src.indexOf(startAnchor, fromIdx);
+  if (startIdx < 0) return "";
+
+  if (endLiteral) {
+    const endIdx = src.indexOf(endLiteral, startIdx + startAnchor.length);
+    return endIdx > 0 ? src.slice(startIdx, endIdx) : src.slice(startIdx);
+  }
+
+  // Heuristic terminators — the next sibling declaration/section.
+  const terminators = [
+    "\n  private ",
+    "\n  public ",
+    "\n  protected ",
+    "\n  static ",
+    "\nexport function ",
+    "\nexport async function ",
+    "\nfunction ",
+    "\nasync function ",
+    "\nexport class ",
+    "\nclass ",
+    "\nexport interface ",
+    "\ninterface ",
+    "\n// ─", // section separator comments
+    "\n/** ", // next docblock
+  ];
+
+  let earliestEnd = -1;
+  for (const t of terminators) {
+    const idx = src.indexOf(t, startIdx + startAnchor.length);
+    if (idx > 0 && (earliestEnd < 0 || idx < earliestEnd)) earliestEnd = idx;
+  }
+
+  return earliestEnd > 0 ? src.slice(startIdx, earliestEnd) : src.slice(startIdx);
+}
+
+/**
+ * Poll `condition()` until it returns truthy or `timeoutMs` elapses.
+ * Returns the truthy value from `condition()`, or throws on timeout.
+ *
+ * Use this instead of `await new Promise(r => setTimeout(r, <magic-ms>))`
+ * when waiting for a state change that tests produce. The fixed-sleep
+ * pattern is flaky: too short → race; too long → slow tests.
+ *
+ * @param condition  Predicate. Returns truthy when done. May be async.
+ * @param opts.timeoutMs  Max total wait. Defaults to 2000ms.
+ * @param opts.intervalMs Poll interval. Defaults to 10ms.
+ * @param opts.description Included in the timeout error for debuggability.
+ */
+export async function waitForCondition<T>(
+  condition: () => T | Promise<T>,
+  opts: { timeoutMs?: number; intervalMs?: number; description?: string } = {},
+): Promise<T> {
+  const timeoutMs = opts.timeoutMs ?? 2000;
+  const intervalMs = opts.intervalMs ?? 10;
+  const deadline = Date.now() + timeoutMs;
+  let lastErr: unknown;
+  while (Date.now() < deadline) {
+    try {
+      const result = await condition();
+      if (result) return result;
+    } catch (e) {
+      lastErr = e;
+    }
+    await new Promise((r) => setTimeout(r, intervalMs));
+  }
+  const desc = opts.description ?? "condition";
+  const errSuffix = lastErr instanceof Error ? ` (last error: ${lastErr.message})` : "";
+  throw new Error(`waitForCondition timed out after ${timeoutMs}ms waiting for ${desc}${errSuffix}`);
+}
+
+/**
+ * Find the first line in rendered output that matches a predicate. Returns
+ * the line's index and text, or throws if no line matches.
+ *
+ * Use this instead of `lines[N]` indexing when the N is positional and
+ * could shift under formatting changes.
+ */
+export function findLine(
+  output: string,
+  predicate: RegExp | ((line: string) => boolean),
+): { index: number; text: string } {
+  const lines = output.split("\n");
+  const fn = predicate instanceof RegExp
+    ? (l: string) => {
+        // RegExp.test is stateful when the pattern has /g or /y flags
+        // (maintains lastIndex across calls). Reset before each test so
+        // matches on different lines don't silently skip.
+        if (predicate.global || predicate.sticky) predicate.lastIndex = 0;
+        return predicate.test(l);
+      }
+    : predicate;
+  for (let i = 0; i < lines.length; i++) {
+    if (fn(lines[i]!)) return { index: i, text: lines[i]! };
+  }
+  const preview = lines.slice(0, 10).join("\n");
+  throw new Error(
+    `findLine: no line matched. First 10 lines were:\n${preview}`,
+  );
+}

package/src/resources/extensions/gsd/tests/token-profile.test.ts

@@ -14,6 +14,7 @@ import assert from "node:assert/strict";
 import { readFileSync } from "node:fs";
 import { join, dirname } from "node:path";
 import { fileURLToPath } from "node:url";
+import { extractSourceRegion } from "./test-helpers.ts";
 
 const __dirname = dirname(fileURLToPath(import.meta.url));
 
@@ -127,7 +128,13 @@ test("profile: balanced profile skips research, reassess, and slice research (AD
 
 test("profile: quality profile skips research, slice research, and reassess (ADR-003)", () => {
   const qualityIdx = preferencesSrc.indexOf('case "quality":');
-  const qualityBlock = preferencesSrc.slice(qualityIdx, qualityIdx + 300);
+  // preferencesSrc is concatenated from multiple modules — bound the region
+  // with the next case marker so the assertion stays tightly scoped.
+  const qualityBlock = extractSourceRegion(
+    preferencesSrc,
+    'case "quality":',
+    'case "burn-max":',
+  );
   assert.ok(qualityBlock.includes("skip_research: true"), "quality should skip research");
   assert.ok(qualityBlock.includes("skip_slice_research: true"), "quality should skip slice research");
   assert.ok(qualityBlock.includes("skip_reassess: true"), "quality should skip reassess");

package/src/resources/extensions/gsd/tests/tool-invocation-error-loop-break.test.ts

@@ -44,7 +44,19 @@ describe("#2883: tool invocation error tracking on AutoSession", () => {
 
 // ─── isToolInvocationError classifier ────────────────────────────────────
 
-import { isToolInvocationError, isQueuedUserMessageSkip } from "../auto-tool-tracking.ts";
+import {
+  isDeterministicPolicyError,
+  isToolInvocationError,
+  isQueuedUserMessageSkip,
+} from "../auto-tool-tracking.ts";
+import {
+  shouldBlockContextWrite,
+  shouldBlockPendingGateInSnapshot,
+  shouldBlockQueueExecutionInSnapshot,
+  resetWriteGateState,
+  type WriteGateSnapshot,
+} from "../bootstrap/write-gate.ts";
+import { BLOCKED_WRITE_ERROR } from "../write-intercept.ts";
 
 describe("#2883: isToolInvocationError classification", () => {
   test("detects JSON validation failure pattern", () => {
@@ -89,11 +101,59 @@ describe("#2883: isToolInvocationError classification", () => {
     );
   });
 
+  test("detects raw write-gate CONTEXT failures for non-GSD write tools", () => {
+    resetWriteGateState();
+    const result = shouldBlockContextWrite(
+      "write",
+      "/tmp/project/.gsd/milestones/M001/M001-CONTEXT.md",
+      "M001",
+      false,
+    );
+
+    assert.equal(result.block, true);
+    assert.equal(isDeterministicPolicyError(result.reason ?? ""), true);
+    assert.equal(isToolInvocationError(result.reason ?? ""), true);
+  });
+
+  test("detects raw pending-gate failures for non-GSD write tools", () => {
+    const snapshot: WriteGateSnapshot = {
+      verifiedDepthMilestones: [],
+      activeQueuePhase: false,
+      pendingGateId: "depth_verification_M001",
+    };
+    const result = shouldBlockPendingGateInSnapshot(snapshot, "write", "M001", false);
+
+    assert.equal(result.block, true);
+    assert.equal(isDeterministicPolicyError(result.reason ?? ""), true);
+    assert.equal(isToolInvocationError(result.reason ?? ""), true);
+  });
+
+  test("detects queue-mode policy blocks for raw write/edit/bash tools", () => {
+    const snapshot: WriteGateSnapshot = {
+      verifiedDepthMilestones: [],
+      activeQueuePhase: true,
+      pendingGateId: null,
+    };
+    const writeResult = shouldBlockQueueExecutionInSnapshot(snapshot, "write", "src/app.ts", true);
+    const bashResult = shouldBlockQueueExecutionInSnapshot(snapshot, "bash", "npm run build", true);
+
+    assert.equal(writeResult.block, true);
+    assert.equal(bashResult.block, true);
+    assert.equal(isDeterministicPolicyError(writeResult.reason ?? ""), true);
+    assert.equal(isToolInvocationError(bashResult.reason ?? ""), true);
+  });
+
+  test("detects direct STATE.md/gsd.db write-intercept policy blocks", () => {
+    assert.equal(isDeterministicPolicyError(BLOCKED_WRITE_ERROR), true);
+    assert.equal(isToolInvocationError(BLOCKED_WRITE_ERROR), true);
+  });
+
   test("returns false for normal tool errors (business logic)", () => {
     assert.equal(
       isToolInvocationError("Slice S01 is already complete"),
       false,
     );
+    assert.equal(isDeterministicPolicyError("Slice S01 is already complete"), false);
   });
 
   test("returns false for empty string", () => {

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

@@ -36,6 +36,9 @@ const RENAME_MAP: Array<{ canonical: string; alias: string }> = [
   { canonical: "gsd_reassess_roadmap", alias: "gsd_roadmap_reassess" },
   { canonical: "gsd_complete_milestone", alias: "gsd_milestone_complete" },
   { canonical: "gsd_validate_milestone", alias: "gsd_milestone_validate" },
+  { canonical: "gsd_task_reopen", alias: "gsd_reopen_task" },
+  { canonical: "gsd_slice_reopen", alias: "gsd_reopen_slice" },
+  { canonical: "gsd_milestone_reopen", alias: "gsd_reopen_milestone" },
 ];
 
 // ─── Registration count ──────────────────────────────────────────────────────
@@ -45,7 +48,11 @@ console.log('\n── Tool naming: registration count ──');
 const pi = makeMockPi();
 registerDbTools(pi);
 
-assert.deepStrictEqual(pi.tools.length, 30, 'Should register exactly 30 tools (14 canonical + 14 aliases + 1 gate tool + 1 gsd_skip_slice)');
+assert.deepStrictEqual(
+  pi.tools.length,
+  RENAME_MAP.length * 2 + 2,
+  'Should register canonical/alias tool pairs plus 1 gate tool and 1 gsd_skip_slice',
+);
 
 // ─── Both names exist for each pair ──────────────────────────────────────────
 

package/src/resources/extensions/gsd/tests/triage-resolution.test.ts

@@ -258,6 +258,55 @@ test("resolution: markCaptureExecuted is idempotent", () => {
   }
 });
 
+// ─── executeTriageResolutions + note execution (#3578) ──────────────────────
+
+test("resolution: executeTriageResolutions stamps note captures as executed", () => {
+  const tmp = makeTempDir("res-exec-note");
+  try {
+    const id = appendCapture(tmp, "FYI the API changed");
+    markCaptureResolved(tmp, id, "note", "acknowledged", "informational");
+
+    const result = executeTriageResolutions(tmp, "M001", "S01");
+
+    // The note should now be marked as executed
+    const all = loadAllCaptures(tmp);
+    assert.strictEqual(all.length, 1);
+    assert.strictEqual(all[0].executed, true, "note capture should be marked as executed");
+
+    // It should appear in the actions log
+    assert.ok(
+      result.actions.some(a => a.includes(id) && a.includes("Note acknowledged")),
+      "actions should include a note-acknowledged entry",
+    );
+  } finally {
+    rmSync(tmp, { recursive: true, force: true });
+  }
+});
+
+test("resolution: executeTriageResolutions does not double-stamp already-executed notes", () => {
+  const tmp = makeTempDir("res-exec-note-idem");
+  try {
+    const id = appendCapture(tmp, "informational note");
+    markCaptureResolved(tmp, id, "note", "acknowledged", "info");
+
+    // First execution — stamps the note
+    executeTriageResolutions(tmp, "M001", "S01");
+
+    // Second execution — should be a no-op for the note
+    const result2 = executeTriageResolutions(tmp, "M001", "S01");
+
+    assert.strictEqual(result2.actions.length, 0, "second call should produce no actions");
+
+    // Verify the Executed field was not duplicated in the file
+    const filePath = join(tmp, ".gsd", "CAPTURES.md");
+    const content = readFileSync(filePath, "utf-8");
+    const executedMatches = content.match(/\*\*Executed:\*\*/g);
+    assert.strictEqual(executedMatches?.length, 1, "should have exactly one Executed field");
+  } finally {
+    rmSync(tmp, { recursive: true, force: true });
+  }
+});
+
 // ─── loadActionableCaptures ──────────────────────────────────────────────────
 
 test("resolution: loadActionableCaptures returns only unexecuted actionable captures", () => {
@@ -387,8 +436,7 @@ test("resolution: executeTriageResolutions handles mixed classifications", () =>
     assert.strictEqual(result.injected, 1, "should inject 1 task");
     assert.strictEqual(result.replanned, 0);
     assert.strictEqual(result.quickTasks.length, 1, "should queue 1 quick-task");
-    // inject + quick-task + note acknowledged = 3 actions (defer still excluded)
-    assert.strictEqual(result.actions.length, 3, "should have 3 action entries (defer excluded, note now included)");
+    assert.strictEqual(result.actions.length, 3, "should have 3 action entries (inject + quick-task + note acknowledged; defer excluded)");
   } finally {
     rmSync(tmp, { recursive: true, force: true });
   }

package/src/resources/extensions/gsd/tests/turn-epoch.test.ts

@@ -0,0 +1,162 @@
+// Project: gsd-pi — Tests for the auto/turn-epoch stale-write guard.
+
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import {
+  _resetTurnEpoch,
+  bumpTurnGeneration,
+  describeTurnEpoch,
+  getCurrentTurnGeneration,
+  isStaleWrite,
+  runWithTurnGeneration,
+} from "../auto/turn-epoch.ts";
+
+test("turn-epoch: generation starts at 0 and bumps monotonically", () => {
+  _resetTurnEpoch();
+  assert.equal(getCurrentTurnGeneration(), 0);
+  assert.equal(bumpTurnGeneration("test-a"), 1);
+  assert.equal(bumpTurnGeneration("test-b"), 2);
+  assert.equal(getCurrentTurnGeneration(), 2);
+});
+
+test("turn-epoch: isStaleWrite returns false when no turn context captured", () => {
+  _resetTurnEpoch();
+  bumpTurnGeneration("no-context");
+  // Called outside runWithTurnGeneration — safe default is false.
+  assert.equal(isStaleWrite("out-of-band"), false);
+});
+
+test("turn-epoch: isStaleWrite returns false inside a fresh turn", () => {
+  _resetTurnEpoch();
+  const captured = getCurrentTurnGeneration();
+  runWithTurnGeneration(captured, () => {
+    assert.equal(isStaleWrite("fresh"), false);
+  });
+});
+
+test("turn-epoch: isStaleWrite returns true after the epoch bumps mid-turn", () => {
+  _resetTurnEpoch();
+  const captured = getCurrentTurnGeneration();
+  runWithTurnGeneration(captured, () => {
+    bumpTurnGeneration("recovery-fires");
+    assert.equal(isStaleWrite("stale"), true);
+  });
+});
+
+test("turn-epoch: nested turns each see their own captured generation", () => {
+  _resetTurnEpoch();
+  const outerGen = getCurrentTurnGeneration();
+  runWithTurnGeneration(outerGen, () => {
+    assert.equal(isStaleWrite("outer-fresh"), false);
+    bumpTurnGeneration("bump-between");
+    const innerGen = getCurrentTurnGeneration();
+    runWithTurnGeneration(innerGen, () => {
+      // Inner context saw the bumped generation at capture time — fresh.
+      assert.equal(isStaleWrite("inner-fresh"), false);
+    });
+    // Back to outer context — still stale because outerGen < current.
+    assert.equal(isStaleWrite("outer-after-bump"), true);
+  });
+});
+
+test("turn-epoch: describeTurnEpoch surfaces captured vs current", () => {
+  _resetTurnEpoch();
+  bumpTurnGeneration("seed");
+  const captured = getCurrentTurnGeneration();
+  runWithTurnGeneration(captured, () => {
+    let snapshot = describeTurnEpoch();
+    assert.equal(snapshot.captured, captured);
+    assert.equal(snapshot.current, captured);
+    assert.equal(snapshot.stale, false);
+
+    bumpTurnGeneration("supersede");
+    snapshot = describeTurnEpoch();
+    assert.equal(snapshot.captured, captured);
+    assert.equal(snapshot.current, captured + 1);
+    assert.equal(snapshot.stale, true);
+  });
+
+  // Outside the turn — captured is null, stale is false.
+  const outside = describeTurnEpoch();
+  assert.equal(outside.captured, null);
+  assert.equal(outside.stale, false);
+});
+
+test("turn-epoch: AsyncLocalStorage propagates across awaits", async () => {
+  _resetTurnEpoch();
+  const captured = getCurrentTurnGeneration();
+  await runWithTurnGeneration(captured, async () => {
+    await Promise.resolve();
+    await new Promise((r) => setTimeout(r, 1));
+    bumpTurnGeneration("async-bump");
+    await Promise.resolve();
+    assert.equal(isStaleWrite("post-await"), true);
+  });
+});
+
+// ─── Source-level invariant checks for recoverTimedOutUnit ──────────────────
+//
+// The recoverTimedOutUnit function has two branch families:
+// - ADVANCE branches: the unit is done, loop moves on — these MUST bump.
+// - STEERING branches: the same LLM turn is kept alive with a steering
+//   message — these MUST NOT bump (otherwise the retry's legitimate writes
+//   get marked stale).
+//
+// The whole function must contain zero raw `resolveAgentEnd` calls with the
+// "timeout-recovery" _synthetic marker — all advance paths go through
+// bumpAndResolveSynthetic. And there must be no top-level bump call.
+
+test("recoverTimedOutUnit: no raw `resolveAgentEnd({ _synthetic: \"timeout-recovery\" })` calls remain", async () => {
+  const fs = await import("node:fs");
+  const path = await import("node:path");
+  const url = await import("node:url");
+  const here = path.dirname(url.fileURLToPath(import.meta.url));
+  const src = fs.readFileSync(
+    path.join(here, "..", "auto-timeout-recovery.ts"),
+    "utf-8",
+  );
+  const rawSyntheticResolve =
+    /resolveAgentEnd\s*\(\s*\{\s*messages:\s*\[\s*\]\s*,\s*_synthetic:\s*["']timeout-recovery/;
+  assert.equal(
+    rawSyntheticResolve.test(src),
+    false,
+    "auto-timeout-recovery.ts must funnel advance paths through bumpAndResolveSynthetic — a raw resolveAgentEnd with _synthetic:\"timeout-recovery\" would leak orphan writes",
+  );
+});
+
+test("recoverTimedOutUnit: no top-level bumpTurnGeneration — steering branches must not supersede", async () => {
+  const fs = await import("node:fs");
+  const path = await import("node:path");
+  const url = await import("node:url");
+  const here = path.dirname(url.fileURLToPath(import.meta.url));
+  const src = fs.readFileSync(
+    path.join(here, "..", "auto-timeout-recovery.ts"),
+    "utf-8",
+  );
+  // The only bump surface allowed is via bumpAndResolveSynthetic (advance
+  // paths) — a direct bumpTurnGeneration call in this file would bump even
+  // when the function later decides to take a steering retry branch.
+  assert.equal(
+    /\bbumpTurnGeneration\s*\(/.test(src),
+    false,
+    "auto-timeout-recovery.ts must not call bumpTurnGeneration directly — use bumpAndResolveSynthetic so bump and supersede are atomic and tied to advance-only branches",
+  );
+});
+
+test("recoverTimedOutUnit: bumpAndResolveSynthetic appears exactly four times (one per advance branch)", async () => {
+  const fs = await import("node:fs");
+  const path = await import("node:path");
+  const url = await import("node:url");
+  const here = path.dirname(url.fileURLToPath(import.meta.url));
+  const src = fs.readFileSync(
+    path.join(here, "..", "auto-timeout-recovery.ts"),
+    "utf-8",
+  );
+  const matches = src.match(/bumpAndResolveSynthetic\s*\(/g) ?? [];
+  assert.equal(
+    matches.length,
+    4,
+    `expected 4 advance-branch supersede sites (durableComplete, execute-task-exhausted, artifact-already-exists, non-execute-exhausted); found ${matches.length}`,
+  );
+});