session-orchestrator 5.7.0 → 5.9.0

package/skills/claude-md-drift-check/checker.mjs

@@ -38,7 +38,7 @@ import { resolveInstructionFile } from '../../scripts/lib/common.mjs';
 import { _parseVaultIntegration } from '../../scripts/lib/config/vault-integration.mjs';
 import { _parseDriftCheck } from '../../scripts/lib/config/drift-check.mjs';
 import { isSessionConfigHeading } from '../../scripts/lib/config/section-extractor.mjs';
-import { FRONTMATTER_NOT_AT_TOP, parseGlobsFrontmatter } from '../../scripts/lib/rule-loader.mjs';
+import { FRONTMATTER_NOT_AT_TOP, isRuleExpired, parseGlobsFrontmatter } from '../../scripts/lib/rule-loader.mjs';
 import { resolveRepoSpec } from '../../scripts/lib/vcs-repo-spec.mjs';
 import { userInvocableSkills } from '../../scripts/lib/user-invocable-skills.mjs';
 import { learningKeyOf } from '../../scripts/lib/learnings/kebab.mjs';
@@ -337,6 +337,44 @@ function instructionFilesAreAliased(pathA, pathB) {
   }
 }
 
+/**
+ * Is `AGENTS.md` a pure import pointer to `CLAUDE.md`?
+ *
+ * Shape: its first non-empty line is a Claude Code import `@CLAUDE.md` (or
+ * `@./CLAUDE.md`), and it carries no config of its own — no
+ * `vault-integration:` block and no `## Session Config` heading. Such a file
+ * cannot disagree with CLAUDE.md at runtime: `resolveInstructionFile()`
+ * (`scripts/lib/common.mjs`) and `parse-config.mjs` read CLAUDE.md first
+ * whenever it is non-empty, so the pointer's missing block is never read.
+ * Comparing them reported `CLAUDE.md='<dir>' vs AGENTS.md='(unset)'` as an
+ * error (peer session gotzendorferv2-6e, 2026-10-04).
+ *
+ * ONE direction only, on purpose. A CLAUDE.md that imports AGENTS.md is the
+ * file those readers pick, and they read it raw — they do not follow `@`
+ * imports — so its missing block IS the runtime value and the disagreement is
+ * real. A pointer that carries its own block stays compared too.
+ *
+ * Check 7 only. Check 9 probe 2a must keep scanning such a file: its citations
+ * are its OWN text (the import is not expanded here), so scanning it reports
+ * no CLAUDE.md defect twice — skipping it would hide its own dangling ones.
+ *
+ * @param {string} agentsPath
+ * @returns {boolean}
+ */
+function agentsMdImportsClaudeMd(agentsPath) {
+  let content;
+  try {
+    content = readFileSync(agentsPath, 'utf8').replace(/^\uFEFF/, '');
+  } catch {
+    return false;
+  }
+  const lines = content.split('\n');
+  const first = lines.find((l) => l.trim().length > 0);
+  if (first === undefined || !/^@(?:\.\/)?CLAUDE\.md\s*$/.test(first.trim())) return false;
+  if (lines.some((l) => isSessionConfigHeading(l))) return false;
+  return !readVaultIntegration(agentsPath).present;
+}
+
 function classifySection(heading) {
   if (!heading) return null;
   if (BACKWARD_HEADING_RE.test(heading)) return 'backward';
@@ -417,8 +455,9 @@ function countTestFiles(vaultDir) {
 }
 
 /**
- * The product's slash-command count: `commands/*.md` ∪ skills carrying an
- * explicit `user-invocable: true`, deduplicated by name — the same union
+ * The product's slash-command count: `commands/*.md` ∪ skills whose
+ * `user-invocable` is `true` or absent (Claude Code picker semantics, #1515),
+ * deduplicated by name — the same union
  * `scripts/site-numbers.mjs` and
  * `tests/commands/headless-bare-command-availability.test.mjs` use. In a
  * consumer repo without `skills/` (or without `commands/`) the union
@@ -669,12 +708,31 @@ function globMatchesAny(pattern, trackedFiles) {
  * Returns the repo's tracked files (relative, forward-slash separated) via
  * `git ls-files`; falls back to a manual walk (excluding dotdirs and
  * node_modules — same exclusions as `walkDir`) when git is unavailable.
+ *
+ * Only "git missing" and "not a git repository" fall back silently. Any other
+ * failure still falls back, but says so on stderr: the walk skips dotdirs, so
+ * every glob under `.claude/` or a dotfile like `.gitleaks.toml` then reports
+ * "matches 0 tracked files" — a false warning that must carry its cause. The
+ * 64 MiB buffer exists because the 1 MiB default threw ENOBUFS on a vault
+ * listing of 1,199,500 bytes (measured 2026-10-04), which the old bare catch
+ * turned into exactly those false warnings, silently.
  */
 function listTrackedFiles(vaultDir) {
   try {
-    const out = execFileSync('git', ['ls-files'], { cwd: vaultDir, encoding: 'utf8' });
+    const out = execFileSync('git', ['ls-files'], {
+      cwd: vaultDir,
+      encoding: 'utf8',
+      stdio: ['ignore', 'pipe', 'pipe'],
+      maxBuffer: 64 * 1024 * 1024,
+    });
     return out.split('\n').filter(Boolean);
-  } catch {
+  } catch (err) {
+    const notARepo = err?.status === 128 && /not a git repository/i.test(String(err?.stderr ?? ''));
+    if (err?.code !== 'ENOENT' && !notARepo) {
+      process.stderr.write(
+        `claude-md-drift-check: git ls-files failed in ${vaultDir} (${err?.code ?? `exit ${err?.status}`}: ${String(err?.message ?? err).split('\n')[0]}) — falling back to a filesystem walk that skips dotdirs; zero-match glob warnings under dot-paths are unreliable this run\n`,
+      );
+    }
     const files = [];
     walkDir(vaultDir, (f) => files.push(relative(vaultDir, f).replace(/\\/g, '/')));
     return files;
@@ -1113,8 +1171,13 @@ function main() {
         // absent vault-integration block is a different (and more informative)
         // outcome than "the files cannot disagree".
         checksSkipped.push('vault-dir-parity: neither file has a vault-integration: block');
-      } else if (instructionFilesAreAliased(claudePath, agentsPath)) {
-        // Identical by construction (generated copy, symlink, or hardlink) —
+      } else if (
+        instructionFilesAreAliased(claudePath, agentsPath) ||
+        agentsMdImportsClaudeMd(agentsPath)
+      ) {
+        // Identical by construction (generated copy, symlink, or hardlink), or
+        // AGENTS.md is a config-free `@CLAUDE.md` import pointer the runtime
+        // never reads (see agentsMdImportsClaudeMd) —
         // parity is SATISFIED, so the check RAN and found nothing. Reporting
         // this as a skip would be wrong too: the invariant Check 7 guards is
         // actively held here, it is simply held mechanically rather than by
@@ -1178,10 +1241,19 @@ function main() {
       const learningKeyM = /^learning-key:\s*(.+)$/m.exec(block);
       const expiresAtM = /^expires-at:\s*(.+)$/m.exec(block);
       const evidenceDigestM = /^evidence-digest:\s*(.+)$/m.exec(block);
+      // `expires-at` is read the loader's way (quote-stripped), so a hand-quoted
+      // `"2026-06-01"` is not handed to V8's legacy local-time parser.
+      let expiresAt = expiresAtM ? expiresAtM[1].trim() : null;
+      try {
+        const loaderValue = parseGlobsFrontmatter(mdContent).meta['expires-at'];
+        if (typeof loaderValue === 'string') expiresAt = loaderValue;
+      } catch {
+        // malformed frontmatter for the loader — keep the raw value (fail-open below)
+      }
       return {
         autoGenerated: autoGenM ? autoGenM[1].trim() === 'true' : false,
         learningKey: learningKeyM ? learningKeyM[1].trim() : null,
-        expiresAt: expiresAtM ? expiresAtM[1].trim() : null,
+        expiresAt,
         evidenceDigest: evidenceDigestM ? evidenceDigestM[1].trim() : null,
       };
     }
@@ -1281,9 +1353,10 @@ function main() {
       // stored expires_at from the learning entry (both should agree, but the
       // rule file is authoritative for its own expiry).
       const expiryStr = expiresAt ?? storedExpiresAt;
+      // Same predicate as the loader (#1521): a bare date is INCLUSIVE, an
+      // unparseable value fails open.
       if (expiryStr) {
-        const expiryMs = Date.parse(expiryStr);
-        if (Number.isFinite(expiryMs) && expiryMs < nowMs) {
+        if (isRuleExpired(expiryStr, nowMs)) {
           warnings.push({
             check: 'generated-rule-staleness',
             file: relPath,

package/skills/discovery/probes/ssot-code-diff.mjs

@@ -24,7 +24,7 @@
  *   2. `.claude/rules/*.md` file count (non-recursive) vs. "N rule files" claims.
  *   3. `skills/` user-facing directory count (excludes `_shared/`) vs.
  *      "N user-facing skills" / "N skills" claims.
- *   4. Slash-command count (`commands/*.md` ∪ skills with `user-invocable: true`)
+ *   4. Slash-command count (`commands/*.md` ∪ skills not marked `user-invocable: false`)
  *      vs. "N slash-commands" / "N commands" claims.
  *
  * Scope exclusions: CHANGELOG.md, docs/adr/, docs/prd/, docs/retro/ are never
@@ -103,8 +103,8 @@ function countUserFacingSkills(root) {
   }
 }
 
-// Slash commands = `commands/*.md` ∪ skills with an explicit
-// `user-invocable: true` (the #1370 commands→skills fold) — the shared census
+// Slash commands = `commands/*.md` ∪ skills whose `user-invocable` is absent or
+// `true` (the #1370 commands→skills fold; absent = true per #1515) — the shared census
 // in user-invocable-skills.mjs, not a second copy. Counting `commands/` alone
 // reported 2 against the real 27 on every run (#1506 pt 7). `null` (entry
 // skipped) only when the repo has neither source.

package/skills/eval/SKILL.md

@@ -15,7 +15,7 @@ args-schema:
   - flag: --verify
     description: "Re-evaluate a stored run-id and diff per-dimension for scoring drift (exit 1 on drift)"
 description: >
-  Use this skill to run an honest session-process evaluation (Standard v1, aiat-llm-eval/1.0) — score the last completed orchestrator session against the pre-registered rubric-v2 dimensions, run /eval, evaluate this session, produce an eval report, or re-verify a stored eval run for reproducibility. Deterministic-first with an optional advisory LLM judge; never produces a global score.
+  Use this skill to run an honest session-process evaluation (Standard v1, aiat-llm-eval/1.0) — score the last completed orchestrator session against the pre-registered rubric-v3 dimensions, run /eval, evaluate this session, produce an eval report, or re-verify a stored eval run for reproducibility. Deterministic-first with an optional advisory LLM judge; never produces a global score.
 ---
 
 > **Platform Note:** State files use the platform's native directory: `.claude/` (Claude Code), `.codex/` (Codex CLI), or `.cursor/` (Cursor IDE). Shared metrics + the eval journal live in `.orchestrator/metrics/`. See `skills/_shared/platform-tools.md`.
@@ -23,7 +23,7 @@ description: >
 # Eval Skill — Session-Process Evaluation (aiat-llm-eval/1.0)
 
 On-demand, honest measurement of ONE completed orchestrator session against the
-pre-registered **rubric-v2** check set. The deterministic engine
+pre-registered **rubric-v3** check set. The deterministic engine
 (`scripts/eval-session.mjs` → `scripts/lib/eval/engine.mjs`) reads only local
 metrics files (`sessions.jsonl` + `events.jsonl`), scores the six deterministic
 dimensions, appends a `session-eval` record to the journal, and optionally
@@ -31,9 +31,10 @@ renders an HTML report. An opt-in LLM judge overlays ONE advisory dimension
 (`instruction-adherence`; `report-quality` was retired in rubric-v2, #1381).
 
 The standard this skill implements is [`docs/eval/aiat-llm-eval-v1.md`](../../docs/eval/aiat-llm-eval-v1.md);
-the frozen, content-hashed check set is [`skills/eval/rubric-v2.md`](./rubric-v2.md)
-(stored records written before 2026-09-19 carry `rubric-v1` and are read against
-[`rubric-v1.md`](./rubric-v1.md), which is never edited again).
+the frozen, content-hashed check set is [`skills/eval/rubric-v3.md`](./rubric-v3.md)
+(stored records carry the version they were scored under: before 2026-09-19
+`rubric-v1` → [`rubric-v1.md`](./rubric-v1.md), 2026-09-19 to 2026-10-05
+`rubric-v2` → [`rubric-v2.md`](./rubric-v2.md); neither is edited again).
 
 ## Invocation
 
@@ -46,7 +47,7 @@ Invoked as `/eval [--session <id>] [--no-write] [--verify <run-id>]` with argume
 
 **On-demand `/eval` runs regardless of `eval.enabled`** — that flag gates only the automatic session-end eval phase (see Phase 1.1).
 
-**Seams used:** `scripts/eval-session.mjs` (deterministic CLI) · `runEvalJudge` / `mergeJudgeDimensions` (`scripts/lib/eval/judge.mjs`, opt-in) · `writeEvalReport` (`scripts/lib/eval/report.mjs`) · `appendEvalRecord` (`scripts/lib/eval/sink.mjs`) · the `eval` config block · [`skills/eval/rubric-v2.md`](./rubric-v2.md) (frozen check set).
+**Seams used:** `scripts/eval-session.mjs` (deterministic CLI) · `runEvalJudge` / `mergeJudgeDimensions` (`scripts/lib/eval/judge.mjs`, opt-in) · `writeEvalReport` (`scripts/lib/eval/report.mjs`) · `appendEvalRecord` (`scripts/lib/eval/sink.mjs`) · the `eval` config block · [`skills/eval/rubric-v3.md`](./rubric-v3.md) (frozen check set).
 
 ## Posture Contract (load-bearing — read before executing)
 
@@ -154,7 +155,7 @@ node scripts/eval-session.mjs [--session <id>] --json \
 Parse the emitted JSON record. It carries `dimensions[]` (6 deterministic
 entries — the two reported-only ones, `guard-friction` and `efficiency-kpis`,
 are always `not-applicable`), `kpis{}`, `provenance.rubric_sha256` (the sha256
-of `rubric-v2.md`; `null` means the rubric file was not found and the append
+of `rubric-v3.md`; `null` means the rubric file was not found and the append
 will fail validation), `model`, `harness`, and `run_id`. Unless `--no-write` was
 passed, the record is already appended to `.orchestrator/metrics/eval.jsonl` by
 the CLI.
@@ -236,7 +237,7 @@ const res = writeEvalReport(record, { generatedAt: new Date().toISOString() });
 Emit a compact, honest per-dimension summary. Status lines only — no global score.
 
 ```
-## /eval — <session_id>  (self-evaluation, aiat-llm-eval/1.0 · rubric-v2 · n=1, no CI)
+## /eval — <session_id>  (self-evaluation, aiat-llm-eval/1.0 · rubric-v3 · n=1, no CI)
 
 Deterministic:
   verification-evidence   PASS   <one-line evidence>
@@ -276,11 +277,9 @@ node scripts/eval-session.mjs --verify <run-id> --json
   changed since the record was written — investigate, do not overwrite.
 - `--verify` reproduces the stored model + timestamp verbatim (no env override),
   so a MATCH is a real reproducibility proof of the scoring, not of model output.
-- **A cross-version DRIFT is not a defect.** A stored `rubric-v1` record
-  re-scored by today's rubric-v2 engine necessarily differs on `process-safety`
-  and reports `present-in-fresh-only: guard-friction`. Read the record's
-  `rubric_version` before treating a diff as a regression (#1400 replaces that
-  report with an explicit version verdict).
+- **A record from another rubric version is not re-scored.** `--verify` reports
+  `VERSION-MISMATCH` (exit 3) when the stored `rubric_version` differs from the
+  engine's (#1400), so a v1 or v2 record never reads as a regression under v3.
 
 ---