@bongos/core 1.19.37 → 1.19.38

package/.bongos-core.json

@@ -2,22 +2,22 @@
   "artifact": "bongos-core",
   "manifest_schema": 1,
   "generator": "scripts/gds/package-core.js",
-  "core_version": "1.19.37",
-  "core_contract": "1.19.37",
-  "source_commit": "606cc770b09032c6ab385e1326f79b08c19e0083",
+  "core_version": "1.19.38",
+  "core_contract": "1.19.38",
+  "source_commit": "ba6a8f6285476ab3cd4e33f9278b333ac0fc3a29",
   "source_ref": "HEAD",
-  "built_at": "2026-08-11T16:06:38.933Z",
+  "built_at": "2026-08-11T16:17:20.571Z",
   "redaction": {
     "model": "docs-redacted+functional-verbatim",
-    "docs_redacted": 287,
+    "docs_redacted": 288,
     "agent_docs_stubbed": 21,
-    "functional_verbatim": 1211,
+    "functional_verbatim": 1212,
     "rules": 3,
     "gate_literals": 3,
     "gate": "passed"
   },
-  "file_count": 1519,
-  "tree_sha256": "faa022ab6fb5e04c311cd4d24d2991b07d15fb5fa565dffa0862d87fdf1e546d",
+  "file_count": 1521,
+  "tree_sha256": "9d55773209aeb85fef771e85aa30e901c52e54e155601030e44cd16fd5c46bda",
   "files": [
     {
       "path": ".claude/skills/blocker-review/SKILL.md",
@@ -159,6 +159,11 @@
       "mode": "0000644",
       "sha256": "e0bf90fa006499aee0a7d75f10fe4b5972c9501129b1e432e7d6e192b367a54c"
     },
+    {
+      "path": ".claude/skills/grader-health/SKILL.md",
+      "mode": "0000644",
+      "sha256": "b9823a6a14b4197f7dae90c8ad6713fcc5b3ce672d03d7b59c8e868d5d6ba70e"
+    },
     {
       "path": ".claude/skills/idea-triage/SKILL.md",
       "mode": "0000644",
@@ -1357,7 +1362,7 @@
     {
       "path": "docs/file-map.md",
       "mode": "0000644",
-      "sha256": "a499551f07ce8f6382aba46641fb06fa8e499d486433ee3bf266f264abac29ca"
+      "sha256": "f51d3978f50ec9b2a34dc12b8fa3b9e8874543e5ab3503f72a09d7d2b06c565b"
     },
     {
       "path": "docs/handoff-template.md",
@@ -4017,12 +4022,12 @@
     {
       "path": "package-lock.json",
       "mode": "0000644",
-      "sha256": "acab5dbbe8ab00da496f534457882de0c115d441e9f0e38fae6f5a901a4b3f47"
+      "sha256": "413f575fce07ba5b167f6492a4f1aed1f7eacb4f4351457303d3875f04557787"
     },
     {
       "path": "package.json",
       "mode": "0000644",
-      "sha256": "c33f5baf3eddd6b7bdcf80af532dc04d290e394b5c51b90931aa52d0ead94e6a"
+      "sha256": "d0d31ac616a949591bca2975236bbf6d98437e969d1618bd68811620c2a130c3"
     },
     {
       "path": "public-docs/index.html",
@@ -5642,7 +5647,7 @@
     {
       "path": "src/module-api.js",
       "mode": "0000644",
-      "sha256": "8066ee5e98e151dfc62228626a99a86f362697410aac483de27c75851cc3bdea"
+      "sha256": "2dfd2867ab83689497aff7a3b7f213e756c182e047ef8da754e098a945491ce7"
     },
     {
       "path": "src/module-loader/loader.js",
@@ -7464,6 +7469,11 @@
       "mode": "0000644",
       "sha256": "76bed8e6fae3bd77d4b45be14d9f80bda71b2f92323fa5697a7e4f07b65d3ba2"
     },
+    {
+      "path": "tests/skill_grader_health.mjs",
+      "mode": "0000644",
+      "sha256": "4557c934704a009612ed8858c0f59c5b53facb5d92fa2c0740b1aecb079aa5c9"
+    },
     {
       "path": "tests/skill_prefs.mjs",
       "mode": "0000644",

package/.claude/skills/grader-health/SKILL.md

@@ -0,0 +1,61 @@
+---
+name: grader-health
+description: Read-only health check over recent grades — consecutive outage runs, pass-rate drift vs baseline, per-grade cost anomalies, and stale model pins. The routine that would have caught the 21-outage 0.0-score streak on day one. Triggers when the user says "/grader-health", "check the grader", "is the grader healthy", "grader health check", "why are grades failing lately", or as part of a periodic ops sweep. Reports and emits exact remediation lines; changes nothing (ADR 0162 — no new gates, describe only). Optional --probe runs a live paid calibration (Metic+).
+---
+
+You are running a health check over the grading panel. Everything here is **read-only**: you read live surfaces, compare against known baselines, and emit findings + exact remediation commands. You never confirm, flip, or re-grade anything yourself — the builder (or /grade-recover) acts on what you report.
+
+## The four axes
+
+### 1 — Outage runs (the pattern this skill exists to catch)
+
+A grader outage used to be recorded as a 0.0 "quality failure": one builder accumulated **21 zero-score rows** over a month before anyone noticed, and the public tile showed a phantom 5-day 0%-pass crater. Since the outage-as-zero fix, aggregates exclude outage rounds and bucket them as `n_unavailable` — so the streak is now directly readable.
+
+1. Read the public trend: `node scripts/gds/api.js GET "/api/gds/public/grades?days=30"` — scan `trend[]` for days where the average is 0 (or null) with passes 0 and n ≥ 2, and for any nonzero `n_unavailable` streak.
+2. (Archon) Read the per-builder split: `node scripts/gds/api.js GET "/api/gds/grades/by-builder?days=30"` — a single builder with pass_rate 0.000 or a large `n_unavailable` bucket is the 21×0.0 shape localized to one machine or one environment.
+3. **Confirm outage vs rejection before alarming anyone.** For each task in a suspect window: `node scripts/gds/api.js GET "/api/gds/tasks/<id>?include=grade"` — `signals.panel_outcome='unavailable'` (with `grader_unavailable_reason` and per-worker errors in `signals.workers`) is an outage, **not a quality rejection**; a real FAIL carries findings in `issues[]`.
+4. For every task parked at `completed` by a confirmed outage, emit the exact remediation line, one per task:
+
+```bash
+node scripts/gds/ship.js <id> --regrade
+```
+
+5. Check the local spawn leg: the panel spawns `claude -p` subprocesses, so verify the CLI is present and authenticated, and run `node scripts/gds/doctor.js` for the hooks/worktree leg.
+
+### 2 — Pass-rate drift
+
+Baselines: the 14-agent evaluation window measured a **90.6%** pass rate, already above the 87% that ADR 0024 called "suspiciously high" and built the adversarial panel to fix. From `/api/gds/public/grades` compute the rolling pass rate and compare:
+
+- **Creeping up** past the baseline → grade inflation; the panel is rubber-stamping (report; candidate causes: score compression, reused grades, trivial-path overuse).
+- **Collapsing** far below it → either a real quality regression or an unconfirmed outage window — axis 1 disambiguates.
+
+### 3 — Cost anomalies
+
+Benchmark: mean panel cost **$2.14** per grade, and ~90% of it is **fixed** re-exploration cost, independent of diff size. Read grader spend: `node scripts/gds/api.js GET "/api/gds/public/cost-summary?source=grader&days=30"` and divide by the grade count from axis 2's window.
+
+- Mean drifting well above the benchmark → panel re-rolls/retries or timeout-and-retry loops (not "big diffs" — the fixed-cost share means diff size barely moves the mean).
+- Near-zero spend with grades still recording → the trivial path or grade reuse dominating; spot-check that gating lenses are actually running.
+
+### 4 — Model pins
+
+The evaluation found every grade ran on a generation-stale pin. Read the live pin from the repo: `DEFAULT_GRADER_MODEL` and `selectGraderModel` in `modules/grading/grader.js`, plus the rubric version in `modules/grading/grader-rubric.json`. Report when the pinned family is a generation behind the current Claude family, or when `selectGraderModel` would throw on a current builder model id (it throws on ids lacking opus/sonnet/haiku — a stranded-ship risk).
+
+## Optional: `--probe` (live calibration, spends real money)
+
+With `--probe`, run the live panel calibration: `node scripts/gds/grade-smoke.js`. It grades known-good and known-bad fixtures against the real panel and reports miscalibration. **Costs ~$0.30–0.90 per run; Metic+ only.** Never run it implicitly — only when the user passed `--probe` or asked for a live calibration.
+
+## Report shape
+
+End with a compact findings list, most severe first: axis, evidence (the numbers you read), verdict (outage / drift / anomaly / stale-pin / healthy), and the exact remediation command or the skill to hand off to (`/grade-recover` for a single stuck task; a GDS task via `capture.js` for systemic fixes). A clean bill of health is one line per axis.
+
+## Constraints
+
+- **Read-only, describe-only (ADR 0162).** No confirms, no regrades run by this skill, no status flips, no new gates. The remediation lines are FOR the builder to run.
+- **Confirm before alarming.** Axis 1's outage-vs-rejection check comes before any "the grader is down" conclusion — a real FAIL streak on one surface is signal, not outage.
+- **`--probe` spends money** — explicit opt-in only, Metic+.
+
+## Files this skill touches
+
+- Reads: `GET /api/gds/public/grades`, `GET /api/gds/grades/by-builder` (Archon), `GET /api/gds/public/cost-summary?source=grader`, `GET /api/gds/tasks/:id?include=grade`, `modules/grading/grader.js`, `modules/grading/grader-rubric.json`.
+- Runs (read-only): `node scripts/gds/doctor.js`; with `--probe` only: `node scripts/gds/grade-smoke.js`.
+- Writes: nothing.

package/docs/file-map.md

@@ -351,6 +351,7 @@ tests/
 ├── goal-create/SKILL.md           ← /goal-create — plan and create ONE goal (the module-scoped workspace between version and criteria, ADR 0086) in one session: <redacted>, scope wall, 2-4 criteria, seed tasks; also the auto-pickup when a task is created with no goal to live in (goal_advisory); Metic+ (Archon for a protected-module scope)
 ├── goal-review/SKILL.md           ← /goal-review — walk the criterion review queue: each auto-flagged 'met — pending review' criterion gets confirm / reject / defer; confirming a goal's last criterion auto-achieves it (ADR 0086 §6); Metic+
 ├── grade-recover/SKILL.md         ← /grade-recover N — the 'my grade failed, now what' loop: diagnose outage vs fabricated fail vs genuine fail from the persisted grade, then run the right recovery (--regrade or the auditable override-request); adds no ship-path gate (ADR 0162, task 1002655)
+├── grader-health/SKILL.md         ← /grader-health — read-only grader health check: outage runs (n_unavailable streaks, outage-vs-rejection confirmation), pass-rate drift vs baseline, per-grade cost anomalies, stale model pins; emits exact --regrade remediation lines; optional --probe live calibration (Metic+, paid) (task 1002656)
 ├── idea-triage/SKILL.md           ← /idea-triage — daily walk through the GDS idea_inbox (promote / discard / merge); Metic+
 ├── ideate/SKILL.md                ← /ideate — ideation-session playbook for ideator-discipline tasks; inverts the loop (human drives, Claude is the sounding board) ([#1267](https://example.com/builders#/task/1267))
 ├── merge-mode/SKILL.md            ← /merge-mode — manual fallback for the auto-merge: land every confirmed task (merge + smoke + deploy)

package/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@bongos/core",
-  "version": "1.19.37",
+  "version": "1.19.38",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@bongos/core",
-      "version": "1.19.37",
+      "version": "1.19.38",
       "license": "AGPL-3.0-or-later",
       "dependencies": {
         "express": "^4.21.2",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bongos/core",
-  "version": "1.19.37",
+  "version": "1.19.38",
   "description": "Cloud Bongos — the AI-first build platform core (GDS + platform surfaces + module system), installed as a versioned dependency (ADR 0108).",
   "license": "AGPL-3.0-or-later",
   "main": "src/platform-server.js",

package/src/module-api.js

@@ -501,8 +501,10 @@ const { validateOrRespond, LIMITS, parseId, asyncHandler, corsPublicGet, parsePa
 //         landed since 1.19.35 with no explicit bump. run 31509715311. (task 1002620)
 //   1.19.37 — CI auto-patch (publish-on-merge, ADR 0161): carrier for merges
 //         landed since 1.19.36 with no explicit bump. run 31510590459. (task 1002620)
+//   1.19.38 — CI auto-patch (publish-on-merge, ADR 0161): carrier for merges
+//         landed since 1.19.37 with no explicit bump. run 31511537558. (task 1002620)
 // ---------------------------------------------------------------------------
-const CORE_VERSION = '1.19.37'; // CI auto-patch carrier (ADR 0161); see changelog above
+const CORE_VERSION = '1.19.38'; // CI auto-patch carrier (ADR 0161); see changelog above
 
 // A namespaced logger so a module's log lines are attributable + consistent.
 // Usage: const log = api.logger('dev-box'); log.info('mounted');

package/tests/skill_grader_health.mjs

@@ -0,0 +1,48 @@
+// tests/skill_grader_health.mjs — the /grader-health skill exists and covers
+// the four health axes from the grader-hardening evaluation §6 (task 1002656):
+// outage runs, pass-rate drift, cost anomalies, and stale model pins — reading
+// LIVE data surfaces, emitting exact --regrade remediation lines, and staying
+// read-only (describe, never confirm/flip — ADR 0162).
+import assert from 'node:assert/strict';
+import { test } from 'node:test';
+import { readFileSync, existsSync } from 'node:fs';
+import { fileURLToPath } from 'node:url';
+import path from 'node:path';
+
+const ROOT = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '..');
+const SKILL = path.join(ROOT, '.claude', 'skills', 'grader-health', 'SKILL.md');
+
+// .claude/skills is INSTANCE-materialized; a neutral core checkout carries none
+// (task 1002470) — skip there, same guard shape as file_map.mjs.
+const skillsDirExists = existsSync(path.join(ROOT, '.claude', 'skills'));
+
+test('/grader-health exists and covers the four health axes on live data', (t) => {
+  if (!skillsDirExists) return t.skip('.claude/skills not materialized in this checkout');
+  assert.ok(existsSync(SKILL), '.claude/skills/grader-health/SKILL.md must exist (task 1002656)');
+  const md = readFileSync(SKILL, 'utf8');
+
+  // Live data surfaces, named concretely.
+  assert.match(md, /\/api\/gds\/public\/grades/, 'reads the public grade trend');
+  assert.match(md, /\/api\/gds\/grades\/by-builder/, 'reads the per-builder split (Archon)');
+  assert.match(md, /\/api\/gds\/public\/cost-summary/, 'reads grader spend');
+
+  // Axis 1 — outage runs: the 21×0.0 pattern and outage-vs-rejection confirmation.
+  assert.match(md, /n_unavailable/, 'names the outage bucket the aggregates expose');
+  assert.match(md, /panel_outcome='unavailable'/, 'confirms outage vs rejection per task');
+
+  // Axis 2 — pass-rate drift vs the evaluation baseline.
+  assert.match(md, /90\.6%/, 'names the eval-window pass-rate baseline');
+
+  // Axis 3 — cost anomalies vs the fixed-cost benchmark.
+  assert.match(md, /\$2\.14/, 'names the mean per-grade cost benchmark');
+
+  // Axis 4 — model pins.
+  assert.match(md, /DEFAULT_GRADER_MODEL|selectGraderModel/, 'checks the live model pin');
+
+  // Remediation + probe + the read-only contract.
+  assert.match(md, /node scripts\/gds\/ship\.js <(?:task-)?id> --regrade/, 'emits the exact remediation line');
+  assert.match(md, /--probe/, 'offers the optional live calibration probe');
+  assert.match(md, /grade-smoke\.js/, 'names the probe script');
+  assert.match(md, /read-only/i, 'states the read-only contract');
+  assert.match(md, /ADR 0162/, 'anchors the no-new-gates constraint');
+});