@claude-flow/cli 3.38.12 → 3.38.13

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783608038317_jrd66e_result.log

@@ -0,0 +1,92 @@
+[2026-07-09T14:43:08.375Z] RESULT
+============================================================
+{
+  "success": true,
+  "output": "## Test Coverage Analysis — `plugins/ruflo-metaharness`\n\n**Infrastructure note:** there's no `package.json`/jest/vitest here. The \"test suite\" is `scripts/test-*.mjs` (6 files) using a hand-rolled `assert()` + pass/fail counter, gated by `scripts/smoke.sh`. All skeletons below follow that existing convention rather than introducing a new framework — nothing here is exported as a library, so most gaps require black-box `execFileSync` tests, not unit imports.\n\nI verified every claim below directly against source (this plugin already had an auto-generated `testgaps` report from today in `.claude-flow/metrics/test-gaps.json` — I cross-checked it rather than trusting it blindly, and corrected one inaccuracy, noted at #5).\n\n### 1. `threat-model.mjs` — stale local `SEVERITY_RANK` (real bug, highest priority)\n\n`scripts/threat-model.mjs:9` keeps its own `{clean:0,low:1,medium:2,high:3}` instead of importing the canonical table from `_harness.mjs:261` (`{clean,info,low,medium,warn,high,error,critical}`). If the harness reports `worst:\"critical\"`, `SEVERITY_RANK['critical']` is `undefined`, so `undefined >= threshold` → `false` — `--fail-on high` **never trips for a critical finding**.\n\n```js\n// scripts/test-threat-model-severity.mjs\nimport { execFileSync } from 'node:child_process';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\n// Requires a mock/fixture harness binary that returns worst:\"critical\"\nconst out = execFileSync('node', ['scripts/threat-model.mjs', '--path', 'FIXTURE_CRITICAL', '--fail-on', 'high'], { encoding: 'utf8' });\nconst payload = JSON.parse(out);\nassert(payload.alert.triggered === true, 'worst=critical must trigger --fail-on high (currently silently false)');\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\nFix: `import { SEVERITY_RANK } from './_harness.mjs'` instead of the local literal.\n\n### 2. `router-parallel-analyze.mjs` — one malformed JSONL line kills the whole run\n\n`scripts/router-parallel-analyze.mjs:90-95`: the `.map(l => JSON.parse(l))` has no per-line try/catch; the *outer* catch aborts with exit 2 for the entire file on a single bad line, discarding all valid rows.\n\n```js\n// scripts/test-router-parallel-malformed.mjs\nimport { execFileSync } from 'node:child_process';\nimport { writeFileSync, mkdtempSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { tmpdir } from 'node:os';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nconst dir = mkdtempSync(join(tmpdir(), 'rpa-'));\nconst file = join(dir, 'router-parallel.jsonl');\nconst goodRow = JSON.stringify({ bandit: {}, ser: {}, outcome: {} }); // must match the real filter shape (bandit && ser && outcome)\nwriteFileSync(file, Array(35).fill(goodRow).join('\\n') + '\\nNOT VALID JSON\\n');\n\nlet exitCode = 0;\ntry {\n  execFileSync('node', ['scripts/router-parallel-analyze.mjs', '--input', file], { encoding: 'utf8' });\n} catch (e) { exitCode = e.status; }\n\nassert(exitCode !== 2, 'one malformed JSONL line should not abort analysis of the other 35 valid rows');\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\nFix: wrap the per-line parse individually, skip+count bad lines instead of aborting.\n\n### 3. `evolve.mjs --diagnose` subsystem — zero coverage\n\n`looksLikeGepaTranscript`, `extractGepaTranscripts`, `summarizeTraces`, `buildDiagnosis` (evolve.mjs:146-239, ~90 lines) are module-private (no `export`) and never exercised by any of the 6 existing test scripts. Must be tested black-box via CLI.\n\n```js\n// scripts/test-evolve-diagnose.mjs\nimport { execFileSync } from 'node:child_process';\nimport { mkdtempSync, writeFileSync, mkdirSync } from 'node:fs';\nimport { tmpdir } from 'node:os';\nimport { join } from 'node:path';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nconst runsDir = mkdtempSync(join(tmpdir(), 'evolve-diag-'));\nmkdirSync(join(runsDir, '.metaharness', 'runs'), { recursive: true });\nwriteFileSync(join(runsDir, '.metaharness', 'runs', 'winner.json'), '{\"id\":\"v1\"}');\nwriteFileSync(join(runsDir, '.metaharness', 'runs', 'v1.json'), JSON.stringify({ id: 'v1', transcript: [] }));\n\nconst out = execFileSync('node', ['scripts/evolve.mjs', runsDir, '--diagnose', '--sandbox', 'mock'], { encoding: 'utf8' });\nassert(typeof JSON.parse(out).diagnosis === 'object', 'diagnosis object present for a valid winner+run record');\n\n// corrupt winner.json must degrade, not throw\nwriteFileSync(join(runsDir, '.metaharness', 'runs', 'winner.json'), '{not json');\nconst out2 = execFileSync('node', ['scripts/evolve.mjs', runsDir, '--diagnose', '--sandbox', 'mock'], { encoding: 'utf8' });\nassert(!!JSON.parse(out2).diagnosis, 'corrupt winner.json does not crash --diagnose');\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\n\n### 4. `gepa.mjs` — `validate`/`render`/`analyze` ops untested\n\nConfirmed all 4 ops (`genome|validate|render|analyze`, gepa.mjs:104) exist but only `genome` is exercised indirectly via MCP. `--alert-on-invalid` exit-1 path and the non-array `--transcript` exit-2 path (gepa.mjs:135-136) are unhit.\n\n```js\n// scripts/test-gepa-ops.mjs\nimport { execFileSync } from 'node:child_process';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nfor (const op of ['validate', 'render']) {\n  const out = execFileSync('node', ['scripts/gepa.mjs', '--op', op], { encoding: 'utf8' });\n  const payload = JSON.parse(out);\n  assert(payload.op === op || payload.degraded, `op=${op} returns a shaped payload or degrades cleanly`);\n}\n\ntry {\n  execFileSync('node', ['scripts/gepa.mjs', '--op', 'analyze', '--transcript', '{\"not\":\"array\"}'], { encoding: 'utf8' });\n  assert(false, 'non-array --transcript should exit non-zero');\n} catch (e) { assert(e.status === 2, 'non-array --transcript exits 2'); }\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\n\n### 5. `mcp-scan.mjs` — string-shaped finding produces silent corruption, **not a crash** (correction)\n\nThe auto-generated report claimed `findings.map` \"throws\" on a raw-string array element. I checked the map body (`mcp-scan.mjs:62-66`): `{...f, message: ...}` on a string `f` spreads its characters into numeric keys and `f.message`/`f.title`/`f.detail` are all `undefined` on a string — so `message` silently falls back to `''`, no throw. The real bug is **silent data corruption** (a finding becomes `{0:'a',1:' ',...,message:''}`), which is worse for debugging than a crash since it never surfaces.\n\n```js\n// scripts/test-mcp-scan-malformed-finding.mjs\nimport { parseMcpScanText } from './_harness.mjs'; // if exported; else drive via fixture stdout + execFileSync\n// Fixture: upstream JSON with findings: [\"a raw string\", {severity:'high', message:'ok'}]\n// Assert: string elements are filtered out or coerced to a real object BEFORE spread,\n// not silently spread into {0:'a',1:' ',...}.\n```\nRecommend a guard: `rawFindings.filter(f => f && typeof f === 'object')` before the `.map`.\n\n### 6. `security-bench.mjs` — `parseSecurityBenchMarkdown` not exported, no isolated test\n\nConfirmed not exported (`security-bench.mjs:77`). Fragile to upstream markdown/emoji/table drift.\n\n```js\n// scripts/test-security-bench-parse.mjs\n// Requires: export function parseSecurityBenchMarkdown(md) { ... }\nimport { parseSecurityBenchMarkdown } from './security-bench.mjs';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nconst wellFormed = `## Overall: ✅ PASS\\n| Gate | Status |\\n|---|---|\\n| injection | ✅ |\\n`;\nassert(parseSecurityBenchMarkdown(wellFormed).gates.length > 0, 'well-formed markdown parses gates');\n\nconst malformed = `unexpected upstream format with no matching tables`;\nconst r2 = parseSecurityBenchMarkdown(malformed);\nassert(r2.gates.length === 0, 'malformed markdown returns empty gates, not throw');\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\n\n### 7. Missing error-handling tests (quick wins, all confirmed against source)\n\n| File:line | Confirmed issue | Test scenario |\n|---|---|---|\n| `audit-trend.mjs:140` | `fingerprint()` truncates `message` at 80 chars via `.slice(0,80)` | Two findings identical in the first 80 chars but differing after — assert they are NOT deduped as \"unchanged\" |\n| `similarity.mjs:51,114` | `Number(process.argv[++i])` on `--alert-below abc` → `NaN`; `result.overall < NaN` is always `false` | `execFileSync(['similarity.mjs',...,'--alert-below','abc'])` should exit non-zero (reject bad arg), not silently succeed |\n| `redblue.mjs:90` | `!next.startsWith('-')` guard means `--some-flag -5` drops `-5` as the value (mistaken for a bare flag) | Passthrough arg with a negative-number value must be preserved, not swallowed |\n| `learn.mjs:92` | `checkout-required` detection is a fixed status string keyed off upstream stderr wording | Fixture stderr with slightly different wording — assert it doesn't silently misreport `status:'failed'` |\n| `_harness.mjs:269` | `rankSeverity` lowercases but doesn't trim — `rankSeverity(' high ')` → 0, not 3 | `assert(rankSeverity(' high ') === rankSeverity('high'))` — currently fails, flags the footgun |\n\n### 8. Integration test gaps\n\n- **`oia-audit.mjs`**: only the \"all 5 subprocesses degraded\" path is tested (`test-graceful-degradation.mjs`). No test for 1–4 of 5 degrading while the composite still reports healthy at the top level.\n- **`mint.mjs`**: the documented pre-0.1.13 `--target`-ignored workaround has no regression test against the currently pinned metaharness version — if upstream fixes it, the workaround's continued correctness is unverified either way.\n- **`test-parallel-pipeline.mjs`**: its TS-source check is a `grep`, not a runtime test — `recordPair`/`recordPairOutcome` in `router-parallel-recorder.ts` are pattern-matched as text, never actually executed against compiled `dist/`.\n- **`test-pipeline-roundtrip.mjs`**: `REPO_ROOT = dirname(dirname(dirname(SCRIPTS_DIR)))` is a fragile path assumption — relocating the plugin silently points at the wrong tree instead of failing loudly. Add a sentinel-file assertion (e.g., root `CLAUDE.md` exists) before proceeding.\n\n### Priority order if picking a subset\n\n1. Fix `threat-model.mjs` `SEVERITY_RANK` (#1) — real bug, silently swallows critical findings\n2. Fix `router-parallel-analyze.mjs` malformed-JSONL abort (#2) — real bug, discards good data\n3. Fix `mcp-scan.mjs` string-finding silent corruption (#5) — real bug, worse than the report's crash claim\n4. `evolve.mjs --diagnose` coverage (#3) — largest untested surface\n5. `gepa.mjs` ops (#4), then the quick-win table (#7) and integration gaps (#8)\n",
+  "parsedOutput": {
+    "sections": [
+      {
+        "title": "Test Coverage Analysis — `plugins/ruflo-metaharness`",
+        "content": "\n**Infrastructure note:** there's no `package.json`/jest/vitest here. The \"test suite\" is `scripts/test-*.mjs` (6 files) using a hand-rolled `assert()` + pass/fail counter, gated by `scripts/smoke.sh`. All skeletons below follow that existing convention rather than introducing a new framework — nothing here is exported as a library, so most gaps require black-box `execFileSync` tests, not unit imports.\n\nI verified every claim below directly against source (this plugin already had an auto-generated `testgaps` report from today in `.claude-flow/metrics/test-gaps.json` — I cross-checked it rather than trusting it blindly, and corrected one inaccuracy, noted at #5).\n\n",
+        "level": 2
+      },
+      {
+        "title": "1. `threat-model.mjs` — stale local `SEVERITY_RANK` (real bug, highest priority)",
+        "content": "\n`scripts/threat-model.mjs:9` keeps its own `{clean:0,low:1,medium:2,high:3}` instead of importing the canonical table from `_harness.mjs:261` (`{clean,info,low,medium,warn,high,error,critical}`). If the harness reports `worst:\"critical\"`, `SEVERITY_RANK['critical']` is `undefined`, so `undefined >= threshold` → `false` — `--fail-on high` **never trips for a critical finding**.\n\n```js\n// scripts/test-threat-model-severity.mjs\nimport { execFileSync } from 'node:child_process';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\n// Requires a mock/fixture harness binary that returns worst:\"critical\"\nconst out = execFileSync('node', ['scripts/threat-model.mjs', '--path', 'FIXTURE_CRITICAL', '--fail-on', 'high'], { encoding: 'utf8' });\nconst payload = JSON.parse(out);\nassert(payload.alert.triggered === true, 'worst=critical must trigger --fail-on high (currently silently false)');\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\nFix: `import { SEVERITY_RANK } from './_harness.mjs'` instead of the local literal.\n\n",
+        "level": 3
+      },
+      {
+        "title": "2. `router-parallel-analyze.mjs` — one malformed JSONL line kills the whole run",
+        "content": "\n`scripts/router-parallel-analyze.mjs:90-95`: the `.map(l => JSON.parse(l))` has no per-line try/catch; the *outer* catch aborts with exit 2 for the entire file on a single bad line, discarding all valid rows.\n\n```js\n// scripts/test-router-parallel-malformed.mjs\nimport { execFileSync } from 'node:child_process';\nimport { writeFileSync, mkdtempSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { tmpdir } from 'node:os';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nconst dir = mkdtempSync(join(tmpdir(), 'rpa-'));\nconst file = join(dir, 'router-parallel.jsonl');\nconst goodRow = JSON.stringify({ bandit: {}, ser: {}, outcome: {} }); // must match the real filter shape (bandit && ser && outcome)\nwriteFileSync(file, Array(35).fill(goodRow).join('\\n') + '\\nNOT VALID JSON\\n');\n\nlet exitCode = 0;\ntry {\n  execFileSync('node', ['scripts/router-parallel-analyze.mjs', '--input', file], { encoding: 'utf8' });\n} catch (e) { exitCode = e.status; }\n\nassert(exitCode !== 2, 'one malformed JSONL line should not abort analysis of the other 35 valid rows');\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\nFix: wrap the per-line parse individually, skip+count bad lines instead of aborting.\n\n",
+        "level": 3
+      },
+      {
+        "title": "3. `evolve.mjs --diagnose` subsystem — zero coverage",
+        "content": "\n`looksLikeGepaTranscript`, `extractGepaTranscripts`, `summarizeTraces`, `buildDiagnosis` (evolve.mjs:146-239, ~90 lines) are module-private (no `export`) and never exercised by any of the 6 existing test scripts. Must be tested black-box via CLI.\n\n```js\n// scripts/test-evolve-diagnose.mjs\nimport { execFileSync } from 'node:child_process';\nimport { mkdtempSync, writeFileSync, mkdirSync } from 'node:fs';\nimport { tmpdir } from 'node:os';\nimport { join } from 'node:path';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nconst runsDir = mkdtempSync(join(tmpdir(), 'evolve-diag-'));\nmkdirSync(join(runsDir, '.metaharness', 'runs'), { recursive: true });\nwriteFileSync(join(runsDir, '.metaharness', 'runs', 'winner.json'), '{\"id\":\"v1\"}');\nwriteFileSync(join(runsDir, '.metaharness', 'runs', 'v1.json'), JSON.stringify({ id: 'v1', transcript: [] }));\n\nconst out = execFileSync('node', ['scripts/evolve.mjs', runsDir, '--diagnose', '--sandbox', 'mock'], { encoding: 'utf8' });\nassert(typeof JSON.parse(out).diagnosis === 'object', 'diagnosis object present for a valid winner+run record');\n\n// corrupt winner.json must degrade, not throw\nwriteFileSync(join(runsDir, '.metaharness', 'runs', 'winner.json'), '{not json');\nconst out2 = execFileSync('node', ['scripts/evolve.mjs', runsDir, '--diagnose', '--sandbox', 'mock'], { encoding: 'utf8' });\nassert(!!JSON.parse(out2).diagnosis, 'corrupt winner.json does not crash --diagnose');\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\n\n",
+        "level": 3
+      },
+      {
+        "title": "4. `gepa.mjs` — `validate`/`render`/`analyze` ops untested",
+        "content": "\nConfirmed all 4 ops (`genome|validate|render|analyze`, gepa.mjs:104) exist but only `genome` is exercised indirectly via MCP. `--alert-on-invalid` exit-1 path and the non-array `--transcript` exit-2 path (gepa.mjs:135-136) are unhit.\n\n```js\n// scripts/test-gepa-ops.mjs\nimport { execFileSync } from 'node:child_process';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nfor (const op of ['validate', 'render']) {\n  const out = execFileSync('node', ['scripts/gepa.mjs', '--op', op], { encoding: 'utf8' });\n  const payload = JSON.parse(out);\n  assert(payload.op === op || payload.degraded, `op=${op} returns a shaped payload or degrades cleanly`);\n}\n\ntry {\n  execFileSync('node', ['scripts/gepa.mjs', '--op', 'analyze', '--transcript', '{\"not\":\"array\"}'], { encoding: 'utf8' });\n  assert(false, 'non-array --transcript should exit non-zero');\n} catch (e) { assert(e.status === 2, 'non-array --transcript exits 2'); }\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\n\n",
+        "level": 3
+      },
+      {
+        "title": "5. `mcp-scan.mjs` — string-shaped finding produces silent corruption, **not a crash** (correction)",
+        "content": "\nThe auto-generated report claimed `findings.map` \"throws\" on a raw-string array element. I checked the map body (`mcp-scan.mjs:62-66`): `{...f, message: ...}` on a string `f` spreads its characters into numeric keys and `f.message`/`f.title`/`f.detail` are all `undefined` on a string — so `message` silently falls back to `''`, no throw. The real bug is **silent data corruption** (a finding becomes `{0:'a',1:' ',...,message:''}`), which is worse for debugging than a crash since it never surfaces.\n\n```js\n// scripts/test-mcp-scan-malformed-finding.mjs\nimport { parseMcpScanText } from './_harness.mjs'; // if exported; else drive via fixture stdout + execFileSync\n// Fixture: upstream JSON with findings: [\"a raw string\", {severity:'high', message:'ok'}]\n// Assert: string elements are filtered out or coerced to a real object BEFORE spread,\n// not silently spread into {0:'a',1:' ',...}.\n```\nRecommend a guard: `rawFindings.filter(f => f && typeof f === 'object')` before the `.map`.\n\n",
+        "level": 3
+      },
+      {
+        "title": "6. `security-bench.mjs` — `parseSecurityBenchMarkdown` not exported, no isolated test",
+        "content": "\nConfirmed not exported (`security-bench.mjs:77`). Fragile to upstream markdown/emoji/table drift.\n\n```js\n// scripts/test-security-bench-parse.mjs\n// Requires: export function parseSecurityBenchMarkdown(md) { ... }\nimport { parseSecurityBenchMarkdown } from './security-bench.mjs';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nconst wellFormed = `## Overall: ✅ PASS\\n| Gate | Status |\\n|---|---|\\n| injection | ✅ |\\n`;\nassert(parseSecurityBenchMarkdown(wellFormed).gates.length > 0, 'well-formed markdown parses gates');\n\nconst malformed = `unexpected upstream format with no matching tables`;\nconst r2 = parseSecurityBenchMarkdown(malformed);\nassert(r2.gates.length === 0, 'malformed markdown returns empty gates, not throw');\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);\n```\n\n",
+        "level": 3
+      },
+      {
+        "title": "7. Missing error-handling tests (quick wins, all confirmed against source)",
+        "content": "\n| File:line | Confirmed issue | Test scenario |\n|---|---|---|\n| `audit-trend.mjs:140` | `fingerprint()` truncates `message` at 80 chars via `.slice(0,80)` | Two findings identical in the first 80 chars but differing after — assert they are NOT deduped as \"unchanged\" |\n| `similarity.mjs:51,114` | `Number(process.argv[++i])` on `--alert-below abc` → `NaN`; `result.overall < NaN` is always `false` | `execFileSync(['similarity.mjs',...,'--alert-below','abc'])` should exit non-zero (reject bad arg), not silently succeed |\n| `redblue.mjs:90` | `!next.startsWith('-')` guard means `--some-flag -5` drops `-5` as the value (mistaken for a bare flag) | Passthrough arg with a negative-number value must be preserved, not swallowed |\n| `learn.mjs:92` | `checkout-required` detection is a fixed status string keyed off upstream stderr wording | Fixture stderr with slightly different wording — assert it doesn't silently misreport `status:'failed'` |\n| `_harness.mjs:269` | `rankSeverity` lowercases but doesn't trim — `rankSeverity(' high ')` → 0, not 3 | `assert(rankSeverity(' high ') === rankSeverity('high'))` — currently fails, flags the footgun |\n\n",
+        "level": 3
+      },
+      {
+        "title": "8. Integration test gaps",
+        "content": "\n- **`oia-audit.mjs`**: only the \"all 5 subprocesses degraded\" path is tested (`test-graceful-degradation.mjs`). No test for 1–4 of 5 degrading while the composite still reports healthy at the top level.\n- **`mint.mjs`**: the documented pre-0.1.13 `--target`-ignored workaround has no regression test against the currently pinned metaharness version — if upstream fixes it, the workaround's continued correctness is unverified either way.\n- **`test-parallel-pipeline.mjs`**: its TS-source check is a `grep`, not a runtime test — `recordPair`/`recordPairOutcome` in `router-parallel-recorder.ts` are pattern-matched as text, never actually executed against compiled `dist/`.\n- **`test-pipeline-roundtrip.mjs`**: `REPO_ROOT = dirname(dirname(dirname(SCRIPTS_DIR)))` is a fragile path assumption — relocating the plugin silently points at the wrong tree instead of failing loudly. Add a sentinel-file assertion (e.g., root `CLAUDE.md` exists) before proceeding.\n\n",
+        "level": 3
+      },
+      {
+        "title": "Priority order if picking a subset",
+        "content": "1. Fix `threat-model.mjs` `SEVERITY_RANK` (#1) — real bug, silently swallows critical findings\n2. Fix `router-parallel-analyze.mjs` malformed-JSONL abort (#2) — real bug, discards good data\n3. Fix `mcp-scan.mjs` string-finding silent corruption (#5) — real bug, worse than the report's crash claim\n4. `evolve.mjs --diagnose` coverage (#3) — largest untested surface\n5. `gepa.mjs` ops (#4), then the quick-win table (#7) and integration gaps (#8)",
+        "level": 3
+      }
+    ],
+    "codeBlocks": [
+      {
+        "language": "js",
+        "code": "// scripts/test-threat-model-severity.mjs\nimport { execFileSync } from 'node:child_process';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\n// Requires a mock/fixture harness binary that returns worst:\"critical\"\nconst out = execFileSync('node', ['scripts/threat-model.mjs', '--path', 'FIXTURE_CRITICAL', '--fail-on', 'high'], { encoding: 'utf8' });\nconst payload = JSON.parse(out);\nassert(payload.alert.triggered === true, 'worst=critical must trigger --fail-on high (currently silently false)');\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);"
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-router-parallel-malformed.mjs\nimport { execFileSync } from 'node:child_process';\nimport { writeFileSync, mkdtempSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { tmpdir } from 'node:os';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nconst dir = mkdtempSync(join(tmpdir(), 'rpa-'));\nconst file = join(dir, 'router-parallel.jsonl');\nconst goodRow = JSON.stringify({ bandit: {}, ser: {}, outcome: {} }); // must match the real filter shape (bandit && ser && outcome)\nwriteFileSync(file, Array(35).fill(goodRow).join('\\n') + '\\nNOT VALID JSON\\n');\n\nlet exitCode = 0;\ntry {\n  execFileSync('node', ['scripts/router-parallel-analyze.mjs', '--input', file], { encoding: 'utf8' });\n} catch (e) { exitCode = e.status; }\n\nassert(exitCode !== 2, 'one malformed JSONL line should not abort analysis of the other 35 valid rows');\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);"
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-evolve-diagnose.mjs\nimport { execFileSync } from 'node:child_process';\nimport { mkdtempSync, writeFileSync, mkdirSync } from 'node:fs';\nimport { tmpdir } from 'node:os';\nimport { join } from 'node:path';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nconst runsDir = mkdtempSync(join(tmpdir(), 'evolve-diag-'));\nmkdirSync(join(runsDir, '.metaharness', 'runs'), { recursive: true });\nwriteFileSync(join(runsDir, '.metaharness', 'runs', 'winner.json'), '{\"id\":\"v1\"}');\nwriteFileSync(join(runsDir, '.metaharness', 'runs', 'v1.json'), JSON.stringify({ id: 'v1', transcript: [] }));\n\nconst out = execFileSync('node', ['scripts/evolve.mjs', runsDir, '--diagnose', '--sandbox', 'mock'], { encoding: 'utf8' });\nassert(typeof JSON.parse(out).diagnosis === 'object', 'diagnosis object present for a valid winner+run record');\n\n// corrupt winner.json must degrade, not throw\nwriteFileSync(join(runsDir, '.metaharness', 'runs', 'winner.json'), '{not json');\nconst out2 = execFileSync('node', ['scripts/evolve.mjs', runsDir, '--diagnose', '--sandbox', 'mock'], { encoding: 'utf8' });\nassert(!!JSON.parse(out2).diagnosis, 'corrupt winner.json does not crash --diagnose');\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);"
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-gepa-ops.mjs\nimport { execFileSync } from 'node:child_process';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nfor (const op of ['validate', 'render']) {\n  const out = execFileSync('node', ['scripts/gepa.mjs', '--op', op], { encoding: 'utf8' });\n  const payload = JSON.parse(out);\n  assert(payload.op === op || payload.degraded, `op=${op} returns a shaped payload or degrades cleanly`);\n}\n\ntry {\n  execFileSync('node', ['scripts/gepa.mjs', '--op', 'analyze', '--transcript', '{\"not\":\"array\"}'], { encoding: 'utf8' });\n  assert(false, 'non-array --transcript should exit non-zero');\n} catch (e) { assert(e.status === 2, 'non-array --transcript exits 2'); }\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);"
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-mcp-scan-malformed-finding.mjs\nimport { parseMcpScanText } from './_harness.mjs'; // if exported; else drive via fixture stdout + execFileSync\n// Fixture: upstream JSON with findings: [\"a raw string\", {severity:'high', message:'ok'}]\n// Assert: string elements are filtered out or coerced to a real object BEFORE spread,\n// not silently spread into {0:'a',1:' ',...}."
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-security-bench-parse.mjs\n// Requires: export function parseSecurityBenchMarkdown(md) { ... }\nimport { parseSecurityBenchMarkdown } from './security-bench.mjs';\nlet passed = 0, failed = 0;\nconst assert = (c, l) => c ? passed++ : (failed++, console.log(`✗ ${l}`));\n\nconst wellFormed = `## Overall: ✅ PASS\\n| Gate | Status |\\n|---|---|\\n| injection | ✅ |\\n`;\nassert(parseSecurityBenchMarkdown(wellFormed).gates.length > 0, 'well-formed markdown parses gates');\n\nconst malformed = `unexpected upstream format with no matching tables`;\nconst r2 = parseSecurityBenchMarkdown(malformed);\nassert(r2.gates.length === 0, 'malformed markdown returns empty gates, not throw');\n\nconsole.log(failed ? `${failed} failed` : `${passed} passed, 0 failed`);\nprocess.exit(failed ? 1 : 0);"
+      }
+    ]
+  },
+  "durationMs": 150058,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T14:43:08.375Z",
+  "executionId": "testgaps_1783608038317_jrd66e"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783609388416_mc3zoe_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T15:03:08.417Z] PROMPT
+============================================================
+Analyze test coverage and identify gaps:
+- Find untested functions and classes
+- Identify edge cases not covered
+- Suggest new test scenarios
+- Check for missing error handling tests
+- Identify integration test gaps
+
+For each gap, provide a test skeleton.
+
+## Instructions
+
+Analyze the codebase and provide your response following the format specified in the task.

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783609388416_mc3zoe_result.log

@@ -0,0 +1,82 @@
+[2026-07-09T15:07:01.357Z] RESULT
+============================================================
+{
+  "success": true,
+  "output": "## Test coverage analysis — `plugins/ruflo-metaharness`\n\nThe plugin has no framework — tests are plain `assert()`-based `.mjs` scripts run via `node scripts/test-*.mjs` (see `test-similarity.mjs` for the convention: `assert(cond, label)`, `--format json`, exit 1 on failure). Existing coverage is almost entirely **black-box CLI/subprocess tests** (`test-graceful-degradation.mjs`, `test-pipeline-roundtrip.mjs`, `test-parallel-pipeline.mjs`, `test-mcp-tools.mjs`) plus one genuine unit-test file (`test-similarity.mjs`, covering `_similarity.mjs` + parts of `_harness.mjs`). Benchmarks (`bench*.mjs`) exercise code paths but only assert performance, not correctness.\n\n### 1. Untested module: `_invoke.mjs` (highest priority — zero direct tests)\n\nThis is the shared plumbing layer (~15 scripts depend on it) and every exported function is pure or near-pure, yet none has a direct unit test — only indirect, best-effort coverage through full CLI subprocess runs.\n\n```js\n// scripts/test-invoke.mjs — unit tests for _invoke.mjs (currently untested)\nimport {\n  classifyDegraded, injectJson, parseTrailingJson,\n  satisfiesTildeRange, cacheBaseDir, DEGRADED_RX,\n} from './_invoke.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? passed++ : (failed++, console.log(`✗ ${label}`)); }\n\n// classifyDegraded\nassert(classifyDegraded('', null, 'x').reason === 'x-timeout', 'null exitCode -> timeout');\nassert(classifyDegraded('npm ERR! 404', 1, 'x').degraded === true, 'npm ERR matches DEGRADED_RX');\nassert(classifyDegraded('boom', 1, 'x').degraded === false, 'unmatched stderr -> not degraded');\nassert(classifyDegraded(undefined, 1, 'x').degraded === false, 'undefined stderr does not throw');\n\n// injectJson\nassert(injectJson(['--foo'], true).includes('--json'), 'appends --json when wanted');\nassert(injectJson(['--json'], true).filter(a => a === '--json').length === 1, 'no duplicate --json');\nassert(injectJson(['--foo'], false).includes('--json') === false, 'no --json when not wanted');\nassert(injectJson(['--foo'], true) !== ['--foo'], 'does not mutate/alias input array');\n\n// parseTrailingJson\nassert(parseTrailingJson('progress {a:1} Result: {\"ok\":true}').ok === true, 'grabs LAST json block, not first-looking-like-json text');\nassert(parseTrailingJson('{\"a\":{\"b\":1}}').a.b === 1, 'nested object via greedy fallback');\nassert(parseTrailingJson('no json here') === null, 'no JSON -> null');\nassert(parseTrailingJson('') === null, 'empty string -> null');\nassert(parseTrailingJson(undefined) === null, 'undefined stdout does not throw');\n\n// satisfiesTildeRange\nassert(satisfiesTildeRange('0.3.2', '~0.3.0') === true, 'patch >= pin patch satisfies');\nassert(satisfiesTildeRange('0.3.0', '~0.3.1') === false, 'patch < pin patch fails');\nassert(satisfiesTildeRange('0.4.0', '~0.3.0') === false, 'minor mismatch fails');\nassert(satisfiesTildeRange('not-a-version', '~0.3.0') === false, 'garbage version -> false, no throw');\nassert(satisfiesTildeRange('0.3.0', 'not-a-range') === false, 'garbage pin -> false, no throw');\n\n// cacheBaseDir (env override test seam)\nprocess.env.RUFLO_METAHARNESS_CACHE_BASE = '/tmp/x';\nassert(cacheBaseDir() === '/tmp/x', 'RUFLO_METAHARNESS_CACHE_BASE override respected');\ndelete process.env.RUFLO_METAHARNESS_CACHE_BASE;\n\nconsole.log(`${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);\n```\n\nAlso missing: `findLocalPackageDir` (stale-major/minor-in-ancestor-node_modules is *skipped*, per its docstring — untested), `ensureCachedInstall` (cache-hit-short-circuit vs cold-install path — can be unit tested by pointing `RUFLO_METAHARNESS_CACHE_BASE` at a temp dir and stubbing `pkg`), and `importOptionalLibrary`'s \"recoverable vs rethrow\" branch (a non-`MODULE_NOT_FOUND` error, e.g. a syntax error in the imported module, should propagate — currently unverified).\n\n### 2. Untested safety guard: `mint.mjs` `safetyChecks()`\n\n`mint.mjs` has destructive-action guards (refuse to target the repo root or a path inside it, `--name`/`--template` required, default `--target` to a tmpdir) but `safetyChecks()` isn't exported and is explicitly excluded from `test-graceful-degradation.mjs`'s skill list (\"mint requires a `--name` argv to even start\"). **No test exercises the refusal paths at all.**\n\n```js\n// scripts/test-mint-safety.mjs — subprocess-level test since safetyChecks() isn't exported\nimport { spawnSync } from 'node:child_process';\nimport { join, dirname } from 'node:path';\nimport { fileURLToPath } from 'node:url';\n\nconst SCRIPT = join(dirname(fileURLToPath(import.meta.url)), 'mint.mjs');\nfunction run(args) { return spawnSync('node', [SCRIPT, ...args], { encoding: 'utf-8' }); }\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? passed++ : (failed++, console.log(`✗ ${label}`)); }\n\nassert(run(['--template', 'minimal']).status === 2, 'missing --name exits 2');\nassert(run(['--name', 'x']).status === 2, 'missing --template exits 2');\n\nconst repoRoot = process.cwd();\nconst r1 = run(['--name', 'x', '--template', 'minimal', '--target', repoRoot]);\nassert(r1.status === 2 && /refusing to write to project root/.test(r1.stderr), 'refuses exact repo root');\n\nconst r2 = run(['--name', 'x', '--template', 'minimal', '--target', join(repoRoot, 'sub')]);\nassert(r2.status === 2 && /refusing to write inside/.test(r2.stderr), 'refuses path inside repo root');\n\nconsole.log(`${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);\n```\n\n**Recommendation independent of the test:** export `safetyChecks` from `mint.mjs` so it can be unit-tested without spawning `node` each time (subprocess tests here cost ~100-300ms/case vs <1ms for an in-process call).\n\n### 3. Documented-but-unfixed bugs with no regression test\n\nPer `test-graceful-degradation.mjs`'s own comments (iter 55), three known bugs are carved out of the drill rather than fixed+covered:\n- `mint`'s degraded payload doesn't include the literal `\"degraded\": true` marker.\n- `drift-from-history` exits `2` (config error) instead of `3` (test-cannot-run) when there's no audit history yet.\n- `oia-audit` can exceed the 180s local timeout in the unreachable-registry case.\n\nNone of these have a regression test asserting the *correct* future behavior — when fixed, nothing will catch a re-regression.\n\n```js\n// Add to test-graceful-degradation.mjs once fixed, or as a standalone xfail-style test:\nconst mintR = run('mint.mjs', ['--name', 'x', '--template', 'minimal', '--target', '/tmp/mint-drill']);\nconst mintJson = JSON.parse(mintR.stdout);\nassert(mintJson.degraded === true, 'mint degraded payload includes literal degraded:true');\n\nconst dfhR = run('drift-from-history.mjs', ['--format', 'json']); // no prior audit records\nassert(dfhR.exitCode === 3, 'drift-from-history exits 3 (test-cannot-run) on empty history, not 2');\n```\n\n### 4. `audit-list.mjs` — untested pure function `parseDurationMs()`\n\n```js\n// scripts/test-audit-list.mjs\n// parseDurationMs isn't exported today — export it first, then:\nimport { parseDurationMs } from './audit-list.mjs';\nassert(parseDurationMs('30d') === 30 * 86_400_000, '30d parses to ms');\nassert(parseDurationMs('1w') !== null, 'w unit supported');\nassert(parseDurationMs('abc') === null, 'garbage spec returns null, not throw');\nassert(parseDurationMs('') === null, 'empty spec returns null');\nassert(parseDurationMs('-5d') === null, 'negative count rejected'); // regex requires \\d+, confirm no coercion bug\n```\n\n### 5. `router-parallel-analyze.mjs` — partial-pass / boundary cases missing\n\n`test-parallel-pipeline.mjs` only checks the all-pass and all-fail fixtures for the 3-criteria AND-gate. It never verifies **AND semantics under partial failure** (e.g. quality passes, cost fails, latency passes → must still be non-promotable) or **exact-threshold boundaries** (`qualityImprovementPct === 2` should fail since the check is `> 2`, not `>= 2`).\n\n```js\n// Extend test-parallel-pipeline.mjs (or a new test-router-gate.mjs) with fixtures where:\n// - exactly 2 of 3 criteria pass -> verdict.promotable === false (guards against an accidental OR)\n// - qualityImprovementPct is exactly 2.00, usdIncreasePct exactly 1.00, latencyIncreasePct exactly 5.00\n//   -> all three individually false (boundary is exclusive)\n```\n\n### 6. Integration gaps: `redblue.mjs`, `learn.mjs`, `gepa.mjs`\n\n`test-mcp-tools.mjs` only proves the MCP tool wrapper's handler is *callable* with minimal input (structural contract), not that `redblue.mjs`/`learn.mjs`/`gepa.mjs`'s actual CLI behavior is correct end-to-end (e.g. redblue's 4-gate human-gated submit flow, `--mock-judge` vs real-judge cost capping, `gepa --op analyze` failure-mode classification). `test-graceful-degradation.mjs` also doesn't include these three in its skill list, so their degraded-path behavior (`_redblue.mjs`'s `emitRedblueDegradedJsonAndExit`, `_darwin.mjs`'s `importGepa`) is unverified with an unreachable registry.\n\n```js\n// Extend test-graceful-degradation.mjs's `skills` array:\n{ name: 'redblue', args: ['run', '--mock-judge', '--tests', '1', '--format', 'json'] },\n{ name: 'gepa',    args: ['--op', 'genome', '--format', 'json'] },\n```\n\n### Summary\n\n| Gap | File | Severity |\n|---|---|---|\n| Zero unit tests for shared plumbing (`classifyDegraded`, `parseTrailingJson`, `satisfiesTildeRange`, etc.) | `_invoke.mjs` | High — used by ~15 scripts |\n| Destructive-action guard untested | `mint.mjs` `safetyChecks()` | High — safety-critical, currently unreachable by any test |\n| 3 known bugs with no regression test | `mint.mjs`, `drift-from-history.mjs`, `oia-audit.mjs` | Medium — will silently re-regress |\n| Pure helper untested | `audit-list.mjs` `parseDurationMs()` | Low |\n| AND-gate boundary/partial-fail cases | `router-parallel-analyze.mjs` | Medium — could hide an OR-vs-AND logic bug |\n| Degraded-path untested for 3 scripts | `redblue.mjs`, `learn.mjs`, `gepa.mjs` | Medium — ADR-150 constraint unverified for these |\n\nI didn't run these skeletons — they're illustrative gaps to fill, following the existing plain-assert convention. Want me to actually write and wire up any of these as real files (e.g. `test-invoke.mjs` and `test-mint-safety.mjs` are the two highest-value ones)?\n",
+  "parsedOutput": {
+    "sections": [
+      {
+        "title": "Test coverage analysis — `plugins/ruflo-metaharness`",
+        "content": "\nThe plugin has no framework — tests are plain `assert()`-based `.mjs` scripts run via `node scripts/test-*.mjs` (see `test-similarity.mjs` for the convention: `assert(cond, label)`, `--format json`, exit 1 on failure). Existing coverage is almost entirely **black-box CLI/subprocess tests** (`test-graceful-degradation.mjs`, `test-pipeline-roundtrip.mjs`, `test-parallel-pipeline.mjs`, `test-mcp-tools.mjs`) plus one genuine unit-test file (`test-similarity.mjs`, covering `_similarity.mjs` + parts of `_harness.mjs`). Benchmarks (`bench*.mjs`) exercise code paths but only assert performance, not correctness.\n\n",
+        "level": 2
+      },
+      {
+        "title": "1. Untested module: `_invoke.mjs` (highest priority — zero direct tests)",
+        "content": "\nThis is the shared plumbing layer (~15 scripts depend on it) and every exported function is pure or near-pure, yet none has a direct unit test — only indirect, best-effort coverage through full CLI subprocess runs.\n\n```js\n// scripts/test-invoke.mjs — unit tests for _invoke.mjs (currently untested)\nimport {\n  classifyDegraded, injectJson, parseTrailingJson,\n  satisfiesTildeRange, cacheBaseDir, DEGRADED_RX,\n} from './_invoke.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? passed++ : (failed++, console.log(`✗ ${label}`)); }\n\n// classifyDegraded\nassert(classifyDegraded('', null, 'x').reason === 'x-timeout', 'null exitCode -> timeout');\nassert(classifyDegraded('npm ERR! 404', 1, 'x').degraded === true, 'npm ERR matches DEGRADED_RX');\nassert(classifyDegraded('boom', 1, 'x').degraded === false, 'unmatched stderr -> not degraded');\nassert(classifyDegraded(undefined, 1, 'x').degraded === false, 'undefined stderr does not throw');\n\n// injectJson\nassert(injectJson(['--foo'], true).includes('--json'), 'appends --json when wanted');\nassert(injectJson(['--json'], true).filter(a => a === '--json').length === 1, 'no duplicate --json');\nassert(injectJson(['--foo'], false).includes('--json') === false, 'no --json when not wanted');\nassert(injectJson(['--foo'], true) !== ['--foo'], 'does not mutate/alias input array');\n\n// parseTrailingJson\nassert(parseTrailingJson('progress {a:1} Result: {\"ok\":true}').ok === true, 'grabs LAST json block, not first-looking-like-json text');\nassert(parseTrailingJson('{\"a\":{\"b\":1}}').a.b === 1, 'nested object via greedy fallback');\nassert(parseTrailingJson('no json here') === null, 'no JSON -> null');\nassert(parseTrailingJson('') === null, 'empty string -> null');\nassert(parseTrailingJson(undefined) === null, 'undefined stdout does not throw');\n\n// satisfiesTildeRange\nassert(satisfiesTildeRange('0.3.2', '~0.3.0') === true, 'patch >= pin patch satisfies');\nassert(satisfiesTildeRange('0.3.0', '~0.3.1') === false, 'patch < pin patch fails');\nassert(satisfiesTildeRange('0.4.0', '~0.3.0') === false, 'minor mismatch fails');\nassert(satisfiesTildeRange('not-a-version', '~0.3.0') === false, 'garbage version -> false, no throw');\nassert(satisfiesTildeRange('0.3.0', 'not-a-range') === false, 'garbage pin -> false, no throw');\n\n// cacheBaseDir (env override test seam)\nprocess.env.RUFLO_METAHARNESS_CACHE_BASE = '/tmp/x';\nassert(cacheBaseDir() === '/tmp/x', 'RUFLO_METAHARNESS_CACHE_BASE override respected');\ndelete process.env.RUFLO_METAHARNESS_CACHE_BASE;\n\nconsole.log(`${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);\n```\n\nAlso missing: `findLocalPackageDir` (stale-major/minor-in-ancestor-node_modules is *skipped*, per its docstring — untested), `ensureCachedInstall` (cache-hit-short-circuit vs cold-install path — can be unit tested by pointing `RUFLO_METAHARNESS_CACHE_BASE` at a temp dir and stubbing `pkg`), and `importOptionalLibrary`'s \"recoverable vs rethrow\" branch (a non-`MODULE_NOT_FOUND` error, e.g. a syntax error in the imported module, should propagate — currently unverified).\n\n",
+        "level": 3
+      },
+      {
+        "title": "2. Untested safety guard: `mint.mjs` `safetyChecks()`",
+        "content": "\n`mint.mjs` has destructive-action guards (refuse to target the repo root or a path inside it, `--name`/`--template` required, default `--target` to a tmpdir) but `safetyChecks()` isn't exported and is explicitly excluded from `test-graceful-degradation.mjs`'s skill list (\"mint requires a `--name` argv to even start\"). **No test exercises the refusal paths at all.**\n\n```js\n// scripts/test-mint-safety.mjs — subprocess-level test since safetyChecks() isn't exported\nimport { spawnSync } from 'node:child_process';\nimport { join, dirname } from 'node:path';\nimport { fileURLToPath } from 'node:url';\n\nconst SCRIPT = join(dirname(fileURLToPath(import.meta.url)), 'mint.mjs');\nfunction run(args) { return spawnSync('node', [SCRIPT, ...args], { encoding: 'utf-8' }); }\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? passed++ : (failed++, console.log(`✗ ${label}`)); }\n\nassert(run(['--template', 'minimal']).status === 2, 'missing --name exits 2');\nassert(run(['--name', 'x']).status === 2, 'missing --template exits 2');\n\nconst repoRoot = process.cwd();\nconst r1 = run(['--name', 'x', '--template', 'minimal', '--target', repoRoot]);\nassert(r1.status === 2 && /refusing to write to project root/.test(r1.stderr), 'refuses exact repo root');\n\nconst r2 = run(['--name', 'x', '--template', 'minimal', '--target', join(repoRoot, 'sub')]);\nassert(r2.status === 2 && /refusing to write inside/.test(r2.stderr), 'refuses path inside repo root');\n\nconsole.log(`${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);\n```\n\n**Recommendation independent of the test:** export `safetyChecks` from `mint.mjs` so it can be unit-tested without spawning `node` each time (subprocess tests here cost ~100-300ms/case vs <1ms for an in-process call).\n\n",
+        "level": 3
+      },
+      {
+        "title": "3. Documented-but-unfixed bugs with no regression test",
+        "content": "\nPer `test-graceful-degradation.mjs`'s own comments (iter 55), three known bugs are carved out of the drill rather than fixed+covered:\n- `mint`'s degraded payload doesn't include the literal `\"degraded\": true` marker.\n- `drift-from-history` exits `2` (config error) instead of `3` (test-cannot-run) when there's no audit history yet.\n- `oia-audit` can exceed the 180s local timeout in the unreachable-registry case.\n\nNone of these have a regression test asserting the *correct* future behavior — when fixed, nothing will catch a re-regression.\n\n```js\n// Add to test-graceful-degradation.mjs once fixed, or as a standalone xfail-style test:\nconst mintR = run('mint.mjs', ['--name', 'x', '--template', 'minimal', '--target', '/tmp/mint-drill']);\nconst mintJson = JSON.parse(mintR.stdout);\nassert(mintJson.degraded === true, 'mint degraded payload includes literal degraded:true');\n\nconst dfhR = run('drift-from-history.mjs', ['--format', 'json']); // no prior audit records\nassert(dfhR.exitCode === 3, 'drift-from-history exits 3 (test-cannot-run) on empty history, not 2');\n```\n\n",
+        "level": 3
+      },
+      {
+        "title": "4. `audit-list.mjs` — untested pure function `parseDurationMs()`",
+        "content": "\n```js\n// scripts/test-audit-list.mjs\n// parseDurationMs isn't exported today — export it first, then:\nimport { parseDurationMs } from './audit-list.mjs';\nassert(parseDurationMs('30d') === 30 * 86_400_000, '30d parses to ms');\nassert(parseDurationMs('1w') !== null, 'w unit supported');\nassert(parseDurationMs('abc') === null, 'garbage spec returns null, not throw');\nassert(parseDurationMs('') === null, 'empty spec returns null');\nassert(parseDurationMs('-5d') === null, 'negative count rejected'); // regex requires \\d+, confirm no coercion bug\n```\n\n",
+        "level": 3
+      },
+      {
+        "title": "5. `router-parallel-analyze.mjs` — partial-pass / boundary cases missing",
+        "content": "\n`test-parallel-pipeline.mjs` only checks the all-pass and all-fail fixtures for the 3-criteria AND-gate. It never verifies **AND semantics under partial failure** (e.g. quality passes, cost fails, latency passes → must still be non-promotable) or **exact-threshold boundaries** (`qualityImprovementPct === 2` should fail since the check is `> 2`, not `>= 2`).\n\n```js\n// Extend test-parallel-pipeline.mjs (or a new test-router-gate.mjs) with fixtures where:\n// - exactly 2 of 3 criteria pass -> verdict.promotable === false (guards against an accidental OR)\n// - qualityImprovementPct is exactly 2.00, usdIncreasePct exactly 1.00, latencyIncreasePct exactly 5.00\n//   -> all three individually false (boundary is exclusive)\n```\n\n",
+        "level": 3
+      },
+      {
+        "title": "6. Integration gaps: `redblue.mjs`, `learn.mjs`, `gepa.mjs`",
+        "content": "\n`test-mcp-tools.mjs` only proves the MCP tool wrapper's handler is *callable* with minimal input (structural contract), not that `redblue.mjs`/`learn.mjs`/`gepa.mjs`'s actual CLI behavior is correct end-to-end (e.g. redblue's 4-gate human-gated submit flow, `--mock-judge` vs real-judge cost capping, `gepa --op analyze` failure-mode classification). `test-graceful-degradation.mjs` also doesn't include these three in its skill list, so their degraded-path behavior (`_redblue.mjs`'s `emitRedblueDegradedJsonAndExit`, `_darwin.mjs`'s `importGepa`) is unverified with an unreachable registry.\n\n```js\n// Extend test-graceful-degradation.mjs's `skills` array:\n{ name: 'redblue', args: ['run', '--mock-judge', '--tests', '1', '--format', 'json'] },\n{ name: 'gepa',    args: ['--op', 'genome', '--format', 'json'] },\n```\n\n",
+        "level": 3
+      },
+      {
+        "title": "Summary",
+        "content": "| Gap | File | Severity |\n|---|---|---|\n| Zero unit tests for shared plumbing (`classifyDegraded`, `parseTrailingJson`, `satisfiesTildeRange`, etc.) | `_invoke.mjs` | High — used by ~15 scripts |\n| Destructive-action guard untested | `mint.mjs` `safetyChecks()` | High — safety-critical, currently unreachable by any test |\n| 3 known bugs with no regression test | `mint.mjs`, `drift-from-history.mjs`, `oia-audit.mjs` | Medium — will silently re-regress |\n| Pure helper untested | `audit-list.mjs` `parseDurationMs()` | Low |\n| AND-gate boundary/partial-fail cases | `router-parallel-analyze.mjs` | Medium — could hide an OR-vs-AND logic bug |\n| Degraded-path untested for 3 scripts | `redblue.mjs`, `learn.mjs`, `gepa.mjs` | Medium — ADR-150 constraint unverified for these |\n\nI didn't run these skeletons — they're illustrative gaps to fill, following the existing plain-assert convention. Want me to actually write and wire up any of these as real files (e.g. `test-invoke.mjs` and `test-mint-safety.mjs` are the two highest-value ones)?",
+        "level": 3
+      }
+    ],
+    "codeBlocks": [
+      {
+        "language": "js",
+        "code": "// scripts/test-invoke.mjs — unit tests for _invoke.mjs (currently untested)\nimport {\n  classifyDegraded, injectJson, parseTrailingJson,\n  satisfiesTildeRange, cacheBaseDir, DEGRADED_RX,\n} from './_invoke.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? passed++ : (failed++, console.log(`✗ ${label}`)); }\n\n// classifyDegraded\nassert(classifyDegraded('', null, 'x').reason === 'x-timeout', 'null exitCode -> timeout');\nassert(classifyDegraded('npm ERR! 404', 1, 'x').degraded === true, 'npm ERR matches DEGRADED_RX');\nassert(classifyDegraded('boom', 1, 'x').degraded === false, 'unmatched stderr -> not degraded');\nassert(classifyDegraded(undefined, 1, 'x').degraded === false, 'undefined stderr does not throw');\n\n// injectJson\nassert(injectJson(['--foo'], true).includes('--json'), 'appends --json when wanted');\nassert(injectJson(['--json'], true).filter(a => a === '--json').length === 1, 'no duplicate --json');\nassert(injectJson(['--foo'], false).includes('--json') === false, 'no --json when not wanted');\nassert(injectJson(['--foo'], true) !== ['--foo'], 'does not mutate/alias input array');\n\n// parseTrailingJson\nassert(parseTrailingJson('progress {a:1} Result: {\"ok\":true}').ok === true, 'grabs LAST json block, not first-looking-like-json text');\nassert(parseTrailingJson('{\"a\":{\"b\":1}}').a.b === 1, 'nested object via greedy fallback');\nassert(parseTrailingJson('no json here') === null, 'no JSON -> null');\nassert(parseTrailingJson('') === null, 'empty string -> null');\nassert(parseTrailingJson(undefined) === null, 'undefined stdout does not throw');\n\n// satisfiesTildeRange\nassert(satisfiesTildeRange('0.3.2', '~0.3.0') === true, 'patch >= pin patch satisfies');\nassert(satisfiesTildeRange('0.3.0', '~0.3.1') === false, 'patch < pin patch fails');\nassert(satisfiesTildeRange('0.4.0', '~0.3.0') === false, 'minor mismatch fails');\nassert(satisfiesTildeRange('not-a-version', '~0.3.0') === false, 'garbage version -> false, no throw');\nassert(satisfiesTildeRange('0.3.0', 'not-a-range') === false, 'garbage pin -> false, no throw');\n\n// cacheBaseDir (env override test seam)\nprocess.env.RUFLO_METAHARNESS_CACHE_BASE = '/tmp/x';\nassert(cacheBaseDir() === '/tmp/x', 'RUFLO_METAHARNESS_CACHE_BASE override respected');\ndelete process.env.RUFLO_METAHARNESS_CACHE_BASE;\n\nconsole.log(`${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);"
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-mint-safety.mjs — subprocess-level test since safetyChecks() isn't exported\nimport { spawnSync } from 'node:child_process';\nimport { join, dirname } from 'node:path';\nimport { fileURLToPath } from 'node:url';\n\nconst SCRIPT = join(dirname(fileURLToPath(import.meta.url)), 'mint.mjs');\nfunction run(args) { return spawnSync('node', [SCRIPT, ...args], { encoding: 'utf-8' }); }\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? passed++ : (failed++, console.log(`✗ ${label}`)); }\n\nassert(run(['--template', 'minimal']).status === 2, 'missing --name exits 2');\nassert(run(['--name', 'x']).status === 2, 'missing --template exits 2');\n\nconst repoRoot = process.cwd();\nconst r1 = run(['--name', 'x', '--template', 'minimal', '--target', repoRoot]);\nassert(r1.status === 2 && /refusing to write to project root/.test(r1.stderr), 'refuses exact repo root');\n\nconst r2 = run(['--name', 'x', '--template', 'minimal', '--target', join(repoRoot, 'sub')]);\nassert(r2.status === 2 && /refusing to write inside/.test(r2.stderr), 'refuses path inside repo root');\n\nconsole.log(`${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);"
+      },
+      {
+        "language": "js",
+        "code": "// Add to test-graceful-degradation.mjs once fixed, or as a standalone xfail-style test:\nconst mintR = run('mint.mjs', ['--name', 'x', '--template', 'minimal', '--target', '/tmp/mint-drill']);\nconst mintJson = JSON.parse(mintR.stdout);\nassert(mintJson.degraded === true, 'mint degraded payload includes literal degraded:true');\n\nconst dfhR = run('drift-from-history.mjs', ['--format', 'json']); // no prior audit records\nassert(dfhR.exitCode === 3, 'drift-from-history exits 3 (test-cannot-run) on empty history, not 2');"
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-audit-list.mjs\n// parseDurationMs isn't exported today — export it first, then:\nimport { parseDurationMs } from './audit-list.mjs';\nassert(parseDurationMs('30d') === 30 * 86_400_000, '30d parses to ms');\nassert(parseDurationMs('1w') !== null, 'w unit supported');\nassert(parseDurationMs('abc') === null, 'garbage spec returns null, not throw');\nassert(parseDurationMs('') === null, 'empty spec returns null');\nassert(parseDurationMs('-5d') === null, 'negative count rejected'); // regex requires \\d+, confirm no coercion bug"
+      },
+      {
+        "language": "js",
+        "code": "// Extend test-parallel-pipeline.mjs (or a new test-router-gate.mjs) with fixtures where:\n// - exactly 2 of 3 criteria pass -> verdict.promotable === false (guards against an accidental OR)\n// - qualityImprovementPct is exactly 2.00, usdIncreasePct exactly 1.00, latencyIncreasePct exactly 5.00\n//   -> all three individually false (boundary is exclusive)"
+      },
+      {
+        "language": "js",
+        "code": "// Extend test-graceful-degradation.mjs's `skills` array:\n{ name: 'redblue', args: ['run', '--mock-judge', '--tests', '1', '--format', 'json'] },\n{ name: 'gepa',    args: ['--op', 'genome', '--format', 'json'] },"
+      }
+    ]
+  },
+  "durationMs": 232941,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T15:07:01.357Z",
+  "executionId": "testgaps_1783609388416_mc3zoe"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783610821384_tzqvlb_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T15:27:01.385Z] PROMPT
+============================================================
+Analyze test coverage and identify gaps:
+- Find untested functions and classes
+- Identify edge cases not covered
+- Suggest new test scenarios
+- Check for missing error handling tests
+- Identify integration test gaps
+
+For each gap, provide a test skeleton.
+
+## Instructions
+
+Analyze the codebase and provide your response following the format specified in the task.

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783610821384_tzqvlb_result.log

@@ -0,0 +1,17 @@
+[2026-07-09T15:27:03.262Z] RESULT
+============================================================
+{
+  "success": false,
+  "output": "You've hit your session limit · resets 2pm (America/Toronto)\n",
+  "parsedOutput": {
+    "sections": [],
+    "codeBlocks": []
+  },
+  "durationMs": 1878,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T15:27:03.262Z",
+  "executionId": "testgaps_1783610821384_tzqvlb",
+  "error": "Process exited with code 1"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783612023285_buygpo_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T15:47:03.286Z] PROMPT
+============================================================
+Analyze test coverage and identify gaps:
+- Find untested functions and classes
+- Identify edge cases not covered
+- Suggest new test scenarios
+- Check for missing error handling tests
+- Identify integration test gaps
+
+For each gap, provide a test skeleton.
+
+## Instructions
+
+Analyze the codebase and provide your response following the format specified in the task.

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783612023285_buygpo_result.log

@@ -0,0 +1,57 @@
+[2026-07-09T15:53:19.262Z] RESULT
+============================================================
+{
+  "success": true,
+  "output": "## Test Coverage Analysis — `plugins/ruflo-metaharness`\n\nNo formal test framework exists here — just 6 ad-hoc `test-*.mjs` smoke/E2E scripts covering ~34 plain `.mjs` files (~7,000 lines). Full report:\n\n## Current Coverage Summary\n- **`test-graceful-degradation.mjs`** — drills ADR-150 rule #3 (registry unreachable → all scripts degrade cleanly) across 8 of the scripts. Misses `mint`, `similarity`, `redblue`, `learn`, `gepa`, `drift-from-history`, `router-parallel-analyze`.\n- **`test-mcp-tools.mjs`** — smoke-tests the 15 MCP tool handlers (envelope shape only for most; deeper checks for 4 tools).\n- **`test-parallel-pipeline.mjs`** — E2E for the router parallel-logging AND-gate, using a hand-synthesized JSONL fixture (not the real TS writer).\n- **`test-pipeline-roundtrip.mjs`** — deepest test: real `oia-audit --dry-run` chained through `audit-trend` → `drift-from-history`, 12 stages.\n- **`test-similarity.mjs`** — pure unit tests of `_similarity.mjs` internals + a frozen regression anchor.\n- **`test-with-openrouter.mjs`** — real-money E2E via GCP-secret-gated OpenRouter key + **upstream** `metaharness` CLI directly (bypasses this plugin's `mint.mjs` wrapper entirely).\n\n## Highest-priority finding — likely a real bug\n`threat-model.mjs` defines its own local `SEVERITY_RANK = {clean,low,medium,high}` instead of importing the shared frozen table from `_harness.mjs` (which uses `info/warn/error/critical`). Since the two vocabularies don't match, `SEVERITY_RANK[worst] >= threshold` evaluates to `undefined >= N === false` whenever upstream reports `critical` — meaning `--fail-on high` **silently never fires** on a critical finding. No test catches this today.\n\n## Untested functions/classes (selected)\n| File | Function | Why it matters |\n|---|---|---|\n| `mint.mjs` | `safetyChecks()` | Path-traversal refusals never exercised by any script in-repo |\n| `_invoke.mjs` | `parseTrailingJson()` | Exists specifically to fix a documented past bug (progress-line JSON shadowing); no regression test |\n| `_invoke.mjs` | `satisfiesTildeRange()` | Pre-release/malformed version strings unhandled by any test |\n| `evolve.mjs` | `buildDiagnosis()`, `extractGepaTranscripts()` | Zero fixture-driven coverage of the 3 fallback branches |\n| `security-bench.mjs` | `parseSecurityBenchMarkdown()` | Regex-based parser never fed a canned markdown fixture |\n| `redblue.mjs` | `buildUpstreamArgs()` | Only the `attack` branch is exercised; `init/run/patch/report` untested |\n| `drift-from-history.mjs` | `--baseline-key` path | Has a documented-but-unfixed exit-code bug (2 vs 3) with no test locking behavior |\n| `gepa.mjs` | `validate`/`render`/`analyze` ops | Only `genome` op tested |\n\n## Edge cases not covered\n- `_similarity.mjs`: zero-length vectors, negative cost inputs (`Math.log10` → NaN risk), all-zero weight overrides\n- `audit-trend.mjs`/`drift-from-history.mjs`: malformed baseline JSON, missing `composite` field, conflicting `--baseline-key`+`--baseline-file`, ignored `--baseline-since`\n- `oia-audit.mjs`: partial degradation (2-of-5 sub-audits down) — only all-degraded/all-healthy tested\n- Concurrency: no test simulates one of 5 parallel sub-audits timing out while others succeed\n\n## Missing error-handling tests\n- `mint.mjs` \"target already exists\" exit-2 guard\n- `gepa.mjs` file-not-found / invalid-JSON `process.exit(2)` branches\n- `evolve.mjs` exit code 99 (safety-disqualified) — zero coverage\n- `_darwin.mjs` subprocess spawn-error path (`npx`/`node` binary missing)\n- `_harness.mjs` `readBinMap()` returning `null` for malformed `package.json`\n\n## Integration gaps\n- `npx @claude-flow/cli memory {store,retrieve,list}` shell-outs — never verified against a real memory backend round-trip\n- `.swarm/router-parallel.jsonl` real writer (`recordPair`/`recordPairOutcome`) never actually invoked — only source-grepped\n- `npx metaharness` cache-install \"stale/partial/corrupted\" middle states — only the two extremes (fully missing / fully available via upstream bypass) are tested\n- OpenRouter real-judge path (`redblue run --patch`, `learn --run`) has no CI-runnable coverage at all\n\n## Suggested new tests (18 concrete scenarios with `node:test` skeletons)\nThe agent produced 18 ready-to-adapt skeletons targeting the gaps above — most notably #1 (threat-model severity bug regression test), #3 (mint path-traversal), #4 (drift-from-history flag precedence), and #12 (security-bench markdown parser). Several target functions (`parseSecurityBenchMarkdown`, `parseDurationMs`, `median`/`pctile`, `buildDiagnosis`) would need an `export` added — harmless since these files run as CLI scripts, not imports — to become unit-testable without full subprocess spawns.\n\nWant me to write these test files (starting with the threat-model severity fix + regression test, since that looks like a live bug) or just leave this as the analysis?\n",
+  "parsedOutput": {
+    "sections": [
+      {
+        "title": "Test Coverage Analysis — `plugins/ruflo-metaharness`",
+        "content": "\nNo formal test framework exists here — just 6 ad-hoc `test-*.mjs` smoke/E2E scripts covering ~34 plain `.mjs` files (~7,000 lines). Full report:\n\n",
+        "level": 2
+      },
+      {
+        "title": "Current Coverage Summary",
+        "content": "- **`test-graceful-degradation.mjs`** — drills ADR-150 rule #3 (registry unreachable → all scripts degrade cleanly) across 8 of the scripts. Misses `mint`, `similarity`, `redblue`, `learn`, `gepa`, `drift-from-history`, `router-parallel-analyze`.\n- **`test-mcp-tools.mjs`** — smoke-tests the 15 MCP tool handlers (envelope shape only for most; deeper checks for 4 tools).\n- **`test-parallel-pipeline.mjs`** — E2E for the router parallel-logging AND-gate, using a hand-synthesized JSONL fixture (not the real TS writer).\n- **`test-pipeline-roundtrip.mjs`** — deepest test: real `oia-audit --dry-run` chained through `audit-trend` → `drift-from-history`, 12 stages.\n- **`test-similarity.mjs`** — pure unit tests of `_similarity.mjs` internals + a frozen regression anchor.\n- **`test-with-openrouter.mjs`** — real-money E2E via GCP-secret-gated OpenRouter key + **upstream** `metaharness` CLI directly (bypasses this plugin's `mint.mjs` wrapper entirely).\n\n",
+        "level": 2
+      },
+      {
+        "title": "Highest-priority finding — likely a real bug",
+        "content": "`threat-model.mjs` defines its own local `SEVERITY_RANK = {clean,low,medium,high}` instead of importing the shared frozen table from `_harness.mjs` (which uses `info/warn/error/critical`). Since the two vocabularies don't match, `SEVERITY_RANK[worst] >= threshold` evaluates to `undefined >= N === false` whenever upstream reports `critical` — meaning `--fail-on high` **silently never fires** on a critical finding. No test catches this today.\n\n",
+        "level": 2
+      },
+      {
+        "title": "Untested functions/classes (selected)",
+        "content": "| File | Function | Why it matters |\n|---|---|---|\n| `mint.mjs` | `safetyChecks()` | Path-traversal refusals never exercised by any script in-repo |\n| `_invoke.mjs` | `parseTrailingJson()` | Exists specifically to fix a documented past bug (progress-line JSON shadowing); no regression test |\n| `_invoke.mjs` | `satisfiesTildeRange()` | Pre-release/malformed version strings unhandled by any test |\n| `evolve.mjs` | `buildDiagnosis()`, `extractGepaTranscripts()` | Zero fixture-driven coverage of the 3 fallback branches |\n| `security-bench.mjs` | `parseSecurityBenchMarkdown()` | Regex-based parser never fed a canned markdown fixture |\n| `redblue.mjs` | `buildUpstreamArgs()` | Only the `attack` branch is exercised; `init/run/patch/report` untested |\n| `drift-from-history.mjs` | `--baseline-key` path | Has a documented-but-unfixed exit-code bug (2 vs 3) with no test locking behavior |\n| `gepa.mjs` | `validate`/`render`/`analyze` ops | Only `genome` op tested |\n\n",
+        "level": 2
+      },
+      {
+        "title": "Edge cases not covered",
+        "content": "- `_similarity.mjs`: zero-length vectors, negative cost inputs (`Math.log10` → NaN risk), all-zero weight overrides\n- `audit-trend.mjs`/`drift-from-history.mjs`: malformed baseline JSON, missing `composite` field, conflicting `--baseline-key`+`--baseline-file`, ignored `--baseline-since`\n- `oia-audit.mjs`: partial degradation (2-of-5 sub-audits down) — only all-degraded/all-healthy tested\n- Concurrency: no test simulates one of 5 parallel sub-audits timing out while others succeed\n\n",
+        "level": 2
+      },
+      {
+        "title": "Missing error-handling tests",
+        "content": "- `mint.mjs` \"target already exists\" exit-2 guard\n- `gepa.mjs` file-not-found / invalid-JSON `process.exit(2)` branches\n- `evolve.mjs` exit code 99 (safety-disqualified) — zero coverage\n- `_darwin.mjs` subprocess spawn-error path (`npx`/`node` binary missing)\n- `_harness.mjs` `readBinMap()` returning `null` for malformed `package.json`\n\n",
+        "level": 2
+      },
+      {
+        "title": "Integration gaps",
+        "content": "- `npx @claude-flow/cli memory {store,retrieve,list}` shell-outs — never verified against a real memory backend round-trip\n- `.swarm/router-parallel.jsonl` real writer (`recordPair`/`recordPairOutcome`) never actually invoked — only source-grepped\n- `npx metaharness` cache-install \"stale/partial/corrupted\" middle states — only the two extremes (fully missing / fully available via upstream bypass) are tested\n- OpenRouter real-judge path (`redblue run --patch`, `learn --run`) has no CI-runnable coverage at all\n\n",
+        "level": 2
+      },
+      {
+        "title": "Suggested new tests (18 concrete scenarios with `node:test` skeletons)",
+        "content": "The agent produced 18 ready-to-adapt skeletons targeting the gaps above — most notably #1 (threat-model severity bug regression test), #3 (mint path-traversal), #4 (drift-from-history flag precedence), and #12 (security-bench markdown parser). Several target functions (`parseSecurityBenchMarkdown`, `parseDurationMs`, `median`/`pctile`, `buildDiagnosis`) would need an `export` added — harmless since these files run as CLI scripts, not imports — to become unit-testable without full subprocess spawns.\n\nWant me to write these test files (starting with the threat-model severity fix + regression test, since that looks like a live bug) or just leave this as the analysis?",
+        "level": 2
+      }
+    ],
+    "codeBlocks": []
+  },
+  "durationMs": 375976,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T15:53:19.261Z",
+  "executionId": "testgaps_1783612023285_buygpo"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783613599301_6f78cw_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T16:13:19.302Z] PROMPT
+============================================================
+Analyze test coverage and identify gaps:
+- Find untested functions and classes
+- Identify edge cases not covered
+- Suggest new test scenarios
+- Check for missing error handling tests
+- Identify integration test gaps
+
+For each gap, provide a test skeleton.
+
+## Instructions
+
+Analyze the codebase and provide your response following the format specified in the task.

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783613599301_6f78cw_result.log

@@ -0,0 +1,60 @@
+[2026-07-09T16:16:40.844Z] RESULT
+============================================================
+{
+  "success": true,
+  "output": "## Test Coverage Analysis — ruflo-metaharness Plugin\n\nThe plugin has 32 `.mjs` scripts (~6,900 lines) with no formal test runner — existing `test-*.mjs` files are manual smoke/integration drills, not unit tests.\n\n### 1. Untested functions/classes\n- **`_invoke.mjs`** shared plumbing (`parseTrailingJson`, `satisfiesTildeRange`, `classifyDegraded`, `findLocalPackageDir`, `ensureCachedInstall`) — no direct unit test.\n- **`threat-model.mjs:9`** has a *local* `SEVERITY_RANK` that's diverged from the shared one in `_harness.mjs` (missing `info`/`warn`/`error`/`critical`) — the same drift class fixed elsewhere in iter 63, but not here, and nothing catches it.\n- **`evolve.mjs`**: the entire `--diagnose` feature (`looksLikeGepaTranscript`, `extractGepaTranscripts`, `summarizeTraces`, `buildDiagnosis`, lines 146-274) is untested.\n- `evolve.mjs`/`security-bench.mjs` `safetyChecks()` bound validation, `audit-list.mjs:parseDurationMs()`, `security-bench.mjs:parseSecurityBenchMarkdown()` — all cheap pure-logic, all untested.\n- **`redblue.mjs`, `gepa.mjs`, `learn.mjs`, `mint.mjs`** — zero dedicated test coverage at any level.\n\n### 2. Edge cases not covered\nEmpty/zero-record paths (`drift-from-history` \"no records found\", `audit-list` empty namespace), malformed JSON in `audit-trend.mjs:loadRecord()` and `gepa.mjs:readJsonFile()`, missing-file argv guards, and CLI arg-parsing edge cases (invalid `--sandbox`/`--fail-on`/`--alert-on-worst` enums, `redblue` unknown subcommand, `mint` missing required flags) — none exercised, despite failing fast before any network call. Also: `OPENROUTER_API_KEY`-absent path for `redblue run --patch`, and partial-degradation races (2-of-5 subcalls degraded in `oia-audit.mjs:runAllParallel()`).\n\n### 3. Missing error-handling tests\n`audit-list.mjs` memory-command failures silently returning `[]`, `oia-audit.mjs:persist()` store-failure path, `redblue.mjs` report-JSON-parse failure, `ensureCachedInstall()`'s cache-dir-create-failed branch — all untested failure paths.\n\n### 4. Integration test gaps\nNo test ever goes through the actual `npx ruflo metaharness <sub>` CLI dispatcher or `commands/` slash-command layer — everything invokes `scripts/*.mjs` directly with `node`. `mint`, `redblue`, `gepa`, `learn` have no integration coverage at all, not even the ADR-150 graceful-degradation contract check that the other 8 skills get. `drift-from-history`'s `--baseline-since`/`--baseline-key` fast-paths are also untested.\n\n### 5. Test skeletons (Node's built-in `node:test`, matching the project's zero-dependency style)\n\n```js\n// tests/test-argv-validation.mjs\nimport { test } from 'node:test';\nimport assert from 'node:assert/strict';\nimport { spawnSync } from 'node:child_process';\n\ntest('evolve.mjs rejects out-of-range --generations', () => {\n  const r = spawnSync('node', ['scripts/evolve.mjs', '--generations', '51']);\n  assert.equal(r.status, 2);\n});\n\ntest('redblue.mjs rejects unknown subcommand', () => {\n  const r = spawnSync('node', ['scripts/redblue.mjs', 'bogus']);\n  assert.equal(r.status, 2);\n});\n```\n\n```js\n// tests/test-invoke.mjs\nimport { test } from 'node:test';\nimport assert from 'node:assert/strict';\nimport { parseTrailingJson, satisfiesTildeRange } from '../scripts/_invoke.mjs';\n\ntest('parseTrailingJson grabs the LAST JSON block, not a progress line', () => {\n  assert.deepEqual(parseTrailingJson('progress {\"x\":1}\\nresult {\"final\":true}'), { final: true });\n});\n```\n\n```js\n// tests/test-threat-model-severity-drift.mjs\nimport { test } from 'node:test';\nimport assert from 'node:assert/strict';\nimport { SEVERITY_RANK as SHARED } from '../scripts/_harness.mjs';\nimport { readFileSync } from 'node:fs';\n\ntest('threat-model.mjs should reuse the shared SEVERITY_RANK, not a local copy', () => {\n  const src = readFileSync('scripts/threat-model.mjs', 'utf-8');\n  assert.ok(Object.keys(SHARED).every(k => src.includes(k)));\n});\n```\n\nHighest-priority fix: the `threat-model.mjs` `SEVERITY_RANK` drift is a real correctness bug (silently ranks `info`/`warn`/`error`/`critical` as `undefined`), not just a coverage gap — worth fixing alongside adding the test. Want me to add `node:test`-based tests for the argv-validation and `_invoke.mjs` gaps, or fix the severity-rank drift first?\n",
+  "parsedOutput": {
+    "sections": [
+      {
+        "title": "Test Coverage Analysis — ruflo-metaharness Plugin",
+        "content": "\nThe plugin has 32 `.mjs` scripts (~6,900 lines) with no formal test runner — existing `test-*.mjs` files are manual smoke/integration drills, not unit tests.\n\n",
+        "level": 2
+      },
+      {
+        "title": "1. Untested functions/classes",
+        "content": "- **`_invoke.mjs`** shared plumbing (`parseTrailingJson`, `satisfiesTildeRange`, `classifyDegraded`, `findLocalPackageDir`, `ensureCachedInstall`) — no direct unit test.\n- **`threat-model.mjs:9`** has a *local* `SEVERITY_RANK` that's diverged from the shared one in `_harness.mjs` (missing `info`/`warn`/`error`/`critical`) — the same drift class fixed elsewhere in iter 63, but not here, and nothing catches it.\n- **`evolve.mjs`**: the entire `--diagnose` feature (`looksLikeGepaTranscript`, `extractGepaTranscripts`, `summarizeTraces`, `buildDiagnosis`, lines 146-274) is untested.\n- `evolve.mjs`/`security-bench.mjs` `safetyChecks()` bound validation, `audit-list.mjs:parseDurationMs()`, `security-bench.mjs:parseSecurityBenchMarkdown()` — all cheap pure-logic, all untested.\n- **`redblue.mjs`, `gepa.mjs`, `learn.mjs`, `mint.mjs`** — zero dedicated test coverage at any level.\n\n",
+        "level": 3
+      },
+      {
+        "title": "2. Edge cases not covered",
+        "content": "Empty/zero-record paths (`drift-from-history` \"no records found\", `audit-list` empty namespace), malformed JSON in `audit-trend.mjs:loadRecord()` and `gepa.mjs:readJsonFile()`, missing-file argv guards, and CLI arg-parsing edge cases (invalid `--sandbox`/`--fail-on`/`--alert-on-worst` enums, `redblue` unknown subcommand, `mint` missing required flags) — none exercised, despite failing fast before any network call. Also: `OPENROUTER_API_KEY`-absent path for `redblue run --patch`, and partial-degradation races (2-of-5 subcalls degraded in `oia-audit.mjs:runAllParallel()`).\n\n",
+        "level": 3
+      },
+      {
+        "title": "3. Missing error-handling tests",
+        "content": "`audit-list.mjs` memory-command failures silently returning `[]`, `oia-audit.mjs:persist()` store-failure path, `redblue.mjs` report-JSON-parse failure, `ensureCachedInstall()`'s cache-dir-create-failed branch — all untested failure paths.\n\n",
+        "level": 3
+      },
+      {
+        "title": "4. Integration test gaps",
+        "content": "No test ever goes through the actual `npx ruflo metaharness <sub>` CLI dispatcher or `commands/` slash-command layer — everything invokes `scripts/*.mjs` directly with `node`. `mint`, `redblue`, `gepa`, `learn` have no integration coverage at all, not even the ADR-150 graceful-degradation contract check that the other 8 skills get. `drift-from-history`'s `--baseline-since`/`--baseline-key` fast-paths are also untested.\n\n",
+        "level": 3
+      },
+      {
+        "title": "5. Test skeletons (Node's built-in `node:test`, matching the project's zero-dependency style)",
+        "content": "```js\n// tests/test-argv-validation.mjs\nimport { test } from 'node:test';\nimport assert from 'node:assert/strict';\nimport { spawnSync } from 'node:child_process';\n\ntest('evolve.mjs rejects out-of-range --generations', () => {\n  const r = spawnSync('node', ['scripts/evolve.mjs', '--generations', '51']);\n  assert.equal(r.status, 2);\n});\n\ntest('redblue.mjs rejects unknown subcommand', () => {\n  const r = spawnSync('node', ['scripts/redblue.mjs', 'bogus']);\n  assert.equal(r.status, 2);\n});\n```\n\n```js\n// tests/test-invoke.mjs\nimport { test } from 'node:test';\nimport assert from 'node:assert/strict';\nimport { parseTrailingJson, satisfiesTildeRange } from '../scripts/_invoke.mjs';\n\ntest('parseTrailingJson grabs the LAST JSON block, not a progress line', () => {\n  assert.deepEqual(parseTrailingJson('progress {\"x\":1}\\nresult {\"final\":true}'), { final: true });\n});\n```\n\n```js\n// tests/test-threat-model-severity-drift.mjs\nimport { test } from 'node:test';\nimport assert from 'node:assert/strict';\nimport { SEVERITY_RANK as SHARED } from '../scripts/_harness.mjs';\nimport { readFileSync } from 'node:fs';\n\ntest('threat-model.mjs should reuse the shared SEVERITY_RANK, not a local copy', () => {\n  const src = readFileSync('scripts/threat-model.mjs', 'utf-8');\n  assert.ok(Object.keys(SHARED).every(k => src.includes(k)));\n});\n```\n\nHighest-priority fix: the `threat-model.mjs` `SEVERITY_RANK` drift is a real correctness bug (silently ranks `info`/`warn`/`error`/`critical` as `undefined`), not just a coverage gap — worth fixing alongside adding the test. Want me to add `node:test`-based tests for the argv-validation and `_invoke.mjs` gaps, or fix the severity-rank drift first?",
+        "level": 3
+      }
+    ],
+    "codeBlocks": [
+      {
+        "language": "js",
+        "code": "// tests/test-argv-validation.mjs\nimport { test } from 'node:test';\nimport assert from 'node:assert/strict';\nimport { spawnSync } from 'node:child_process';\n\ntest('evolve.mjs rejects out-of-range --generations', () => {\n  const r = spawnSync('node', ['scripts/evolve.mjs', '--generations', '51']);\n  assert.equal(r.status, 2);\n});\n\ntest('redblue.mjs rejects unknown subcommand', () => {\n  const r = spawnSync('node', ['scripts/redblue.mjs', 'bogus']);\n  assert.equal(r.status, 2);\n});"
+      },
+      {
+        "language": "js",
+        "code": "// tests/test-invoke.mjs\nimport { test } from 'node:test';\nimport assert from 'node:assert/strict';\nimport { parseTrailingJson, satisfiesTildeRange } from '../scripts/_invoke.mjs';\n\ntest('parseTrailingJson grabs the LAST JSON block, not a progress line', () => {\n  assert.deepEqual(parseTrailingJson('progress {\"x\":1}\\nresult {\"final\":true}'), { final: true });\n});"
+      },
+      {
+        "language": "js",
+        "code": "// tests/test-threat-model-severity-drift.mjs\nimport { test } from 'node:test';\nimport assert from 'node:assert/strict';\nimport { SEVERITY_RANK as SHARED } from '../scripts/_harness.mjs';\nimport { readFileSync } from 'node:fs';\n\ntest('threat-model.mjs should reuse the shared SEVERITY_RANK, not a local copy', () => {\n  const src = readFileSync('scripts/threat-model.mjs', 'utf-8');\n  assert.ok(Object.keys(SHARED).every(k => src.includes(k)));\n});"
+      }
+    ]
+  },
+  "durationMs": 201543,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T16:16:40.844Z",
+  "executionId": "testgaps_1783613599301_6f78cw"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783615000844_v95ues_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T16:36:40.845Z] PROMPT
+============================================================
+Analyze test coverage and identify gaps:
+- Find untested functions and classes
+- Identify edge cases not covered
+- Suggest new test scenarios
+- Check for missing error handling tests
+- Identify integration test gaps
+
+For each gap, provide a test skeleton.
+
+## Instructions
+
+Analyze the codebase and provide your response following the format specified in the task.