@claude-flow/cli 3.38.12 → 3.38.14

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783615000844_v95ues_result.log

@@ -0,0 +1,69 @@
+[2026-07-09T16:39:01.688Z] RESULT
+============================================================
+{
+  "success": true,
+  "output": "Good, all claims verified against source. I have enough to write an accurate, verified report with skeletons matching this repo's actual testing convention (the custom `assert()`/`approx()` harness used in `test-similarity.mjs`, not `node:test`).\n\n## Test Coverage Analysis — `plugins/ruflo-metaharness`\n\nThe plugin has 34 `.mjs` scripts (~6,900 LOC), no test runner/framework — existing `test-*.mjs` files are self-contained scripts using a hand-rolled `assert()`/`approx()` counter (see `test-similarity.mjs`), invoked directly with `node`. Only two modules (`_similarity.mjs`, `_harness.mjs`) export functions and get true unit coverage; everything else is either integration-tested via `spawnSync` subprocess (`test-mcp-tools.mjs`, `test-parallel-pipeline.mjs`, `test-pipeline-roundtrip.mjs`, `test-graceful-degradation.mjs`, `test-with-openrouter.mjs`) or **not tested at all**.\n\n### 1. Correctness bug hiding behind the coverage gap (highest priority)\n\n**`threat-model.mjs:9`** — `const SEVERITY_RANK = { clean: 0, low: 1, medium: 2, high: 3 }` is a **local, stale copy**. `_harness.mjs:261` documents (iter 63) that this exact class of bug was fixed everywhere else by centralizing on `rankSeverity()`, which safely maps `critical → 4` and returns `0` for unknowns instead of `undefined`. `threat-model.mjs` never migrated:\n- Line 37: `SEVERITY_RANK[worst] >= threshold` — if the upstream harness ever reports `worst: \"critical\"`, this is `undefined >= threshold` → **always `false`** → a critical finding silently never triggers the alert/exit-1, the exact NaN-compare hazard the shared module exists to prevent.\n- Line 23-26 also rejects `--fail-on critical` as invalid input, even though the underlying tool can emit that severity.\n\nNo test exists to catch this because `threat-model.mjs` exports nothing.\n\n### 2. Untested modules (zero exports, zero coverage)\n\n`evolve.mjs`, `audit-list.mjs`, `oia-audit.mjs`, `mint.mjs`, `gepa.mjs`, `audit-trend.mjs`, `drift-from-history.mjs`, `security-bench.mjs`, `learn.mjs`, `redblue.mjs`, `threat-model.mjs` — confirmed via `grep '^export'`: **none** export their internal functions, so pure logic is untestable except by black-box CLI subprocess. Highest-value functions currently untestable in isolation:\n- `evolve.mjs`: `looksLikeGepaTranscript`, `extractGepaTranscripts`, `summarizeTraces`, `buildDiagnosis` (the entire `--diagnose` feature, lines 146–274) and `safetyChecks()` bound validation\n- `security-bench.mjs`: `parseSecurityBenchMarkdown()`, `safetyChecks()`\n- `audit-list.mjs`: `parseDurationMs()`\n- `audit-trend.mjs`: `loadRecord()` malformed-JSON / missing-file branches\n- `mint.mjs`: `safetyChecks()` — path-escape guards (`targetAbs === repoRoot`, `startsWith(repoRoot + '/')`) are exactly the kind of security-relevant logic that should never rely on subprocess-only testing\n\n### 3. Edge cases not covered\n\n- Malformed/missing JSON: `audit-trend.mjs:loadRecord()` (bad `--baseline`/`--current` file), `gepa.mjs:readJsonFile()`\n- Empty/zero-record paths: `drift-from-history.mjs` \"no records found\", `audit-list.mjs` empty namespace\n- CLI arg-parsing enums never exercised with invalid input: `--fail-on`, `--sandbox` (`evolve.mjs` only allows `real|mock|agent`), `--mutator`, `--selection`, `redblue` unknown subcommand (`redblue.mjs:182`), `mint` missing `--name`/`--template` (`mint.mjs:38-45`)\n- `mint.mjs` path-escape guard itself (target == repo root, target inside repo root) — untested despite being a safety-critical guard\n- `redblue.mjs` report-JSON parse failure at line 220-225 (upstream wrote a corrupt/truncated report file)\n\n### 4. Missing error-handling tests\n\n- `audit-list.mjs` memory-command failures silently returning `[]`/`null` on non-zero exit\n- `oia-audit.mjs:persist()` store-failure path\n- `redblue.mjs` corrupt report JSON (line 220-225, shown above)\n- `ensureCachedInstall()` in `_invoke.mjs` — cache-dir-create-failed branch\n\n### 5. Integration test gaps\n\nNothing drives the real `npx ruflo metaharness <sub>` CLI dispatcher or the `commands/` slash-command layer — every existing integration test 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 (`degraded: true` on `MODULE_NOT_FOUND`) that the other scripts get via `test-graceful-degradation.mjs`.\n\n---\n\n## Test skeletons\n\nFollowing this repo's actual convention (custom assert harness, not `node:test`):\n\n```js\n// scripts/test-threat-model-severity.mjs — regression test for the SEVERITY_RANK drift bug\n//\n// USAGE: node scripts/test-threat-model-severity.mjs\n\nimport { readFileSync } from 'node:fs';\nimport { spawnSync } from 'node:child_process';\nimport { SEVERITY_RANK as SHARED } from './_harness.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\n// (a) static check: threat-model.mjs must not carry its own stale rank map\nconst src = readFileSync(new URL('./threat-model.mjs', import.meta.url), 'utf-8');\nassert(\n  !/const SEVERITY_RANK = \\{/.test(src),\n  'threat-model.mjs has no local SEVERITY_RANK literal (uses shared rankSeverity instead)'\n);\n\n// (b) behavioral check: --fail-on critical must be accepted once fixed\nconst r = spawnSync('node', ['scripts/threat-model.mjs', '--fail-on', 'critical', '--path', '/nonexistent'], { encoding: 'utf-8' });\nassert(r.status !== 2 || !/must be one of clean\\|low\\|medium\\|high/.test(r.stderr || ''),\n  '--fail-on critical is accepted (not rejected as invalid enum)');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);\n```\n\n```js\n// scripts/test-mint-safety.mjs — unit tests for mint.mjs path-escape guards\n// Requires exporting `safetyChecks` from mint.mjs (currently module-private).\n//\n// USAGE: node scripts/test-mint-safety.mjs\n\nimport { spawnSync } from 'node:child_process';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\nconst cwd = process.cwd();\n\n// missing --name\nlet r = spawnSync('node', ['scripts/mint.mjs', '--template', 'minimal'], { encoding: 'utf-8' });\nassert(r.status === 2 && /--name is required/.test(r.stderr), 'rejects missing --name');\n\n// missing --template\nr = spawnSync('node', ['scripts/mint.mjs', '--name', 'foo'], { encoding: 'utf-8' });\nassert(r.status === 2 && /--template is required/.test(r.stderr), 'rejects missing --template');\n\n// refuses project root as target\nr = spawnSync('node', ['scripts/mint.mjs', '--name', 'foo', '--template', 'minimal', '--target', cwd], { encoding: 'utf-8' });\nassert(r.status === 2 && /refusing to write to project root/.test(r.stderr), 'refuses target === repo root');\n\n// refuses a path inside the repo root\nr = spawnSync('node', ['scripts/mint.mjs', '--name', 'foo', '--template', 'minimal', '--target', `${cwd}/subdir`], { encoding: 'utf-8' });\nassert(r.status === 2 && /refusing to write inside the calling repo root/.test(r.stderr), 'refuses target inside repo root');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);\n```\n\n```js\n// scripts/test-evolve-diagnose.mjs — unit tests for evolve.mjs's --diagnose helpers\n// Requires exporting looksLikeGepaTranscript/extractGepaTranscripts/summarizeTraces\n// from evolve.mjs (currently module-private, untestable in isolation today).\n//\n// USAGE: node scripts/test-evolve-diagnose.mjs\n\nimport { looksLikeGepaTranscript, extractGepaTranscripts, summarizeTraces } from './evolve.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\nassert(looksLikeGepaTranscript([{ actionRaw: 'x' }]) === true, 'accepts actionRaw-shaped entries');\nassert(looksLikeGepaTranscript([{ obs: 'y' }]) === true, 'accepts obs-shaped entries');\nassert(looksLikeGepaTranscript([]) === false, 'rejects empty array');\nassert(looksLikeGepaTranscript([{ foo: 1 }]) === false, 'rejects entries missing actionRaw/obs');\nassert(looksLikeGepaTranscript(null) === false, 'rejects non-array input without throwing');\n\nassert(\n  extractGepaTranscripts({ transcript: [{ actionRaw: 'x' }] }).length === 1,\n  'extracts record.transcript when gepa-shaped'\n);\nassert(\n  extractGepaTranscripts({ traces: [{ taskId: 't1', transcript: [{ obs: 'y' }] }] }).length === 1,\n  'extracts per-trace transcript with taskId preserved'\n);\nassert(\n  extractGepaTranscripts({}).length === 0,\n  'returns empty array, not throw, when no transcript-shaped data exists'\n);\n\nconst summary = summarizeTraces({ traces: [{ exitCode: 1 }, { exitCode: 0, timedOut: true }, { exitCode: 0, blockedActions: ['a', 'b'] }] });\nassert(summary.tasks === 3, 'counts total tasks');\nassert(summary.failed === 1, 'counts non-zero exitCode as failed');\nassert(summary.timedOut === 1, 'counts timedOut entries');\nassert(summary.blockedActions === 2, 'sums blockedActions across traces');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);\n```\n\n```js\n// scripts/test-redblue-argv.mjs — CLI arg-parsing edge cases (no `export` needed, pure subprocess)\n//\n// USAGE: node scripts/test-redblue-argv.mjs\n\nimport { spawnSync } from 'node:child_process';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\nlet r = spawnSync('node', ['scripts/redblue.mjs', 'bogus-subcommand'], { encoding: 'utf-8' });\nassert(r.status === 2 && /unknown subcommand/.test(r.stderr), 'rejects unknown subcommand with exit 2');\n\nr = spawnSync('node', ['scripts/redblue.mjs'], { encoding: 'utf-8' });\nassert(r.status === 0, 'no subcommand prints help and exits 0');\n\nr = spawnSync('node', ['scripts/redblue.mjs', '--help'], { encoding: 'utf-8' });\nassert(r.status === 0 && /Subcommands:/.test(r.stdout), '--help prints usage');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);\n```\n\n---\n\n**Priority order to actually fix, not just cover:**\n1. `threat-model.mjs` `SEVERITY_RANK` drift — this is a live correctness bug (critical findings never trigger `--fail-on`), fix by importing `rankSeverity`/`SEVERITY_RANK` from `_harness.mjs` and widening the `--fail-on` enum to include `critical`.\n2. Export the pure-logic functions in `evolve.mjs`, `mint.mjs`, `security-bench.mjs`, `audit-list.mjs`, `audit-trend.mjs` so they're unit-testable without subprocess overhead — currently every one of these 8+ modules is untestable in isolation by construction.\n3. Then land the skeletons above plus graceful-degradation integration coverage for `mint`/`redblue`/`gepa`/`learn` to match the other 8 scripts.\n\nI found a pre-existing `testgaps` worker run in `.claude-flow/metrics/test-gaps.json` covering similar ground — I independently verified its core claims against the source (confirmed the `SEVERITY_RANK` bug and the zero-export pattern are real) and corrected its test skeletons to match this repo's actual convention (custom assert harness, not `node:test`, which isn't used anywhere in this plugin).\n",
+  "parsedOutput": {
+    "sections": [
+      {
+        "title": "Test Coverage Analysis — `plugins/ruflo-metaharness`",
+        "content": "\nThe plugin has 34 `.mjs` scripts (~6,900 LOC), no test runner/framework — existing `test-*.mjs` files are self-contained scripts using a hand-rolled `assert()`/`approx()` counter (see `test-similarity.mjs`), invoked directly with `node`. Only two modules (`_similarity.mjs`, `_harness.mjs`) export functions and get true unit coverage; everything else is either integration-tested via `spawnSync` subprocess (`test-mcp-tools.mjs`, `test-parallel-pipeline.mjs`, `test-pipeline-roundtrip.mjs`, `test-graceful-degradation.mjs`, `test-with-openrouter.mjs`) or **not tested at all**.\n\n",
+        "level": 2
+      },
+      {
+        "title": "1. Correctness bug hiding behind the coverage gap (highest priority)",
+        "content": "\n**`threat-model.mjs:9`** — `const SEVERITY_RANK = { clean: 0, low: 1, medium: 2, high: 3 }` is a **local, stale copy**. `_harness.mjs:261` documents (iter 63) that this exact class of bug was fixed everywhere else by centralizing on `rankSeverity()`, which safely maps `critical → 4` and returns `0` for unknowns instead of `undefined`. `threat-model.mjs` never migrated:\n- Line 37: `SEVERITY_RANK[worst] >= threshold` — if the upstream harness ever reports `worst: \"critical\"`, this is `undefined >= threshold` → **always `false`** → a critical finding silently never triggers the alert/exit-1, the exact NaN-compare hazard the shared module exists to prevent.\n- Line 23-26 also rejects `--fail-on critical` as invalid input, even though the underlying tool can emit that severity.\n\nNo test exists to catch this because `threat-model.mjs` exports nothing.\n\n",
+        "level": 3
+      },
+      {
+        "title": "2. Untested modules (zero exports, zero coverage)",
+        "content": "\n`evolve.mjs`, `audit-list.mjs`, `oia-audit.mjs`, `mint.mjs`, `gepa.mjs`, `audit-trend.mjs`, `drift-from-history.mjs`, `security-bench.mjs`, `learn.mjs`, `redblue.mjs`, `threat-model.mjs` — confirmed via `grep '^export'`: **none** export their internal functions, so pure logic is untestable except by black-box CLI subprocess. Highest-value functions currently untestable in isolation:\n- `evolve.mjs`: `looksLikeGepaTranscript`, `extractGepaTranscripts`, `summarizeTraces`, `buildDiagnosis` (the entire `--diagnose` feature, lines 146–274) and `safetyChecks()` bound validation\n- `security-bench.mjs`: `parseSecurityBenchMarkdown()`, `safetyChecks()`\n- `audit-list.mjs`: `parseDurationMs()`\n- `audit-trend.mjs`: `loadRecord()` malformed-JSON / missing-file branches\n- `mint.mjs`: `safetyChecks()` — path-escape guards (`targetAbs === repoRoot`, `startsWith(repoRoot + '/')`) are exactly the kind of security-relevant logic that should never rely on subprocess-only testing\n\n",
+        "level": 3
+      },
+      {
+        "title": "3. Edge cases not covered",
+        "content": "\n- Malformed/missing JSON: `audit-trend.mjs:loadRecord()` (bad `--baseline`/`--current` file), `gepa.mjs:readJsonFile()`\n- Empty/zero-record paths: `drift-from-history.mjs` \"no records found\", `audit-list.mjs` empty namespace\n- CLI arg-parsing enums never exercised with invalid input: `--fail-on`, `--sandbox` (`evolve.mjs` only allows `real|mock|agent`), `--mutator`, `--selection`, `redblue` unknown subcommand (`redblue.mjs:182`), `mint` missing `--name`/`--template` (`mint.mjs:38-45`)\n- `mint.mjs` path-escape guard itself (target == repo root, target inside repo root) — untested despite being a safety-critical guard\n- `redblue.mjs` report-JSON parse failure at line 220-225 (upstream wrote a corrupt/truncated report file)\n\n",
+        "level": 3
+      },
+      {
+        "title": "4. Missing error-handling tests",
+        "content": "\n- `audit-list.mjs` memory-command failures silently returning `[]`/`null` on non-zero exit\n- `oia-audit.mjs:persist()` store-failure path\n- `redblue.mjs` corrupt report JSON (line 220-225, shown above)\n- `ensureCachedInstall()` in `_invoke.mjs` — cache-dir-create-failed branch\n\n",
+        "level": 3
+      },
+      {
+        "title": "5. Integration test gaps",
+        "content": "\nNothing drives the real `npx ruflo metaharness <sub>` CLI dispatcher or the `commands/` slash-command layer — every existing integration test 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 (`degraded: true` on `MODULE_NOT_FOUND`) that the other scripts get via `test-graceful-degradation.mjs`.\n\n---\n\n",
+        "level": 3
+      },
+      {
+        "title": "Test skeletons",
+        "content": "Following this repo's actual convention (custom assert harness, not `node:test`):\n\n```js\n// scripts/test-threat-model-severity.mjs — regression test for the SEVERITY_RANK drift bug\n//\n// USAGE: node scripts/test-threat-model-severity.mjs\n\nimport { readFileSync } from 'node:fs';\nimport { spawnSync } from 'node:child_process';\nimport { SEVERITY_RANK as SHARED } from './_harness.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\n// (a) static check: threat-model.mjs must not carry its own stale rank map\nconst src = readFileSync(new URL('./threat-model.mjs', import.meta.url), 'utf-8');\nassert(\n  !/const SEVERITY_RANK = \\{/.test(src),\n  'threat-model.mjs has no local SEVERITY_RANK literal (uses shared rankSeverity instead)'\n);\n\n// (b) behavioral check: --fail-on critical must be accepted once fixed\nconst r = spawnSync('node', ['scripts/threat-model.mjs', '--fail-on', 'critical', '--path', '/nonexistent'], { encoding: 'utf-8' });\nassert(r.status !== 2 || !/must be one of clean\\|low\\|medium\\|high/.test(r.stderr || ''),\n  '--fail-on critical is accepted (not rejected as invalid enum)');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);\n```\n\n```js\n// scripts/test-mint-safety.mjs — unit tests for mint.mjs path-escape guards\n// Requires exporting `safetyChecks` from mint.mjs (currently module-private).\n//\n// USAGE: node scripts/test-mint-safety.mjs\n\nimport { spawnSync } from 'node:child_process';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\nconst cwd = process.cwd();\n\n// missing --name\nlet r = spawnSync('node', ['scripts/mint.mjs', '--template', 'minimal'], { encoding: 'utf-8' });\nassert(r.status === 2 && /--name is required/.test(r.stderr), 'rejects missing --name');\n\n// missing --template\nr = spawnSync('node', ['scripts/mint.mjs', '--name', 'foo'], { encoding: 'utf-8' });\nassert(r.status === 2 && /--template is required/.test(r.stderr), 'rejects missing --template');\n\n// refuses project root as target\nr = spawnSync('node', ['scripts/mint.mjs', '--name', 'foo', '--template', 'minimal', '--target', cwd], { encoding: 'utf-8' });\nassert(r.status === 2 && /refusing to write to project root/.test(r.stderr), 'refuses target === repo root');\n\n// refuses a path inside the repo root\nr = spawnSync('node', ['scripts/mint.mjs', '--name', 'foo', '--template', 'minimal', '--target', `${cwd}/subdir`], { encoding: 'utf-8' });\nassert(r.status === 2 && /refusing to write inside the calling repo root/.test(r.stderr), 'refuses target inside repo root');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);\n```\n\n```js\n// scripts/test-evolve-diagnose.mjs — unit tests for evolve.mjs's --diagnose helpers\n// Requires exporting looksLikeGepaTranscript/extractGepaTranscripts/summarizeTraces\n// from evolve.mjs (currently module-private, untestable in isolation today).\n//\n// USAGE: node scripts/test-evolve-diagnose.mjs\n\nimport { looksLikeGepaTranscript, extractGepaTranscripts, summarizeTraces } from './evolve.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\nassert(looksLikeGepaTranscript([{ actionRaw: 'x' }]) === true, 'accepts actionRaw-shaped entries');\nassert(looksLikeGepaTranscript([{ obs: 'y' }]) === true, 'accepts obs-shaped entries');\nassert(looksLikeGepaTranscript([]) === false, 'rejects empty array');\nassert(looksLikeGepaTranscript([{ foo: 1 }]) === false, 'rejects entries missing actionRaw/obs');\nassert(looksLikeGepaTranscript(null) === false, 'rejects non-array input without throwing');\n\nassert(\n  extractGepaTranscripts({ transcript: [{ actionRaw: 'x' }] }).length === 1,\n  'extracts record.transcript when gepa-shaped'\n);\nassert(\n  extractGepaTranscripts({ traces: [{ taskId: 't1', transcript: [{ obs: 'y' }] }] }).length === 1,\n  'extracts per-trace transcript with taskId preserved'\n);\nassert(\n  extractGepaTranscripts({}).length === 0,\n  'returns empty array, not throw, when no transcript-shaped data exists'\n);\n\nconst summary = summarizeTraces({ traces: [{ exitCode: 1 }, { exitCode: 0, timedOut: true }, { exitCode: 0, blockedActions: ['a', 'b'] }] });\nassert(summary.tasks === 3, 'counts total tasks');\nassert(summary.failed === 1, 'counts non-zero exitCode as failed');\nassert(summary.timedOut === 1, 'counts timedOut entries');\nassert(summary.blockedActions === 2, 'sums blockedActions across traces');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);\n```\n\n```js\n// scripts/test-redblue-argv.mjs — CLI arg-parsing edge cases (no `export` needed, pure subprocess)\n//\n// USAGE: node scripts/test-redblue-argv.mjs\n\nimport { spawnSync } from 'node:child_process';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\nlet r = spawnSync('node', ['scripts/redblue.mjs', 'bogus-subcommand'], { encoding: 'utf-8' });\nassert(r.status === 2 && /unknown subcommand/.test(r.stderr), 'rejects unknown subcommand with exit 2');\n\nr = spawnSync('node', ['scripts/redblue.mjs'], { encoding: 'utf-8' });\nassert(r.status === 0, 'no subcommand prints help and exits 0');\n\nr = spawnSync('node', ['scripts/redblue.mjs', '--help'], { encoding: 'utf-8' });\nassert(r.status === 0 && /Subcommands:/.test(r.stdout), '--help prints usage');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);\n```\n\n---\n\n**Priority order to actually fix, not just cover:**\n1. `threat-model.mjs` `SEVERITY_RANK` drift — this is a live correctness bug (critical findings never trigger `--fail-on`), fix by importing `rankSeverity`/`SEVERITY_RANK` from `_harness.mjs` and widening the `--fail-on` enum to include `critical`.\n2. Export the pure-logic functions in `evolve.mjs`, `mint.mjs`, `security-bench.mjs`, `audit-list.mjs`, `audit-trend.mjs` so they're unit-testable without subprocess overhead — currently every one of these 8+ modules is untestable in isolation by construction.\n3. Then land the skeletons above plus graceful-degradation integration coverage for `mint`/`redblue`/`gepa`/`learn` to match the other 8 scripts.\n\nI found a pre-existing `testgaps` worker run in `.claude-flow/metrics/test-gaps.json` covering similar ground — I independently verified its core claims against the source (confirmed the `SEVERITY_RANK` bug and the zero-export pattern are real) and corrected its test skeletons to match this repo's actual convention (custom assert harness, not `node:test`, which isn't used anywhere in this plugin).",
+        "level": 2
+      }
+    ],
+    "codeBlocks": [
+      {
+        "language": "js",
+        "code": "// scripts/test-threat-model-severity.mjs — regression test for the SEVERITY_RANK drift bug\n//\n// USAGE: node scripts/test-threat-model-severity.mjs\n\nimport { readFileSync } from 'node:fs';\nimport { spawnSync } from 'node:child_process';\nimport { SEVERITY_RANK as SHARED } from './_harness.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\n// (a) static check: threat-model.mjs must not carry its own stale rank map\nconst src = readFileSync(new URL('./threat-model.mjs', import.meta.url), 'utf-8');\nassert(\n  !/const SEVERITY_RANK = \\{/.test(src),\n  'threat-model.mjs has no local SEVERITY_RANK literal (uses shared rankSeverity instead)'\n);\n\n// (b) behavioral check: --fail-on critical must be accepted once fixed\nconst r = spawnSync('node', ['scripts/threat-model.mjs', '--fail-on', 'critical', '--path', '/nonexistent'], { encoding: 'utf-8' });\nassert(r.status !== 2 || !/must be one of clean\\|low\\|medium\\|high/.test(r.stderr || ''),\n  '--fail-on critical is accepted (not rejected as invalid enum)');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);"
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-mint-safety.mjs — unit tests for mint.mjs path-escape guards\n// Requires exporting `safetyChecks` from mint.mjs (currently module-private).\n//\n// USAGE: node scripts/test-mint-safety.mjs\n\nimport { spawnSync } from 'node:child_process';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\nconst cwd = process.cwd();\n\n// missing --name\nlet r = spawnSync('node', ['scripts/mint.mjs', '--template', 'minimal'], { encoding: 'utf-8' });\nassert(r.status === 2 && /--name is required/.test(r.stderr), 'rejects missing --name');\n\n// missing --template\nr = spawnSync('node', ['scripts/mint.mjs', '--name', 'foo'], { encoding: 'utf-8' });\nassert(r.status === 2 && /--template is required/.test(r.stderr), 'rejects missing --template');\n\n// refuses project root as target\nr = spawnSync('node', ['scripts/mint.mjs', '--name', 'foo', '--template', 'minimal', '--target', cwd], { encoding: 'utf-8' });\nassert(r.status === 2 && /refusing to write to project root/.test(r.stderr), 'refuses target === repo root');\n\n// refuses a path inside the repo root\nr = spawnSync('node', ['scripts/mint.mjs', '--name', 'foo', '--template', 'minimal', '--target', `${cwd}/subdir`], { encoding: 'utf-8' });\nassert(r.status === 2 && /refusing to write inside the calling repo root/.test(r.stderr), 'refuses target inside repo root');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);"
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-evolve-diagnose.mjs — unit tests for evolve.mjs's --diagnose helpers\n// Requires exporting looksLikeGepaTranscript/extractGepaTranscripts/summarizeTraces\n// from evolve.mjs (currently module-private, untestable in isolation today).\n//\n// USAGE: node scripts/test-evolve-diagnose.mjs\n\nimport { looksLikeGepaTranscript, extractGepaTranscripts, summarizeTraces } from './evolve.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\nassert(looksLikeGepaTranscript([{ actionRaw: 'x' }]) === true, 'accepts actionRaw-shaped entries');\nassert(looksLikeGepaTranscript([{ obs: 'y' }]) === true, 'accepts obs-shaped entries');\nassert(looksLikeGepaTranscript([]) === false, 'rejects empty array');\nassert(looksLikeGepaTranscript([{ foo: 1 }]) === false, 'rejects entries missing actionRaw/obs');\nassert(looksLikeGepaTranscript(null) === false, 'rejects non-array input without throwing');\n\nassert(\n  extractGepaTranscripts({ transcript: [{ actionRaw: 'x' }] }).length === 1,\n  'extracts record.transcript when gepa-shaped'\n);\nassert(\n  extractGepaTranscripts({ traces: [{ taskId: 't1', transcript: [{ obs: 'y' }] }] }).length === 1,\n  'extracts per-trace transcript with taskId preserved'\n);\nassert(\n  extractGepaTranscripts({}).length === 0,\n  'returns empty array, not throw, when no transcript-shaped data exists'\n);\n\nconst summary = summarizeTraces({ traces: [{ exitCode: 1 }, { exitCode: 0, timedOut: true }, { exitCode: 0, blockedActions: ['a', 'b'] }] });\nassert(summary.tasks === 3, 'counts total tasks');\nassert(summary.failed === 1, 'counts non-zero exitCode as failed');\nassert(summary.timedOut === 1, 'counts timedOut entries');\nassert(summary.blockedActions === 2, 'sums blockedActions across traces');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);"
+      },
+      {
+        "language": "js",
+        "code": "// scripts/test-redblue-argv.mjs — CLI arg-parsing edge cases (no `export` needed, pure subprocess)\n//\n// USAGE: node scripts/test-redblue-argv.mjs\n\nimport { spawnSync } from 'node:child_process';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failed++; }\n}\n\nlet r = spawnSync('node', ['scripts/redblue.mjs', 'bogus-subcommand'], { encoding: 'utf-8' });\nassert(r.status === 2 && /unknown subcommand/.test(r.stderr), 'rejects unknown subcommand with exit 2');\n\nr = spawnSync('node', ['scripts/redblue.mjs'], { encoding: 'utf-8' });\nassert(r.status === 0, 'no subcommand prints help and exits 0');\n\nr = spawnSync('node', ['scripts/redblue.mjs', '--help'], { encoding: 'utf-8' });\nassert(r.status === 0 && /Subcommands:/.test(r.stdout), '--help prints usage');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) process.exit(1);"
+      }
+    ]
+  },
+  "durationMs": 140844,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T16:39:01.688Z",
+  "executionId": "testgaps_1783615000844_v95ues"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783616341693_bl5d9o_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T16:59:01.693Z] 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_1783616341693_bl5d9o_result.log

@@ -0,0 +1,64 @@
+[2026-07-09T17:03:52.819Z] RESULT
+============================================================
+{
+  "success": true,
+  "output": "## Test Coverage Analysis — `ruflo-metaharness`\n\nThe plugin has no formal test framework — coverage comes from 6 ad-hoc `test-*.mjs` scripts (real `assert()`/exit-code based) plus `scripts/smoke.sh` (123 grep-based structural checks, not runtime tests). Of 34 scripts, **6 have zero test coverage** and the most load-bearing shared module (`_invoke.mjs`) is never unit-tested at all.\n\n### 1. Untested functions/classes (no coverage in any of the 6 test files)\n\n| File | Status |\n|---|---|\n| `_invoke.mjs` | **Never unit-tested.** `findLocalPackageDir`, `satisfiesTildeRange`, `ensureCachedInstall`, `importOptionalLibrary`, `classifyDegraded`, `injectJson`, `parseTrailingJson` — only touched transitively via subprocess drills that force the single \"not-available\" branch |\n| `_darwin.mjs` | No unit test. `runDarwinAsync`'s `onProgress`/`signal` cancellation path is completely unexercised |\n| `_redblue.mjs` | No unit test at all (not even in the degradation drill) |\n| `mint.mjs` | Zero test coverage — including its root/self-repo write-refusal safety check |\n| `audit-list.mjs` | Zero test coverage |\n| `redblue.mjs`, `learn.mjs`, `gepa.mjs` | Zero test coverage — missing even from `test-graceful-degradation.mjs`'s skill list |\n| `router-parallel-analyze.mjs` | CLI tested end-to-end, but internal `median()`, `pctile()`, `mean()` never unit-tested directly (no verification of interpolation, even-length arrays, single-element arrays) |\n| `threat-model.mjs` | Has its own **stale, divergent** `SEVERITY_RANK` literal (missing `info/warn/error/critical` keys present in the shared `_harness.mjs` version) — this drift is invisible to every existing test |\n\n### 2. Edge cases not covered\n\n- `_invoke.mjs::satisfiesTildeRange` — no test for cross-major mismatch (`~0.3.0` vs `1.0.0`), pre-release suffixes (`0.3.0-beta.1`), or malformed version strings\n- `_invoke.mjs::parseTrailingJson` — the greedy whole-span fallback branch (used when no clean trailing `{...}` block exists) is never exercised\n- `_invoke.mjs::classifyDegraded` — `null` exitCode (timeout) vs `DEGRADED_RX` stderr match vs healthy are three distinct branches; only the stderr-match path is drilled\n- `_similarity.mjs::perDimensionBreakdown()` — only key-presence is checked in `test-similarity.mjs`, not the actual per-dimension numeric contribution math\n- `_similarity.mjs::similarity()` — no test with genuinely malformed input (`null`, missing `score`/`genome` keys entirely, non-object)\n- `oia-audit.mjs` composite worst-severity — no test of tie-breaking when two components report equal severity from different sources\n- `mint.mjs` — the upstream `--target`-ignored workaround (cwd/basename trick) has no regression test; a re-fix upstream could silently break it\n- `evolve.mjs --diagnose` — `buildDiagnosis`/`extractGepaTranscripts`/`looksLikeGepaTranscript` transcript-classification heuristics have no fixture-based test\n\n### 3. Missing error handling tests\n\n- **`threat-model.mjs`**: an unrecognized severity value (e.g. upstream adds a `\"warn\"` finding) falls through the local `SEVERITY_RANK` to `undefined`, producing a `NaN` comparison in the `--fail-on` gate instead of failing safe. No test catches this.\n- **`_invoke.mjs::ensureCachedInstall`**: no test of the `npm install` subprocess itself failing (disk full, permission denied, corrupt cache dir) as distinct from \"package unreachable\"\n- **`redblue.mjs`**: no test of the upstream `isMain` npx-symlink workaround failing to resolve a real path (falls back to what?)\n- **`test-graceful-degradation.mjs`** itself documents (in comments) two known-but-unfixed bugs — `mint.mjs`'s degraded payload possibly missing the `\"degraded\": true` marker, and `drift-from-history.mjs` returning exit 2 instead of documented exit 3 on upstream-absent — neither has a regression test asserting the *correct* behavior once fixed\n- **`gepa.mjs`**: no test of `analyzeTranscript` receiving a non-GEPA-shaped transcript file (malformed JSON, wrong schema)\n\n### 4. Integration test gaps\n\n- **`redblue.mjs`, `learn.mjs`, `gepa.mjs`, `mint.mjs`, `audit-list.mjs`** are absent from `test-graceful-degradation.mjs`'s subprocess drill — none of the \"does this script behave correctly when upstream is completely unreachable\" guarantees are verified for them\n- **`test-mcp-tools.mjs` silently no-ops** or when the sibling monorepo's compiled `dist/` isn't present (plugin-only checkout) — this means the 15-tool MCP handler contract goes completely unverified in that environment, with `exit 0` masking zero assertions run\n- **`_darwin.mjs`/`_redblue.mjs` \"already cached\" and \"local install found\" resolution paths** are never exercised — only \"must install fresh\" and \"unreachable\" are covered\n- **`audit-trend.mjs` ↔ `_similarity.mjs` composition** is only tested via `test-pipeline-roundtrip.mjs`'s self-match/synthetic-mutation cases — no test of genuinely divergent real-world audit pairs (e.g. different repo entirely, comparing incompatible genomes)\n- **`test-with-openrouter.mjs`** bypasses `_harness.mjs`'s pinned-version resolution entirely (uses `metaharness@latest`) — it looks like an integration test of the harness CLI surface but never actually validates the plugin's own caching/pin logic under real network conditions\n\n---\n\n### Test skeletons\n\n**A. `_invoke.mjs` unit tests (highest-priority gap — currently zero coverage)**\n\n```javascript\n#!/usr/bin/env node\n// test-invoke.mjs — unit tests for the shared subprocess/plumbing layer\n// (_invoke.mjs). Currently untested despite being imported by\n// _harness.mjs, _darwin.mjs, _redblue.mjs, and gepa.mjs.\n//\n// USAGE\n//   node scripts/test-invoke.mjs\n// EXIT\n//   0 all pass, 1 at least one failure\n\nimport {\n  classifyDegraded, injectJson, parseTrailingJson,\n  satisfiesTildeRange, cacheBaseDir,\n} from './_invoke.mjs';\n\nlet passed = 0, failed = 0;\nconst failures = [];\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failures.push(label); failed++; }\n}\n\n// ── classifyDegraded ──────────────────────────────────────────────\n{\n  const timeout = classifyDegraded('', null, 'harness');\n  assert(timeout.degraded === true, 'classifyDegraded: null exitCode (timeout) is degraded');\n\n  const notFound = classifyDegraded('npm ERR! 404 Not Found', 1, 'harness');\n  assert(notFound.degraded === true, 'classifyDegraded: DEGRADED_RX stderr match is degraded');\n\n  const healthy = classifyDegraded('', 0, 'harness');\n  assert(healthy.degraded === false, 'classifyDegraded: clean exit is not degraded');\n\n  const nonzeroButHealthy = classifyDegraded('some real validation error', 1, 'harness');\n  assert(nonzeroButHealthy.degraded === false,\n    'classifyDegraded: nonzero exit without DEGRADED_RX match is NOT treated as degraded');\n}\n\n// ── injectJson ─────────────────────────────────────────────────────\n{\n  assert(injectJson(['score', '.'], true).includes('--json'),\n    'injectJson: appends --json when wantJson=true and absent');\n  assert(!injectJson(['score', '.'], false).includes('--json'),\n    'injectJson: does not append when wantJson=false');\n  const already = injectJson(['score', '.', '--json'], true);\n  assert(already.filter(a => a === '--json').length === 1,\n    'injectJson: does not duplicate --json if already present');\n}\n\n// ── parseTrailingJson ────────────────────────────────────────────\n{\n  const clean = parseTrailingJson('some log line\\n{\"a\":1}');\n  assert(clean && clean.a === 1, 'parseTrailingJson: extracts trailing JSON block');\n\n  const noisy = parseTrailingJson('{\"a\":1}\\nnot json trailing garbage');\n  assert(noisy === null || typeof noisy === 'object',\n    'parseTrailingJson: gracefully handles non-JSON trailing content (greedy fallback)');\n\n  assert(parseTrailingJson('') === null, 'parseTrailingJson: empty string returns null, not throw');\n  assert(parseTrailingJson('no json here at all') === null,\n    'parseTrailingJson: no JSON present returns null');\n}\n\n// ── satisfiesTildeRange ──────────────────────────────────────────\n{\n  assert(satisfiesTildeRange('0.3.2', '~0.3.0') === true,\n    'satisfiesTildeRange: same major.minor, patch >= pin is satisfied');\n  assert(satisfiesTildeRange('0.3.0', '~0.3.0') === true,\n    'satisfiesTildeRange: exact match is satisfied');\n  assert(satisfiesTildeRange('0.4.0', '~0.3.0') === false,\n    'satisfiesTildeRange: different minor is NOT satisfied');\n  assert(satisfiesTildeRange('1.3.0', '~0.3.0') === false,\n    'satisfiesTildeRange: different major is NOT satisfied');\n  assert(satisfiesTildeRange('0.3.0-beta.1', '~0.3.0') === false,\n    'satisfiesTildeRange: pre-release suffix does not satisfy plain pin');\n  assert(satisfiesTildeRange('not-a-version', '~0.3.0') === false,\n    'satisfiesTildeRange: malformed version string fails closed, not throw');\n}\n\n// ── cacheBaseDir ──────────────────────────────────────────────────\n{\n  const original = process.env.RUFLO_METAHARNESS_CACHE_BASE;\n  delete process.env.RUFLO_METAHARNESS_CACHE_BASE;\n  assert(cacheBaseDir().includes('.ruflo'), 'cacheBaseDir: defaults under ~/.ruflo');\n  process.env.RUFLO_METAHARNESS_CACHE_BASE = '/tmp/custom-cache';\n  assert(cacheBaseDir() === '/tmp/custom-cache', 'cacheBaseDir: honors env override test seam');\n  if (original === undefined) delete process.env.RUFLO_METAHARNESS_CACHE_BASE;\n  else process.env.RUFLO_METAHARNESS_CACHE_BASE = original;\n}\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) { console.log('Failures:', failures.join(', ')); process.exit(1); }\nprocess.exit(0);\n```\n\n**B. `threat-model.mjs` — regression test for the divergent `SEVERITY_RANK` literal**\n\n```javascript\n#!/usr/bin/env node\n// test-threat-model-severity.mjs — regression guard for the iter-63\n// SEVERITY_RANK unification. threat-model.mjs still carries its own\n// local literal that is missing info/warn/error/critical keys present\n// in the shared _harness.mjs version — this drift was never caught\n// by an existing test.\n//\n// USAGE: node scripts/test-threat-model-severity.mjs\n\nimport { readFileSync } from 'node:fs';\nimport { SEVERITY_RANK as SHARED_RANK } from './_harness.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? (console.log(`  ✓ ${label}`), passed++)\n                                     : (console.log(`  ✗ ${label}`), failed++); }\n\nconst src = readFileSync(new URL('./threat-model.mjs', import.meta.url), 'utf8');\n\n// Fails today (documents the known gap) — once threat-model.mjs is fixed\n// to import SEVERITY_RANK from _harness.mjs, this assertion should hold.\nconst importsShared = /import\\s*\\{[^}]*SEVERITY_RANK[^}]*\\}\\s*from\\s*['\"]\\.\\/_harness\\.mjs['\"]/.test(src);\nassert(importsShared,\n  'threat-model.mjs imports the shared SEVERITY_RANK instead of a local literal');\n\n// Guards against future drift even if a local literal is reintroduced elsewhere.\nconst sharedKeys = Object.keys(SHARED_RANK).sort();\nassert(sharedKeys.includes('info') && sharedKeys.includes('critical'),\n  'shared SEVERITY_RANK in _harness.mjs still covers the full severity vocabulary');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);\n```\n\n**C. Extend `test-graceful-degradation.mjs`'s skill coverage list**\n\n```javascript\n// Add to the SKILLS array in test-graceful-degradation.mjs — currently\n// missing redblue.mjs, learn.mjs, gepa.mjs, mint.mjs, audit-list.mjs.\nconst ADDITIONAL_SKILLS = [\n  { script: 'mint.mjs', args: ['--name', 'tmp-test', '--template', 'vertical:coding'] },\n  { script: 'audit-list.mjs', args: ['--limit', '5'] },\n  { script: 'learn.mjs', args: ['--host', 'claude-code', '--model', 'haiku'] },\n  { script: 'gepa.mjs', args: ['--op', 'genome'] },\n  { script: 'redblue.mjs', args: ['run', '--mock-judge', '--tests', '1'] },\n];\n// Each should assert: exit code in [0], stdout parses as JSON, and\n// stdout contains `\"degraded\": true` when RUFLO_METAHARNESS_SKIP_LOCAL=1\n// + unreachable registry are set — same pattern as the existing 8 skills.\n```\n\n**D. `router-parallel-analyze.mjs` internal stats helpers (currently only reachable via full CLI)**\n\n```javascript\n#!/usr/bin/env node\n// test-router-stats.mjs — direct unit tests for median/pctile/mean.\n// These are currently unexported and only reachable through main()'s\n// aggregate CLI output — export them from router-parallel-analyze.mjs\n// for direct testability, then add:\n\n// After exporting: import { median, pctile, mean } from './router-parallel-analyze.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? (console.log(`  ✓ ${label}`), passed++)\n                                     : (console.log(`  ✗ ${label}`), failed++); }\n\n// median()\n// assert(median([1,2,3]) === 2, 'median: odd-length array');\n// assert(median([1,2,3,4]) === 2.5, 'median: even-length array interpolates');\n// assert(median([5]) === 5, 'median: single-element array');\n// assert(median([]) === undefined || Number.isNaN(median([])), 'median: empty array does not throw');\n\n// pctile()\n// assert(pctile([1,2,3,4,5,6,7,8,9,10], 0.5) === median([1,2,3,4,5,6,7,8,9,10]), 'pctile(0.5) matches median');\n// assert(pctile([1,2,3,4,5,6,7,8,9,10], 0.95) >= 9, 'pctile(0.95) is near the top of the range');\n// assert(pctile([1], 0.95) === 1, 'pctile: single-element array returns that element regardless of pct');\n\n// mean()\n// assert(mean([2,4,6]) === 4, 'mean: simple average');\n// assert(mean([]) === 0 || Number.isNaN(mean([])), 'mean: empty array does not throw');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);\n```\n\nNone of these were written to disk — they're skeletons for you to place under `scripts/` if you want to close the gaps. The single highest-leverage fix is **(A)** since `_invoke.mjs` underpins every upstream-CLI-wrapping script in the plugin and currently has no direct test at all; **(B)** is a one-line correctness bug (`threat-model.mjs`'s stale severity map) worth fixing regardless of the test.\n",
+  "parsedOutput": {
+    "sections": [
+      {
+        "title": "Test Coverage Analysis — `ruflo-metaharness`",
+        "content": "\nThe plugin has no formal test framework — coverage comes from 6 ad-hoc `test-*.mjs` scripts (real `assert()`/exit-code based) plus `scripts/smoke.sh` (123 grep-based structural checks, not runtime tests). Of 34 scripts, **6 have zero test coverage** and the most load-bearing shared module (`_invoke.mjs`) is never unit-tested at all.\n\n",
+        "level": 2
+      },
+      {
+        "title": "1. Untested functions/classes (no coverage in any of the 6 test files)",
+        "content": "\n| File | Status |\n|---|---|\n| `_invoke.mjs` | **Never unit-tested.** `findLocalPackageDir`, `satisfiesTildeRange`, `ensureCachedInstall`, `importOptionalLibrary`, `classifyDegraded`, `injectJson`, `parseTrailingJson` — only touched transitively via subprocess drills that force the single \"not-available\" branch |\n| `_darwin.mjs` | No unit test. `runDarwinAsync`'s `onProgress`/`signal` cancellation path is completely unexercised |\n| `_redblue.mjs` | No unit test at all (not even in the degradation drill) |\n| `mint.mjs` | Zero test coverage — including its root/self-repo write-refusal safety check |\n| `audit-list.mjs` | Zero test coverage |\n| `redblue.mjs`, `learn.mjs`, `gepa.mjs` | Zero test coverage — missing even from `test-graceful-degradation.mjs`'s skill list |\n| `router-parallel-analyze.mjs` | CLI tested end-to-end, but internal `median()`, `pctile()`, `mean()` never unit-tested directly (no verification of interpolation, even-length arrays, single-element arrays) |\n| `threat-model.mjs` | Has its own **stale, divergent** `SEVERITY_RANK` literal (missing `info/warn/error/critical` keys present in the shared `_harness.mjs` version) — this drift is invisible to every existing test |\n\n",
+        "level": 3
+      },
+      {
+        "title": "2. Edge cases not covered",
+        "content": "\n- `_invoke.mjs::satisfiesTildeRange` — no test for cross-major mismatch (`~0.3.0` vs `1.0.0`), pre-release suffixes (`0.3.0-beta.1`), or malformed version strings\n- `_invoke.mjs::parseTrailingJson` — the greedy whole-span fallback branch (used when no clean trailing `{...}` block exists) is never exercised\n- `_invoke.mjs::classifyDegraded` — `null` exitCode (timeout) vs `DEGRADED_RX` stderr match vs healthy are three distinct branches; only the stderr-match path is drilled\n- `_similarity.mjs::perDimensionBreakdown()` — only key-presence is checked in `test-similarity.mjs`, not the actual per-dimension numeric contribution math\n- `_similarity.mjs::similarity()` — no test with genuinely malformed input (`null`, missing `score`/`genome` keys entirely, non-object)\n- `oia-audit.mjs` composite worst-severity — no test of tie-breaking when two components report equal severity from different sources\n- `mint.mjs` — the upstream `--target`-ignored workaround (cwd/basename trick) has no regression test; a re-fix upstream could silently break it\n- `evolve.mjs --diagnose` — `buildDiagnosis`/`extractGepaTranscripts`/`looksLikeGepaTranscript` transcript-classification heuristics have no fixture-based test\n\n",
+        "level": 3
+      },
+      {
+        "title": "3. Missing error handling tests",
+        "content": "\n- **`threat-model.mjs`**: an unrecognized severity value (e.g. upstream adds a `\"warn\"` finding) falls through the local `SEVERITY_RANK` to `undefined`, producing a `NaN` comparison in the `--fail-on` gate instead of failing safe. No test catches this.\n- **`_invoke.mjs::ensureCachedInstall`**: no test of the `npm install` subprocess itself failing (disk full, permission denied, corrupt cache dir) as distinct from \"package unreachable\"\n- **`redblue.mjs`**: no test of the upstream `isMain` npx-symlink workaround failing to resolve a real path (falls back to what?)\n- **`test-graceful-degradation.mjs`** itself documents (in comments) two known-but-unfixed bugs — `mint.mjs`'s degraded payload possibly missing the `\"degraded\": true` marker, and `drift-from-history.mjs` returning exit 2 instead of documented exit 3 on upstream-absent — neither has a regression test asserting the *correct* behavior once fixed\n- **`gepa.mjs`**: no test of `analyzeTranscript` receiving a non-GEPA-shaped transcript file (malformed JSON, wrong schema)\n\n",
+        "level": 3
+      },
+      {
+        "title": "4. Integration test gaps",
+        "content": "\n- **`redblue.mjs`, `learn.mjs`, `gepa.mjs`, `mint.mjs`, `audit-list.mjs`** are absent from `test-graceful-degradation.mjs`'s subprocess drill — none of the \"does this script behave correctly when upstream is completely unreachable\" guarantees are verified for them\n- **`test-mcp-tools.mjs` silently no-ops** or when the sibling monorepo's compiled `dist/` isn't present (plugin-only checkout) — this means the 15-tool MCP handler contract goes completely unverified in that environment, with `exit 0` masking zero assertions run\n- **`_darwin.mjs`/`_redblue.mjs` \"already cached\" and \"local install found\" resolution paths** are never exercised — only \"must install fresh\" and \"unreachable\" are covered\n- **`audit-trend.mjs` ↔ `_similarity.mjs` composition** is only tested via `test-pipeline-roundtrip.mjs`'s self-match/synthetic-mutation cases — no test of genuinely divergent real-world audit pairs (e.g. different repo entirely, comparing incompatible genomes)\n- **`test-with-openrouter.mjs`** bypasses `_harness.mjs`'s pinned-version resolution entirely (uses `metaharness@latest`) — it looks like an integration test of the harness CLI surface but never actually validates the plugin's own caching/pin logic under real network conditions\n\n---\n\n",
+        "level": 3
+      },
+      {
+        "title": "Test skeletons",
+        "content": "**A. `_invoke.mjs` unit tests (highest-priority gap — currently zero coverage)**\n\n```javascript\n#!/usr/bin/env node\n// test-invoke.mjs — unit tests for the shared subprocess/plumbing layer\n// (_invoke.mjs). Currently untested despite being imported by\n// _harness.mjs, _darwin.mjs, _redblue.mjs, and gepa.mjs.\n//\n// USAGE\n//   node scripts/test-invoke.mjs\n// EXIT\n//   0 all pass, 1 at least one failure\n\nimport {\n  classifyDegraded, injectJson, parseTrailingJson,\n  satisfiesTildeRange, cacheBaseDir,\n} from './_invoke.mjs';\n\nlet passed = 0, failed = 0;\nconst failures = [];\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failures.push(label); failed++; }\n}\n\n// ── classifyDegraded ──────────────────────────────────────────────\n{\n  const timeout = classifyDegraded('', null, 'harness');\n  assert(timeout.degraded === true, 'classifyDegraded: null exitCode (timeout) is degraded');\n\n  const notFound = classifyDegraded('npm ERR! 404 Not Found', 1, 'harness');\n  assert(notFound.degraded === true, 'classifyDegraded: DEGRADED_RX stderr match is degraded');\n\n  const healthy = classifyDegraded('', 0, 'harness');\n  assert(healthy.degraded === false, 'classifyDegraded: clean exit is not degraded');\n\n  const nonzeroButHealthy = classifyDegraded('some real validation error', 1, 'harness');\n  assert(nonzeroButHealthy.degraded === false,\n    'classifyDegraded: nonzero exit without DEGRADED_RX match is NOT treated as degraded');\n}\n\n// ── injectJson ─────────────────────────────────────────────────────\n{\n  assert(injectJson(['score', '.'], true).includes('--json'),\n    'injectJson: appends --json when wantJson=true and absent');\n  assert(!injectJson(['score', '.'], false).includes('--json'),\n    'injectJson: does not append when wantJson=false');\n  const already = injectJson(['score', '.', '--json'], true);\n  assert(already.filter(a => a === '--json').length === 1,\n    'injectJson: does not duplicate --json if already present');\n}\n\n// ── parseTrailingJson ────────────────────────────────────────────\n{\n  const clean = parseTrailingJson('some log line\\n{\"a\":1}');\n  assert(clean && clean.a === 1, 'parseTrailingJson: extracts trailing JSON block');\n\n  const noisy = parseTrailingJson('{\"a\":1}\\nnot json trailing garbage');\n  assert(noisy === null || typeof noisy === 'object',\n    'parseTrailingJson: gracefully handles non-JSON trailing content (greedy fallback)');\n\n  assert(parseTrailingJson('') === null, 'parseTrailingJson: empty string returns null, not throw');\n  assert(parseTrailingJson('no json here at all') === null,\n    'parseTrailingJson: no JSON present returns null');\n}\n\n// ── satisfiesTildeRange ──────────────────────────────────────────\n{\n  assert(satisfiesTildeRange('0.3.2', '~0.3.0') === true,\n    'satisfiesTildeRange: same major.minor, patch >= pin is satisfied');\n  assert(satisfiesTildeRange('0.3.0', '~0.3.0') === true,\n    'satisfiesTildeRange: exact match is satisfied');\n  assert(satisfiesTildeRange('0.4.0', '~0.3.0') === false,\n    'satisfiesTildeRange: different minor is NOT satisfied');\n  assert(satisfiesTildeRange('1.3.0', '~0.3.0') === false,\n    'satisfiesTildeRange: different major is NOT satisfied');\n  assert(satisfiesTildeRange('0.3.0-beta.1', '~0.3.0') === false,\n    'satisfiesTildeRange: pre-release suffix does not satisfy plain pin');\n  assert(satisfiesTildeRange('not-a-version', '~0.3.0') === false,\n    'satisfiesTildeRange: malformed version string fails closed, not throw');\n}\n\n// ── cacheBaseDir ──────────────────────────────────────────────────\n{\n  const original = process.env.RUFLO_METAHARNESS_CACHE_BASE;\n  delete process.env.RUFLO_METAHARNESS_CACHE_BASE;\n  assert(cacheBaseDir().includes('.ruflo'), 'cacheBaseDir: defaults under ~/.ruflo');\n  process.env.RUFLO_METAHARNESS_CACHE_BASE = '/tmp/custom-cache';\n  assert(cacheBaseDir() === '/tmp/custom-cache', 'cacheBaseDir: honors env override test seam');\n  if (original === undefined) delete process.env.RUFLO_METAHARNESS_CACHE_BASE;\n  else process.env.RUFLO_METAHARNESS_CACHE_BASE = original;\n}\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) { console.log('Failures:', failures.join(', ')); process.exit(1); }\nprocess.exit(0);\n```\n\n**B. `threat-model.mjs` — regression test for the divergent `SEVERITY_RANK` literal**\n\n```javascript\n#!/usr/bin/env node\n// test-threat-model-severity.mjs — regression guard for the iter-63\n// SEVERITY_RANK unification. threat-model.mjs still carries its own\n// local literal that is missing info/warn/error/critical keys present\n// in the shared _harness.mjs version — this drift was never caught\n// by an existing test.\n//\n// USAGE: node scripts/test-threat-model-severity.mjs\n\nimport { readFileSync } from 'node:fs';\nimport { SEVERITY_RANK as SHARED_RANK } from './_harness.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? (console.log(`  ✓ ${label}`), passed++)\n                                     : (console.log(`  ✗ ${label}`), failed++); }\n\nconst src = readFileSync(new URL('./threat-model.mjs', import.meta.url), 'utf8');\n\n// Fails today (documents the known gap) — once threat-model.mjs is fixed\n// to import SEVERITY_RANK from _harness.mjs, this assertion should hold.\nconst importsShared = /import\\s*\\{[^}]*SEVERITY_RANK[^}]*\\}\\s*from\\s*['\"]\\.\\/_harness\\.mjs['\"]/.test(src);\nassert(importsShared,\n  'threat-model.mjs imports the shared SEVERITY_RANK instead of a local literal');\n\n// Guards against future drift even if a local literal is reintroduced elsewhere.\nconst sharedKeys = Object.keys(SHARED_RANK).sort();\nassert(sharedKeys.includes('info') && sharedKeys.includes('critical'),\n  'shared SEVERITY_RANK in _harness.mjs still covers the full severity vocabulary');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);\n```\n\n**C. Extend `test-graceful-degradation.mjs`'s skill coverage list**\n\n```javascript\n// Add to the SKILLS array in test-graceful-degradation.mjs — currently\n// missing redblue.mjs, learn.mjs, gepa.mjs, mint.mjs, audit-list.mjs.\nconst ADDITIONAL_SKILLS = [\n  { script: 'mint.mjs', args: ['--name', 'tmp-test', '--template', 'vertical:coding'] },\n  { script: 'audit-list.mjs', args: ['--limit', '5'] },\n  { script: 'learn.mjs', args: ['--host', 'claude-code', '--model', 'haiku'] },\n  { script: 'gepa.mjs', args: ['--op', 'genome'] },\n  { script: 'redblue.mjs', args: ['run', '--mock-judge', '--tests', '1'] },\n];\n// Each should assert: exit code in [0], stdout parses as JSON, and\n// stdout contains `\"degraded\": true` when RUFLO_METAHARNESS_SKIP_LOCAL=1\n// + unreachable registry are set — same pattern as the existing 8 skills.\n```\n\n**D. `router-parallel-analyze.mjs` internal stats helpers (currently only reachable via full CLI)**\n\n```javascript\n#!/usr/bin/env node\n// test-router-stats.mjs — direct unit tests for median/pctile/mean.\n// These are currently unexported and only reachable through main()'s\n// aggregate CLI output — export them from router-parallel-analyze.mjs\n// for direct testability, then add:\n\n// After exporting: import { median, pctile, mean } from './router-parallel-analyze.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? (console.log(`  ✓ ${label}`), passed++)\n                                     : (console.log(`  ✗ ${label}`), failed++); }\n\n// median()\n// assert(median([1,2,3]) === 2, 'median: odd-length array');\n// assert(median([1,2,3,4]) === 2.5, 'median: even-length array interpolates');\n// assert(median([5]) === 5, 'median: single-element array');\n// assert(median([]) === undefined || Number.isNaN(median([])), 'median: empty array does not throw');\n\n// pctile()\n// assert(pctile([1,2,3,4,5,6,7,8,9,10], 0.5) === median([1,2,3,4,5,6,7,8,9,10]), 'pctile(0.5) matches median');\n// assert(pctile([1,2,3,4,5,6,7,8,9,10], 0.95) >= 9, 'pctile(0.95) is near the top of the range');\n// assert(pctile([1], 0.95) === 1, 'pctile: single-element array returns that element regardless of pct');\n\n// mean()\n// assert(mean([2,4,6]) === 4, 'mean: simple average');\n// assert(mean([]) === 0 || Number.isNaN(mean([])), 'mean: empty array does not throw');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);\n```\n\nNone of these were written to disk — they're skeletons for you to place under `scripts/` if you want to close the gaps. The single highest-leverage fix is **(A)** since `_invoke.mjs` underpins every upstream-CLI-wrapping script in the plugin and currently has no direct test at all; **(B)** is a one-line correctness bug (`threat-model.mjs`'s stale severity map) worth fixing regardless of the test.",
+        "level": 3
+      }
+    ],
+    "codeBlocks": [
+      {
+        "language": "javascript",
+        "code": "#!/usr/bin/env node\n// test-invoke.mjs — unit tests for the shared subprocess/plumbing layer\n// (_invoke.mjs). Currently untested despite being imported by\n// _harness.mjs, _darwin.mjs, _redblue.mjs, and gepa.mjs.\n//\n// USAGE\n//   node scripts/test-invoke.mjs\n// EXIT\n//   0 all pass, 1 at least one failure\n\nimport {\n  classifyDegraded, injectJson, parseTrailingJson,\n  satisfiesTildeRange, cacheBaseDir,\n} from './_invoke.mjs';\n\nlet passed = 0, failed = 0;\nconst failures = [];\nfunction assert(cond, label) {\n  if (cond) { console.log(`  ✓ ${label}`); passed++; }\n  else { console.log(`  ✗ ${label}`); failures.push(label); failed++; }\n}\n\n// ── classifyDegraded ──────────────────────────────────────────────\n{\n  const timeout = classifyDegraded('', null, 'harness');\n  assert(timeout.degraded === true, 'classifyDegraded: null exitCode (timeout) is degraded');\n\n  const notFound = classifyDegraded('npm ERR! 404 Not Found', 1, 'harness');\n  assert(notFound.degraded === true, 'classifyDegraded: DEGRADED_RX stderr match is degraded');\n\n  const healthy = classifyDegraded('', 0, 'harness');\n  assert(healthy.degraded === false, 'classifyDegraded: clean exit is not degraded');\n\n  const nonzeroButHealthy = classifyDegraded('some real validation error', 1, 'harness');\n  assert(nonzeroButHealthy.degraded === false,\n    'classifyDegraded: nonzero exit without DEGRADED_RX match is NOT treated as degraded');\n}\n\n// ── injectJson ─────────────────────────────────────────────────────\n{\n  assert(injectJson(['score', '.'], true).includes('--json'),\n    'injectJson: appends --json when wantJson=true and absent');\n  assert(!injectJson(['score', '.'], false).includes('--json'),\n    'injectJson: does not append when wantJson=false');\n  const already = injectJson(['score', '.', '--json'], true);\n  assert(already.filter(a => a === '--json').length === 1,\n    'injectJson: does not duplicate --json if already present');\n}\n\n// ── parseTrailingJson ────────────────────────────────────────────\n{\n  const clean = parseTrailingJson('some log line\\n{\"a\":1}');\n  assert(clean && clean.a === 1, 'parseTrailingJson: extracts trailing JSON block');\n\n  const noisy = parseTrailingJson('{\"a\":1}\\nnot json trailing garbage');\n  assert(noisy === null || typeof noisy === 'object',\n    'parseTrailingJson: gracefully handles non-JSON trailing content (greedy fallback)');\n\n  assert(parseTrailingJson('') === null, 'parseTrailingJson: empty string returns null, not throw');\n  assert(parseTrailingJson('no json here at all') === null,\n    'parseTrailingJson: no JSON present returns null');\n}\n\n// ── satisfiesTildeRange ──────────────────────────────────────────\n{\n  assert(satisfiesTildeRange('0.3.2', '~0.3.0') === true,\n    'satisfiesTildeRange: same major.minor, patch >= pin is satisfied');\n  assert(satisfiesTildeRange('0.3.0', '~0.3.0') === true,\n    'satisfiesTildeRange: exact match is satisfied');\n  assert(satisfiesTildeRange('0.4.0', '~0.3.0') === false,\n    'satisfiesTildeRange: different minor is NOT satisfied');\n  assert(satisfiesTildeRange('1.3.0', '~0.3.0') === false,\n    'satisfiesTildeRange: different major is NOT satisfied');\n  assert(satisfiesTildeRange('0.3.0-beta.1', '~0.3.0') === false,\n    'satisfiesTildeRange: pre-release suffix does not satisfy plain pin');\n  assert(satisfiesTildeRange('not-a-version', '~0.3.0') === false,\n    'satisfiesTildeRange: malformed version string fails closed, not throw');\n}\n\n// ── cacheBaseDir ──────────────────────────────────────────────────\n{\n  const original = process.env.RUFLO_METAHARNESS_CACHE_BASE;\n  delete process.env.RUFLO_METAHARNESS_CACHE_BASE;\n  assert(cacheBaseDir().includes('.ruflo'), 'cacheBaseDir: defaults under ~/.ruflo');\n  process.env.RUFLO_METAHARNESS_CACHE_BASE = '/tmp/custom-cache';\n  assert(cacheBaseDir() === '/tmp/custom-cache', 'cacheBaseDir: honors env override test seam');\n  if (original === undefined) delete process.env.RUFLO_METAHARNESS_CACHE_BASE;\n  else process.env.RUFLO_METAHARNESS_CACHE_BASE = original;\n}\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nif (failed > 0) { console.log('Failures:', failures.join(', ')); process.exit(1); }\nprocess.exit(0);"
+      },
+      {
+        "language": "javascript",
+        "code": "#!/usr/bin/env node\n// test-threat-model-severity.mjs — regression guard for the iter-63\n// SEVERITY_RANK unification. threat-model.mjs still carries its own\n// local literal that is missing info/warn/error/critical keys present\n// in the shared _harness.mjs version — this drift was never caught\n// by an existing test.\n//\n// USAGE: node scripts/test-threat-model-severity.mjs\n\nimport { readFileSync } from 'node:fs';\nimport { SEVERITY_RANK as SHARED_RANK } from './_harness.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? (console.log(`  ✓ ${label}`), passed++)\n                                     : (console.log(`  ✗ ${label}`), failed++); }\n\nconst src = readFileSync(new URL('./threat-model.mjs', import.meta.url), 'utf8');\n\n// Fails today (documents the known gap) — once threat-model.mjs is fixed\n// to import SEVERITY_RANK from _harness.mjs, this assertion should hold.\nconst importsShared = /import\\s*\\{[^}]*SEVERITY_RANK[^}]*\\}\\s*from\\s*['\"]\\.\\/_harness\\.mjs['\"]/.test(src);\nassert(importsShared,\n  'threat-model.mjs imports the shared SEVERITY_RANK instead of a local literal');\n\n// Guards against future drift even if a local literal is reintroduced elsewhere.\nconst sharedKeys = Object.keys(SHARED_RANK).sort();\nassert(sharedKeys.includes('info') && sharedKeys.includes('critical'),\n  'shared SEVERITY_RANK in _harness.mjs still covers the full severity vocabulary');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);"
+      },
+      {
+        "language": "javascript",
+        "code": "// Add to the SKILLS array in test-graceful-degradation.mjs — currently\n// missing redblue.mjs, learn.mjs, gepa.mjs, mint.mjs, audit-list.mjs.\nconst ADDITIONAL_SKILLS = [\n  { script: 'mint.mjs', args: ['--name', 'tmp-test', '--template', 'vertical:coding'] },\n  { script: 'audit-list.mjs', args: ['--limit', '5'] },\n  { script: 'learn.mjs', args: ['--host', 'claude-code', '--model', 'haiku'] },\n  { script: 'gepa.mjs', args: ['--op', 'genome'] },\n  { script: 'redblue.mjs', args: ['run', '--mock-judge', '--tests', '1'] },\n];\n// Each should assert: exit code in [0], stdout parses as JSON, and\n// stdout contains `\"degraded\": true` when RUFLO_METAHARNESS_SKIP_LOCAL=1\n// + unreachable registry are set — same pattern as the existing 8 skills."
+      },
+      {
+        "language": "javascript",
+        "code": "#!/usr/bin/env node\n// test-router-stats.mjs — direct unit tests for median/pctile/mean.\n// These are currently unexported and only reachable through main()'s\n// aggregate CLI output — export them from router-parallel-analyze.mjs\n// for direct testability, then add:\n\n// After exporting: import { median, pctile, mean } from './router-parallel-analyze.mjs';\n\nlet passed = 0, failed = 0;\nfunction assert(cond, label) { cond ? (console.log(`  ✓ ${label}`), passed++)\n                                     : (console.log(`  ✗ ${label}`), failed++); }\n\n// median()\n// assert(median([1,2,3]) === 2, 'median: odd-length array');\n// assert(median([1,2,3,4]) === 2.5, 'median: even-length array interpolates');\n// assert(median([5]) === 5, 'median: single-element array');\n// assert(median([]) === undefined || Number.isNaN(median([])), 'median: empty array does not throw');\n\n// pctile()\n// assert(pctile([1,2,3,4,5,6,7,8,9,10], 0.5) === median([1,2,3,4,5,6,7,8,9,10]), 'pctile(0.5) matches median');\n// assert(pctile([1,2,3,4,5,6,7,8,9,10], 0.95) >= 9, 'pctile(0.95) is near the top of the range');\n// assert(pctile([1], 0.95) === 1, 'pctile: single-element array returns that element regardless of pct');\n\n// mean()\n// assert(mean([2,4,6]) === 4, 'mean: simple average');\n// assert(mean([]) === 0 || Number.isNaN(mean([])), 'mean: empty array does not throw');\n\nconsole.log(`\\n${passed} passed, ${failed} failed`);\nprocess.exit(failed > 0 ? 1 : 0);"
+      }
+    ]
+  },
+  "durationMs": 291126,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T17:03:52.819Z",
+  "executionId": "testgaps_1783616341693_bl5d9o"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783617832831_ha6s8d_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T17:23:52.831Z] 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_1783617832831_ha6s8d_result.log

@@ -0,0 +1,42 @@
+[2026-07-09T17:29:44.995Z] RESULT
+============================================================
+{
+  "success": true,
+  "output": "## Test Coverage Analysis — `plugins/ruflo-metaharness/scripts/`\n\n**Bottom line:** ~7,000 lines across 34 `.mjs` files, no `package.json`, no test framework (no jest/vitest/`node:test` config). The `test-*.mjs` files are hand-rolled assertion scripts, not a real suite — coverage is uneven and several high-risk files (`evolve.mjs`, `mint.mjs` safety checks, `_invoke.mjs` core plumbing) have **zero direct unit tests**.\n\n### Coverage by file (best → worst)\n\n| File | Coverage | Key gap |\n|---|---|---|\n| `_similarity.mjs` | Strong — near-complete unit tests | `perDimension` contribution formulas type-checked, not value-asserted |\n| `audit-trend.mjs` | Strong e2e via `test-pipeline-roundtrip.mjs` | `loadRecord` file-not-found / invalid-JSON paths untested |\n| `router-parallel-analyze.mjs` | Good | `<30` samples, malformed JSONL, missing-file branches untested |\n| `drift-from-history.mjs` | Good (fast paths) | Default slow path (real `audit-list.mjs` call), zero-history branch untested |\n| `oia-audit.mjs` | Decent (dry-run heavy) | `persist()`'s real memory-store call and its failure path **never** executed; partial-degradation (1-4 of 5 sub-audits fail) untested |\n| `mcp-scan.mjs`, `similarity.mjs`, `score.mjs`, `genome.mjs` | Shallow — positive path only | `--format table`, invalid `--fail-on`, alert-triggered exit-1 paths untested |\n| `threat-model.mjs`, `audit-list.mjs` | Weak — grep-only in `smoke.sh` | Pure functions (`parseDurationMs`) never called directly; a **divergent duplicate `SEVERITY_RANK`** in `threat-model.mjs:9` (missing keys vs. `_harness.mjs`) has no drift test |\n| `security-bench.mjs` | Weak | `parseSecurityBenchMarkdown` — a 3-regex markdown parser — has **zero** unit tests (same bug class the iter-50 fix addressed for `parseMcpScanText`, but unprotected here) |\n| `mint.mjs` | Weak | `safetyChecks()` — the \"refuse to write to project root / repo root\" guard, explicitly documented as safety-critical — **completely untested** |\n| `redblue.mjs`, `learn.mjs`, `gepa.mjs` | Weak | Only 1 of ~4 subcommands/ops exercised each; arg-guards and error branches untested |\n| `bench.mjs` | **Zero** functional coverage | Only `node --check` syntax validation |\n| `evolve.mjs` (404 lines, most complex file) | **Near zero** | The entire `--diagnose` subsystem (`buildDiagnosis`, `extractGepaTranscripts`, `summarizeTraces`, `looksLikeGepaTranscript`) — ~130 lines of failure-classification logic — has no test at all |\n| `_invoke.mjs` (shared plumbing, 10 exports) | **Zero direct tests** | `classifyDegraded` — the single most load-bearing function in the whole degradation contract (timeout vs. not-available vs. healthy) — untested in isolation. `parseTrailingJson`'s \"LAST JSON block wins\" behavior, an explicitly-documented prior bug fix, has **no regression test** |\n| `_darwin.mjs` | **Zero direct tests** | `runDarwinAsync`'s timeout-kill and `onProgress` streaming paths untested; used by `evolve.mjs`/`security-bench.mjs`, both safety-critical |\n\n### Meta-testing gap\n`smoke.sh` verifies `test-*.mjs` scripts via `grep -qE \"(N+) passed, 0 failed\"` — a floor-check on the printed count, not real assertion inspection. A test file that silently stops asserting (but still prints \"0 failed\") would pass CI undetected.\n\n### Integration gap\n`test-mcp-tools.mjs` covers the MCP-tool-handler → script path, but **silently no-ops** if `dist/` isn't built (exits 0 with a SKIPPED message). The plain CLI path (`ruflo metaharness score`) is only checked by `smoke.sh` for \"does the dispatcher map mention the filename\" — never actually invoked end-to-end.\n\n### Top 3 gaps worth fixing first\n1. **`_invoke.mjs::classifyDegraded` + `parseTrailingJson`** — foundational, every script depends on these; a `scripts/test-invoke.mjs` with `node:test` would be cheap and high-value (skeleton above).\n2. **`mint.mjs::safetyChecks`** — a documented safety invariant (never scaffold into project root/repo root) with no regression test guarding it.\n3. **`evolve.mjs`'s `--diagnose` subsystem** — most complex untested logic in the plugin; would need exporting 4 currently-module-local helpers first.\n\nFull file-by-file breakdown with file:line references and 8 ready-to-use `node:test` skeletons (covering `_invoke.mjs`, `mint.mjs` safety, `security-bench.mjs` parser, `evolve.mjs` diagnosis, `audit-trend.mjs` edge cases, `router-parallel-analyze.mjs`, `_darwin.mjs` async) is above. Want me to actually write any of these test files, or export the currently-unexported helpers (`buildDiagnosis`, `parseSecurityBenchMarkdown`) so they can be tested?\n",
+  "parsedOutput": {
+    "sections": [
+      {
+        "title": "Test Coverage Analysis — `plugins/ruflo-metaharness/scripts/`",
+        "content": "\n**Bottom line:** ~7,000 lines across 34 `.mjs` files, no `package.json`, no test framework (no jest/vitest/`node:test` config). The `test-*.mjs` files are hand-rolled assertion scripts, not a real suite — coverage is uneven and several high-risk files (`evolve.mjs`, `mint.mjs` safety checks, `_invoke.mjs` core plumbing) have **zero direct unit tests**.\n\n",
+        "level": 2
+      },
+      {
+        "title": "Coverage by file (best → worst)",
+        "content": "\n| File | Coverage | Key gap |\n|---|---|---|\n| `_similarity.mjs` | Strong — near-complete unit tests | `perDimension` contribution formulas type-checked, not value-asserted |\n| `audit-trend.mjs` | Strong e2e via `test-pipeline-roundtrip.mjs` | `loadRecord` file-not-found / invalid-JSON paths untested |\n| `router-parallel-analyze.mjs` | Good | `<30` samples, malformed JSONL, missing-file branches untested |\n| `drift-from-history.mjs` | Good (fast paths) | Default slow path (real `audit-list.mjs` call), zero-history branch untested |\n| `oia-audit.mjs` | Decent (dry-run heavy) | `persist()`'s real memory-store call and its failure path **never** executed; partial-degradation (1-4 of 5 sub-audits fail) untested |\n| `mcp-scan.mjs`, `similarity.mjs`, `score.mjs`, `genome.mjs` | Shallow — positive path only | `--format table`, invalid `--fail-on`, alert-triggered exit-1 paths untested |\n| `threat-model.mjs`, `audit-list.mjs` | Weak — grep-only in `smoke.sh` | Pure functions (`parseDurationMs`) never called directly; a **divergent duplicate `SEVERITY_RANK`** in `threat-model.mjs:9` (missing keys vs. `_harness.mjs`) has no drift test |\n| `security-bench.mjs` | Weak | `parseSecurityBenchMarkdown` — a 3-regex markdown parser — has **zero** unit tests (same bug class the iter-50 fix addressed for `parseMcpScanText`, but unprotected here) |\n| `mint.mjs` | Weak | `safetyChecks()` — the \"refuse to write to project root / repo root\" guard, explicitly documented as safety-critical — **completely untested** |\n| `redblue.mjs`, `learn.mjs`, `gepa.mjs` | Weak | Only 1 of ~4 subcommands/ops exercised each; arg-guards and error branches untested |\n| `bench.mjs` | **Zero** functional coverage | Only `node --check` syntax validation |\n| `evolve.mjs` (404 lines, most complex file) | **Near zero** | The entire `--diagnose` subsystem (`buildDiagnosis`, `extractGepaTranscripts`, `summarizeTraces`, `looksLikeGepaTranscript`) — ~130 lines of failure-classification logic — has no test at all |\n| `_invoke.mjs` (shared plumbing, 10 exports) | **Zero direct tests** | `classifyDegraded` — the single most load-bearing function in the whole degradation contract (timeout vs. not-available vs. healthy) — untested in isolation. `parseTrailingJson`'s \"LAST JSON block wins\" behavior, an explicitly-documented prior bug fix, has **no regression test** |\n| `_darwin.mjs` | **Zero direct tests** | `runDarwinAsync`'s timeout-kill and `onProgress` streaming paths untested; used by `evolve.mjs`/`security-bench.mjs`, both safety-critical |\n\n",
+        "level": 3
+      },
+      {
+        "title": "Meta-testing gap",
+        "content": "`smoke.sh` verifies `test-*.mjs` scripts via `grep -qE \"(N+) passed, 0 failed\"` — a floor-check on the printed count, not real assertion inspection. A test file that silently stops asserting (but still prints \"0 failed\") would pass CI undetected.\n\n",
+        "level": 3
+      },
+      {
+        "title": "Integration gap",
+        "content": "`test-mcp-tools.mjs` covers the MCP-tool-handler → script path, but **silently no-ops** if `dist/` isn't built (exits 0 with a SKIPPED message). The plain CLI path (`ruflo metaharness score`) is only checked by `smoke.sh` for \"does the dispatcher map mention the filename\" — never actually invoked end-to-end.\n\n",
+        "level": 3
+      },
+      {
+        "title": "Top 3 gaps worth fixing first",
+        "content": "1. **`_invoke.mjs::classifyDegraded` + `parseTrailingJson`** — foundational, every script depends on these; a `scripts/test-invoke.mjs` with `node:test` would be cheap and high-value (skeleton above).\n2. **`mint.mjs::safetyChecks`** — a documented safety invariant (never scaffold into project root/repo root) with no regression test guarding it.\n3. **`evolve.mjs`'s `--diagnose` subsystem** — most complex untested logic in the plugin; would need exporting 4 currently-module-local helpers first.\n\nFull file-by-file breakdown with file:line references and 8 ready-to-use `node:test` skeletons (covering `_invoke.mjs`, `mint.mjs` safety, `security-bench.mjs` parser, `evolve.mjs` diagnosis, `audit-trend.mjs` edge cases, `router-parallel-analyze.mjs`, `_darwin.mjs` async) is above. Want me to actually write any of these test files, or export the currently-unexported helpers (`buildDiagnosis`, `parseSecurityBenchMarkdown`) so they can be tested?",
+        "level": 3
+      }
+    ],
+    "codeBlocks": []
+  },
+  "durationMs": 352164,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T17:29:44.995Z",
+  "executionId": "testgaps_1783617832831_ha6s8d"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783619384959_s5iiwf_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T17:49:44.960Z] 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_1783619384959_s5iiwf_result.log

@@ -0,0 +1,47 @@
+[2026-07-09T17:55:46.253Z] RESULT
+============================================================
+{
+  "success": true,
+  "output": "## Test Coverage Gap Analysis — `ruflo-metaharness`\n\nThe plugin has no formal test framework — just 26 `.mjs` scripts (~6,900 lines) and 6 ad-hoc `assert`-based `test-*.mjs` runners, plus a `smoke.sh` that only does static grep + `node --check` (never actually executes scripts like `mint.mjs`).\n\n### Untested / weakly-tested functions\n- **`_invoke.mjs`** (the shared dependency of everything) has **zero direct unit tests**: `classifyDegraded`, `injectJson`, `parseTrailingJson`, `satisfiesTildeRange`, `findLocalPackageDir`, `ensureCachedInstall`, `importOptionalLibrary` (the plugin's only `throw` path, at `_invoke.mjs:204`), `makeDegradedEmitter`.\n- **`_darwin.mjs`** and **`_redblue.mjs`**: only their *degraded* branches are ever exercised — the success/happy-path branches (`runDarwin`, `runDarwinAsync`, `runRedblue`) have never been run against a real or mocked binary.\n- **`_harness.mjs`**: `parseMcpScanText`/`rankSeverity` are well-tested; `resolveMetaharnessBins`, `execBin`/`execBinAsync`, `metaharnessResolution` are not.\n- Best-covered module: `_similarity.mjs` (extensively tested), though weight-validation and NaN/negative-input handling are still gaps.\n- CLI scripts with essentially no coverage: `mint.mjs` (grep-only), `gepa.mjs` (validate/render/analyze ops), `learn.mjs`, `redblue.mjs` (init/run/patch/report), `audit-list.mjs`, `security-bench.mjs`'s markdown parser.\n\n### Key edge cases missed\n- NaN/negative/out-of-range numbers in similarity/scoring math (`_similarity.mjs`, `genome.mjs`).\n- Path traversal / non-UTF8 content in user-supplied file args (`audit-trend.mjs --baseline/--current`, `similarity.mjs --a/--b`, `gepa.mjs --path/--transcript`).\n- Concurrent partial-failure races in `oia-audit.mjs::runAllParallel` and `drift-from-history.mjs`'s `Promise.all`.\n- Graceful-degradation drill explicitly omits `gepa.mjs`, `learn.mjs`, `redblue.mjs`, and `mint.mjs` from its skill list — their exit-0/`degraded:true` contract is unverified.\n\n### Missing error-handling tests (highest-value finding)\n`audit-list.mjs`, `audit-trend.mjs`, `oia-audit.mjs::persist`, and `similarity.mjs` all call `spawnSync('npx', ...)` for memory-CLI access **without a timeout** — inconsistent with every other subprocess call in the plugin (which sets `timeoutMs`) — and no test has ever caught this. A hung `npx` would block these calls indefinitely.\n\n**A live regression bug was found during this analysis**: `threat-model.mjs:9` redeclares a stale local `SEVERITY_RANK` map missing `info`/`warn`/`error`/`critical` — the exact bug class that iter 62/63 fixed everywhere else by centralizing the map in `_harness.mjs`. Every sibling script (`mcp-scan.mjs`, `oia-audit.mjs`, `audit-trend.mjs`) imports the shared map; `threat-model.mjs` was missed. If upstream ever emits `worst: \"warn\"` or `\"critical\"`, `--fail-on` silently fails to trigger (`undefined >= threshold` → `false`).\n\n### Integration gaps\n- `drift-from-history.mjs`'s `--baseline-since` variant and its no-flag default path are never exercised end-to-end.\n- `redblue.mjs`'s `init`/`run`/`patch`/`report` subcommands (including the documented $0 `--mock-judge` CI path) have never been run by any test — only `attack` once, through the MCP layer.\n- MCP tool surface vs. raw CLI surface isn't symmetric: some scripts (`mint.mjs`, `audit-list.mjs`) have no MCP tool *and* no CLI-level test — a double gap.\n\n### Test skeletons\nThe full report includes 9 ready-to-adapt skeletons (matching the existing plain-`assert` idiom) for: `_invoke.mjs` primitives, `_darwin.mjs::runDarwin`, a `threat-model.mjs` severity-rank regression lock, a static check that all memory-CLI `spawnSync` calls set a timeout, `gepa.mjs` op dispatch, `security-bench.mjs`'s markdown parser, `redblue.mjs` subcommand dispatch, `oia-audit.mjs::persist` failure path, and `drift-from-history.mjs --baseline-since`.\n\nWant me to write any of these test files (e.g. `scripts/test-invoke.mjs` or the `threat-model.mjs` severity-rank fix) into the repo?\n",
+  "parsedOutput": {
+    "sections": [
+      {
+        "title": "Test Coverage Gap Analysis — `ruflo-metaharness`",
+        "content": "\nThe plugin has no formal test framework — just 26 `.mjs` scripts (~6,900 lines) and 6 ad-hoc `assert`-based `test-*.mjs` runners, plus a `smoke.sh` that only does static grep + `node --check` (never actually executes scripts like `mint.mjs`).\n\n",
+        "level": 2
+      },
+      {
+        "title": "Untested / weakly-tested functions",
+        "content": "- **`_invoke.mjs`** (the shared dependency of everything) has **zero direct unit tests**: `classifyDegraded`, `injectJson`, `parseTrailingJson`, `satisfiesTildeRange`, `findLocalPackageDir`, `ensureCachedInstall`, `importOptionalLibrary` (the plugin's only `throw` path, at `_invoke.mjs:204`), `makeDegradedEmitter`.\n- **`_darwin.mjs`** and **`_redblue.mjs`**: only their *degraded* branches are ever exercised — the success/happy-path branches (`runDarwin`, `runDarwinAsync`, `runRedblue`) have never been run against a real or mocked binary.\n- **`_harness.mjs`**: `parseMcpScanText`/`rankSeverity` are well-tested; `resolveMetaharnessBins`, `execBin`/`execBinAsync`, `metaharnessResolution` are not.\n- Best-covered module: `_similarity.mjs` (extensively tested), though weight-validation and NaN/negative-input handling are still gaps.\n- CLI scripts with essentially no coverage: `mint.mjs` (grep-only), `gepa.mjs` (validate/render/analyze ops), `learn.mjs`, `redblue.mjs` (init/run/patch/report), `audit-list.mjs`, `security-bench.mjs`'s markdown parser.\n\n",
+        "level": 3
+      },
+      {
+        "title": "Key edge cases missed",
+        "content": "- NaN/negative/out-of-range numbers in similarity/scoring math (`_similarity.mjs`, `genome.mjs`).\n- Path traversal / non-UTF8 content in user-supplied file args (`audit-trend.mjs --baseline/--current`, `similarity.mjs --a/--b`, `gepa.mjs --path/--transcript`).\n- Concurrent partial-failure races in `oia-audit.mjs::runAllParallel` and `drift-from-history.mjs`'s `Promise.all`.\n- Graceful-degradation drill explicitly omits `gepa.mjs`, `learn.mjs`, `redblue.mjs`, and `mint.mjs` from its skill list — their exit-0/`degraded:true` contract is unverified.\n\n",
+        "level": 3
+      },
+      {
+        "title": "Missing error-handling tests (highest-value finding)",
+        "content": "`audit-list.mjs`, `audit-trend.mjs`, `oia-audit.mjs::persist`, and `similarity.mjs` all call `spawnSync('npx', ...)` for memory-CLI access **without a timeout** — inconsistent with every other subprocess call in the plugin (which sets `timeoutMs`) — and no test has ever caught this. A hung `npx` would block these calls indefinitely.\n\n**A live regression bug was found during this analysis**: `threat-model.mjs:9` redeclares a stale local `SEVERITY_RANK` map missing `info`/`warn`/`error`/`critical` — the exact bug class that iter 62/63 fixed everywhere else by centralizing the map in `_harness.mjs`. Every sibling script (`mcp-scan.mjs`, `oia-audit.mjs`, `audit-trend.mjs`) imports the shared map; `threat-model.mjs` was missed. If upstream ever emits `worst: \"warn\"` or `\"critical\"`, `--fail-on` silently fails to trigger (`undefined >= threshold` → `false`).\n\n",
+        "level": 3
+      },
+      {
+        "title": "Integration gaps",
+        "content": "- `drift-from-history.mjs`'s `--baseline-since` variant and its no-flag default path are never exercised end-to-end.\n- `redblue.mjs`'s `init`/`run`/`patch`/`report` subcommands (including the documented $0 `--mock-judge` CI path) have never been run by any test — only `attack` once, through the MCP layer.\n- MCP tool surface vs. raw CLI surface isn't symmetric: some scripts (`mint.mjs`, `audit-list.mjs`) have no MCP tool *and* no CLI-level test — a double gap.\n\n",
+        "level": 3
+      },
+      {
+        "title": "Test skeletons",
+        "content": "The full report includes 9 ready-to-adapt skeletons (matching the existing plain-`assert` idiom) for: `_invoke.mjs` primitives, `_darwin.mjs::runDarwin`, a `threat-model.mjs` severity-rank regression lock, a static check that all memory-CLI `spawnSync` calls set a timeout, `gepa.mjs` op dispatch, `security-bench.mjs`'s markdown parser, `redblue.mjs` subcommand dispatch, `oia-audit.mjs::persist` failure path, and `drift-from-history.mjs --baseline-since`.\n\nWant me to write any of these test files (e.g. `scripts/test-invoke.mjs` or the `threat-model.mjs` severity-rank fix) into the repo?",
+        "level": 3
+      }
+    ],
+    "codeBlocks": []
+  },
+  "durationMs": 361294,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T17:55:46.253Z",
+  "executionId": "testgaps_1783619384959_s5iiwf"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783620946263_d7ovai_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T18:15:46.263Z] 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_1783620946263_d7ovai_result.log

@@ -0,0 +1,52 @@
+[2026-07-09T18:21:13.824Z] RESULT
+============================================================
+{
+  "success": true,
+  "output": "## Test Coverage Analysis — `plugins/ruflo-metaharness`\n\nThis plugin has no formal test framework (no package.json, no jest/vitest) — coverage comes from ad hoc `test-*.mjs` scripts using a custom `assert(cond, label)` + `process.exit()` pattern across ~6,900 lines in `scripts/`. Full findings below.\n\n### 1. Untested functions and classes\n\n**`_invoke.mjs`** — the shared plumbing used by ~15 scripts has **zero direct unit tests**: `classifyDegraded`, `injectJson`, `parseTrailingJson`, `satisfiesTildeRange`, `findLocalPackageDir`, `ensureCachedInstall`, `importOptionalLibrary`, `makeDegradedEmitter` are only exercised indirectly (and only the degraded branch, via subprocess string-matching).\n\n**`_harness.mjs`** — `runMetaharness`/`runHarness`/`runMetaharnessAsync`/`runHarnessAsync`/`metaharnessResolution`/`emitDegradedJsonAndExit` are untested as functions (only side effects checked transitively).\n\n**CLI scripts with zero or degraded-only coverage**: `mint.mjs` (its `safetyChecks()` root-refusal logic is completely unverified — notable since it's write-capable), `redblue.mjs` (only `attack` generically probed; `init`/`run`/`patch`/`report` untested), `learn.mjs`, `gepa.mjs`'s `validate`/`render`/`analyze` ops, `audit-list.mjs` (never invoked directly by any test), `evolve.mjs`'s pure helpers (`looksLikeGepaTranscript`, `extractGepaTranscripts`, `summarizeTraces` — zero-cost, zero-subprocess, yet untested), `security-bench.mjs`'s `parseSecurityBenchMarkdown` (a 3-regex parser with no analog to the test that exists for its sibling `bench-parse-mcp-scan.mjs`).\n\n`_spike-similarity.mjs` has its own built-in invariant checks but isn't wired into any `test-*.mjs` — a regression would only surface if someone runs it manually.\n\nExample skeleton (pattern repeats for each untested export):\n```js\n// test-invoke.mjs — zero-coverage core plumbing\nimport { classifyDegraded, injectJson, parseTrailingJson, satisfiesTildeRange } from './_invoke.mjs';\nlet passed=0, failed=0;\nfunction assert(c,l){ console.log(`  ${c?'✓':'✗'} ${l}`); c?passed++:failed++; }\nassert(classifyDegraded('', null, 'x').reason === 'x-timeout', 'null exitCode -> timeout');\nassert(parseTrailingJson('progress {\"a\":1}\\ndone {\"b\":2}').b === 2, 'grabs LAST json block');\nassert(satisfiesTildeRange('0.3.2', '~0.3.0') === true, 'tilde range patch match');\nprocess.exit(failed ? 1 : 0);\n```\n\n### 2. Edge cases not covered\n\n- `_similarity.mjs` `cosine`/`jaccard`: no test for `NaN`/`Infinity` inputs, negative scores, or duplicate `agent_topology` entries.\n- `parseMcpScanText`: no test for malformed `Result:` lines (non-numeric count → `NaN` propagates unasserted) or CRLF endings.\n- `mcp-scan.mjs --fail-on medium|low` and its invalid-value exit-2 path: never exercised (only default `high` tested).\n- `router-parallel-analyze.mjs`: no test for a partially-corrupt JSONL file (one malformed line would throw synchronously in `.map(JSON.parse)`).\n- `audit-trend.mjs`'s `verdict: 'unavailable'` fallback (pre-iter-38 records without fingerprints) is never triggered by any test.\n\n### 3. Missing error handling tests\n\n- **Likely latent bug, not just a test gap**: `audit-list.mjs`, `audit-trend.mjs`, and `similarity.mjs`'s `memList()`/`memRetrieve()` `spawnSync` calls have **no `timeout` option**, unlike every other subprocess call in this plugin family — a hung `npx ... memory list` would hang the script forever.\n- `oia-audit.mjs`'s `persist()` failure path (`payload.persisted.error`) is never forced/verified.\n- `mint.mjs`'s non-degraded failure branch (`metaharness new` exiting non-zero) is never simulated.\n- 9 of ~18 CLI scripts (`redblue`, `learn`, `gepa`, `mint`, `drift-from-history`, `audit-trend`, `audit-list`, `similarity`, `router-parallel-analyze`) are **excluded** from `test-graceful-degradation.mjs`'s skill list entirely — no automated \"upstream unavailable → exit 0 + degraded:true\" check for them.\n\n### 4. Integration test gaps\n\n- `drift-from-history.mjs`'s **default slow path** (no `--baseline-key`/`--baseline-file`, which actually calls `audit-list.mjs`) — the doc-advertised \"composes 3 primitives\" workflow — is never exercised; every existing test uses a fast-path flag that skips it.\n- `redblue.mjs`'s full `init → run → patch → report` lifecycle has no end-to-end test.\n- Of 15 registered MCP tools, only ~4 get real positive-data-shape assertions in `test-mcp-tools.mjs`; the rest (`redblue`, `learn`, `gepa`, `evolve`, `bench`, `security_bench`, `score`, `genome`, `threat_model`, `audit_list`) only get generic \"didn't throw\" checks.\n- The actual production \"weekly cron\" workflow (`oia-audit` persist → `audit-list` → `audit-trend` diff on **real**, non-fixture, non-dry-run data) has never been driven end-to-end.\n\n### 5. Top 15 prioritized new test scenarios (risk/value order)\n\n1. Unit-test `_invoke.mjs` directly — shared plumbing for ~15 scripts, currently zero direct coverage.\n2. Add `timeout` to `audit-list.mjs`/`audit-trend.mjs`/`similarity.mjs`'s `memList`/`memRetrieve` calls + regression test (hang risk).\n3. Fix `threat-model.mjs:9`'s local, out-of-sync `SEVERITY_RANK` literal (missing `info`/`warn`/`critical`/`error` vs. the shared one in `_harness.mjs`) — same bug class iter-63 fixed elsewhere.\n4. E2E test of `drift-from-history.mjs`'s default slow path.\n5. Add `mint.mjs` (write-capable, zero coverage) to the test suite — start with `safetyChecks()` in isolation.\n6. Unit-test `evolve.mjs`'s pure helpers (zero subprocess cost, currently untested).\n7. Unit-test `security-bench.mjs`'s `parseSecurityBenchMarkdown` against a realistic fixture.\n8. Extend `test-graceful-degradation.mjs` to cover the 9 currently-excluded scripts.\n9. Deepen `test-mcp-tools.mjs` Phase 4 with positive-data assertions for the 10 shallow-covered tools.\n10. Test `parseMcpScanText` against malformed `Result:` lines and CRLF input.\n11. Test `oia-audit.mjs`'s persist-failure surfacing.\n12. Real (non-dry-run) `oia-audit → audit-list → audit-trend` chain test.\n13. Test `_similarity.mjs` numeric edge cases (NaN/Infinity/negative/duplicates).\n14. Test `router-parallel-analyze.mjs` against partially-corrupt JSONL.\n15. Wire `_spike-similarity.mjs`'s invariant checks into the automated suite so CI catches ADR-152 regressions.\n\nNote: item 3 is a genuine bug find (not just missing coverage) — `threat-model.mjs` duplicates `SEVERITY_RANK` locally instead of importing the shared one, and its copy is missing values. Worth fixing regardless of test-writing priority. Want me to dig into that one now, or start writing any of these test skeletons for real?\n",
+  "parsedOutput": {
+    "sections": [
+      {
+        "title": "Test Coverage Analysis — `plugins/ruflo-metaharness`",
+        "content": "\nThis plugin has no formal test framework (no package.json, no jest/vitest) — coverage comes from ad hoc `test-*.mjs` scripts using a custom `assert(cond, label)` + `process.exit()` pattern across ~6,900 lines in `scripts/`. Full findings below.\n\n",
+        "level": 2
+      },
+      {
+        "title": "1. Untested functions and classes",
+        "content": "\n**`_invoke.mjs`** — the shared plumbing used by ~15 scripts has **zero direct unit tests**: `classifyDegraded`, `injectJson`, `parseTrailingJson`, `satisfiesTildeRange`, `findLocalPackageDir`, `ensureCachedInstall`, `importOptionalLibrary`, `makeDegradedEmitter` are only exercised indirectly (and only the degraded branch, via subprocess string-matching).\n\n**`_harness.mjs`** — `runMetaharness`/`runHarness`/`runMetaharnessAsync`/`runHarnessAsync`/`metaharnessResolution`/`emitDegradedJsonAndExit` are untested as functions (only side effects checked transitively).\n\n**CLI scripts with zero or degraded-only coverage**: `mint.mjs` (its `safetyChecks()` root-refusal logic is completely unverified — notable since it's write-capable), `redblue.mjs` (only `attack` generically probed; `init`/`run`/`patch`/`report` untested), `learn.mjs`, `gepa.mjs`'s `validate`/`render`/`analyze` ops, `audit-list.mjs` (never invoked directly by any test), `evolve.mjs`'s pure helpers (`looksLikeGepaTranscript`, `extractGepaTranscripts`, `summarizeTraces` — zero-cost, zero-subprocess, yet untested), `security-bench.mjs`'s `parseSecurityBenchMarkdown` (a 3-regex parser with no analog to the test that exists for its sibling `bench-parse-mcp-scan.mjs`).\n\n`_spike-similarity.mjs` has its own built-in invariant checks but isn't wired into any `test-*.mjs` — a regression would only surface if someone runs it manually.\n\nExample skeleton (pattern repeats for each untested export):\n```js\n// test-invoke.mjs — zero-coverage core plumbing\nimport { classifyDegraded, injectJson, parseTrailingJson, satisfiesTildeRange } from './_invoke.mjs';\nlet passed=0, failed=0;\nfunction assert(c,l){ console.log(`  ${c?'✓':'✗'} ${l}`); c?passed++:failed++; }\nassert(classifyDegraded('', null, 'x').reason === 'x-timeout', 'null exitCode -> timeout');\nassert(parseTrailingJson('progress {\"a\":1}\\ndone {\"b\":2}').b === 2, 'grabs LAST json block');\nassert(satisfiesTildeRange('0.3.2', '~0.3.0') === true, 'tilde range patch match');\nprocess.exit(failed ? 1 : 0);\n```\n\n",
+        "level": 3
+      },
+      {
+        "title": "2. Edge cases not covered",
+        "content": "\n- `_similarity.mjs` `cosine`/`jaccard`: no test for `NaN`/`Infinity` inputs, negative scores, or duplicate `agent_topology` entries.\n- `parseMcpScanText`: no test for malformed `Result:` lines (non-numeric count → `NaN` propagates unasserted) or CRLF endings.\n- `mcp-scan.mjs --fail-on medium|low` and its invalid-value exit-2 path: never exercised (only default `high` tested).\n- `router-parallel-analyze.mjs`: no test for a partially-corrupt JSONL file (one malformed line would throw synchronously in `.map(JSON.parse)`).\n- `audit-trend.mjs`'s `verdict: 'unavailable'` fallback (pre-iter-38 records without fingerprints) is never triggered by any test.\n\n",
+        "level": 3
+      },
+      {
+        "title": "3. Missing error handling tests",
+        "content": "\n- **Likely latent bug, not just a test gap**: `audit-list.mjs`, `audit-trend.mjs`, and `similarity.mjs`'s `memList()`/`memRetrieve()` `spawnSync` calls have **no `timeout` option**, unlike every other subprocess call in this plugin family — a hung `npx ... memory list` would hang the script forever.\n- `oia-audit.mjs`'s `persist()` failure path (`payload.persisted.error`) is never forced/verified.\n- `mint.mjs`'s non-degraded failure branch (`metaharness new` exiting non-zero) is never simulated.\n- 9 of ~18 CLI scripts (`redblue`, `learn`, `gepa`, `mint`, `drift-from-history`, `audit-trend`, `audit-list`, `similarity`, `router-parallel-analyze`) are **excluded** from `test-graceful-degradation.mjs`'s skill list entirely — no automated \"upstream unavailable → exit 0 + degraded:true\" check for them.\n\n",
+        "level": 3
+      },
+      {
+        "title": "4. Integration test gaps",
+        "content": "\n- `drift-from-history.mjs`'s **default slow path** (no `--baseline-key`/`--baseline-file`, which actually calls `audit-list.mjs`) — the doc-advertised \"composes 3 primitives\" workflow — is never exercised; every existing test uses a fast-path flag that skips it.\n- `redblue.mjs`'s full `init → run → patch → report` lifecycle has no end-to-end test.\n- Of 15 registered MCP tools, only ~4 get real positive-data-shape assertions in `test-mcp-tools.mjs`; the rest (`redblue`, `learn`, `gepa`, `evolve`, `bench`, `security_bench`, `score`, `genome`, `threat_model`, `audit_list`) only get generic \"didn't throw\" checks.\n- The actual production \"weekly cron\" workflow (`oia-audit` persist → `audit-list` → `audit-trend` diff on **real**, non-fixture, non-dry-run data) has never been driven end-to-end.\n\n",
+        "level": 3
+      },
+      {
+        "title": "5. Top 15 prioritized new test scenarios (risk/value order)",
+        "content": "1. Unit-test `_invoke.mjs` directly — shared plumbing for ~15 scripts, currently zero direct coverage.\n2. Add `timeout` to `audit-list.mjs`/`audit-trend.mjs`/`similarity.mjs`'s `memList`/`memRetrieve` calls + regression test (hang risk).\n3. Fix `threat-model.mjs:9`'s local, out-of-sync `SEVERITY_RANK` literal (missing `info`/`warn`/`critical`/`error` vs. the shared one in `_harness.mjs`) — same bug class iter-63 fixed elsewhere.\n4. E2E test of `drift-from-history.mjs`'s default slow path.\n5. Add `mint.mjs` (write-capable, zero coverage) to the test suite — start with `safetyChecks()` in isolation.\n6. Unit-test `evolve.mjs`'s pure helpers (zero subprocess cost, currently untested).\n7. Unit-test `security-bench.mjs`'s `parseSecurityBenchMarkdown` against a realistic fixture.\n8. Extend `test-graceful-degradation.mjs` to cover the 9 currently-excluded scripts.\n9. Deepen `test-mcp-tools.mjs` Phase 4 with positive-data assertions for the 10 shallow-covered tools.\n10. Test `parseMcpScanText` against malformed `Result:` lines and CRLF input.\n11. Test `oia-audit.mjs`'s persist-failure surfacing.\n12. Real (non-dry-run) `oia-audit → audit-list → audit-trend` chain test.\n13. Test `_similarity.mjs` numeric edge cases (NaN/Infinity/negative/duplicates).\n14. Test `router-parallel-analyze.mjs` against partially-corrupt JSONL.\n15. Wire `_spike-similarity.mjs`'s invariant checks into the automated suite so CI catches ADR-152 regressions.\n\nNote: item 3 is a genuine bug find (not just missing coverage) — `threat-model.mjs` duplicates `SEVERITY_RANK` locally instead of importing the shared one, and its copy is missing values. Worth fixing regardless of test-writing priority. Want me to dig into that one now, or start writing any of these test skeletons for real?",
+        "level": 3
+      }
+    ],
+    "codeBlocks": [
+      {
+        "language": "js",
+        "code": "// test-invoke.mjs — zero-coverage core plumbing\nimport { classifyDegraded, injectJson, parseTrailingJson, satisfiesTildeRange } from './_invoke.mjs';\nlet passed=0, failed=0;\nfunction assert(c,l){ console.log(`  ${c?'✓':'✗'} ${l}`); c?passed++:failed++; }\nassert(classifyDegraded('', null, 'x').reason === 'x-timeout', 'null exitCode -> timeout');\nassert(parseTrailingJson('progress {\"a\":1}\\ndone {\"b\":2}').b === 2, 'grabs LAST json block');\nassert(satisfiesTildeRange('0.3.2', '~0.3.0') === true, 'tilde range patch match');\nprocess.exit(failed ? 1 : 0);"
+      }
+    ]
+  },
+  "durationMs": 327560,
+  "model": "sonnet",
+  "sandboxMode": "permissive",
+  "workerType": "testgaps",
+  "timestamp": "2026-07-09T18:21:13.823Z",
+  "executionId": "testgaps_1783620946263_d7ovai"
+}

package/plugins/ruflo-metaharness/.claude-flow/logs/headless/testgaps_1783622473846_t839e5_prompt.log

@@ -0,0 +1,14 @@
+[2026-07-09T18:41:13.847Z] 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.