@claude-flow/cli 3.64.0 → 3.65.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@claude-flow/cli",
-  "version": "3.64.0",
+  "version": "3.65.0",
   "type": "module",
   "description": "Ruflo CLI - Enterprise AI agent orchestration with 60+ specialized agents, swarm coordination, MCP server, self-learning hooks, and vector memory for Claude Code",
   "main": "dist/src/index.js",
@@ -86,6 +86,7 @@
     "build": "tsc",
     "check:metaharness-pins": "node ../../../scripts/check-metaharness-pins.mjs",
     "test": "vitest run",
+    "test:real-binary": "vitest run --config vitest.real-binary.config.ts",
     "test:plugin-store": "npx tsx src/plugins/tests/standalone-test.ts",
     "test:pattern-store": "npx tsx src/transfer/store/tests/standalone-test.ts",
     "postinstall": "node ./scripts/postinstall.cjs",

package/plugins/ruflo-metaharness/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "ruflo-metaharness",
   "description": "MetaHarness integration for ruflo — surfaces score/genome/mint/mcp-scan/threat-model via skills; pairs with @metaharness/router (ADR-148/149) for cost-optimal model routing; honors ADR-150's architectural constraint that MetaHarness remains an optional augmentation, never a required runtime dependency. As a mod (ADR-445): confirms flywheel promotion, keeps secrets out of inputs, /metaharness-mod, and a status file the console shows",
-  "version": "0.2.6",
+  "version": "0.4.0",
   "author": {
     "name": "ruvnet",
     "url": "https://github.com/ruvnet"
@@ -43,6 +43,13 @@
         "on",
         "off"
       ]
+    },
+    "openrouterApiKey": {
+      "type": "string",
+      "title": "OpenRouter API key",
+      "description": "Key for the real model judge in redblue runs (--mock-judge needs none). Kept in the credential store, never in settings; OPENROUTER_API_KEY still works when this is empty.",
+      "required": false,
+      "sensitive": true
     }
   }
 }

package/plugins/ruflo-metaharness/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 Newest first. One `## <version> — <date>` heading per version, then `feat:`, `fix:`, `breaking:` and `chore:` bullets (ADR-478). Built from git history; this is the whole history.
 
+## 0.4.0 — 2026-10-29
+- chore: operational skills are user-invoked only (`disable-model-invocation: true`) and long skill/command descriptions are cut to one sentence, so they cost less prompt on every turn (ADR-507, #4588). Every skill is still available as `/<skill>`.
+
+## 0.3.0 — 2026-10-29
+- feat: `openrouterApiKey` is a `userConfig` option with `sensitive: true` (ADR-506, #4587), read through `readSecret` before the `OPENROUTER_API_KEY` environment fallback.
+
 ## 0.2.6 — 2026-10-12
 - fix: the shared text screen now catches `.env` and e-mailed credentials, no longer flags a model's token talk (`--max-tokens`, `/oauth/token`, "token usage"), and no longer flags values that only name where a secret lives (`./service-account.json`, `--set-secrets=NAME=NAME:latest`); every rule stays linear (#4094, thanks @proffesor-for-testing)
 

package/plugins/ruflo-metaharness/README.md

@@ -29,6 +29,19 @@ MetaHarness integration plugin for ruflo. Surfaces the upstream `metaharness` /
 | `harness-learn` | `/harness-learn --host <h> --model <m> --slice <manifest> [--repo <checkout>] [--run]` | metaharness@0.3.0 / upstream ADR-235 — GEPA learning run; $0 dry-run default, `--run` to spend; needs a metaharness repo checkout |
 | `harness-gepa` | `/harness-gepa --op genome\|validate\|render\|analyze [--path <genome.json>]` | darwin@0.8.0 GEPA library surface — genome load/validate/render + transcript failure analysis; `gepaOptimize` stays library-only |
 
+## User-invoked skills (ADR-507)
+
+These skills carry `disable-model-invocation: true`: Claude does not pick them on its own, and they cost nothing in the per-turn skill listing. Type them to run them. They run benchmarks, spend-gated learning or evolution runs, scheduled audits, or a one-shot scaffold.
+
+| Skill | Run it with |
+|---|---|
+| harness-bench | `/ruflo-metaharness:harness-bench` |
+| harness-evolve | `/ruflo-metaharness:harness-evolve` |
+| harness-learn | `/ruflo-metaharness:harness-learn` |
+| harness-mint | `/ruflo-metaharness:harness-mint` |
+| harness-oia-audit | `/ruflo-metaharness:harness-oia-audit` |
+| harness-security-bench | `/ruflo-metaharness:harness-security-bench` |
+
 ## Phase-0 baseline (ruflo itself, 2026-06-16)
 
 ```json
@@ -79,6 +92,6 @@ This plugin ships a function-hook mod (ADR-445, `hooks/register.ts`). It makes n
 - **`/metaharness-mod`**: `status`, `recent`, `tools` (which of this plugin's tools are connected) and `scan <text>` (would the guard refuse it). Answered locally.
 - **Status file**: `.claude-flow/metaharness-mod/status.json` (`{version: 1, updatedMs, guard, calls, total, blocked, recent}`, counters only, never tool input), written at session start and after every call to this plugin's tools.
 
-Options (`userConfig`): `guard` (on).
+Options (`userConfig`): `guard` (on); `openrouterApiKey` (sensitive, empty by default): the key for redblue's real model judge, kept in the credential store. A classic SessionStart hook exports it as `OPENROUTER_API_KEY` to the session's Bash commands when that variable is unset, so a redblue run started from a skill uses it; it never goes in argv. The `metaharness_redblue` MCP tool runs inside ruflo-core's MCP server, which does not see it: export `OPENROUTER_API_KEY` before starting Claude Code for that path (ADR-506).
 
 Test it: `claude plugin validate plugins/ruflo-metaharness`, `claude plugin test plugins/ruflo-metaharness`, `bash plugins/ruflo-metaharness/scripts/smoke.sh`.

package/plugins/ruflo-metaharness/agents/metaharness-architect.md

@@ -36,6 +36,8 @@ needs its own ADR.
 | `harness-oia-audit` | Composite weekly audit worker (iter 7) | Cron-scheduled; bundles oia+threat+mcp into one timestamped record in `metaharness-audit` namespace |
 | `harness-mint` | Scaffold a custom harness | User wants to fork; ALWAYS dry-run first, never write to project root |
 
+`harness-oia-audit` and `harness-mint` are user-invoked (ADR-507), as are `harness-bench`, `harness-evolve`, `harness-learn` and `harness-security-bench`: the Skill tool refuses them. Tell the user which `/ruflo-metaharness:<skill>` to run rather than running it yourself.
+
 ## Tools (subprocess-only, no library imports)
 
 - All skills shell out to the pinned `metaharness` / `harness` binaries (`metaharness@~0.4.1`, local install or one-time versioned cache — never `@latest`) via the `_harness.mjs` shared helper.

package/plugins/ruflo-metaharness/hooks/hooks.json

@@ -1,4 +1,18 @@
 {
-  "description": "MetaHarness as a mod (ADR-445): a tighten-only guard on the metaharness tools (promotion needs an explicit confirm, no secrets in inputs), a status file the console reads, and /metaharness-mod.",
-  "modules": ["./register.ts"]
+  "description": "MetaHarness as a mod (ADR-445): a tighten-only guard on the metaharness tools (promotion needs an explicit confirm, no secrets in inputs), a status file the console reads, and /metaharness-mod. The classic SessionStart hook hands the sensitive option openrouterApiKey to the Bash tool as OPENROUTER_API_KEY through CLAUDE_ENV_FILE, only when that variable is not already set, so a redblue run started from a skill can use a key kept in the credential store (scripts/_secret-env.cjs, ADR-506).",
+  "hooks": {
+    "SessionStart": [
+      {
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node -e \"require(require('path').join(process.env.CLAUDE_PLUGIN_ROOT,'scripts','_secret-env.cjs')).bridge({openrouterApiKey:'OPENROUTER_API_KEY'})\""
+          }
+        ]
+      }
+    ]
+  },
+  "modules": [
+    "./register.ts"
+  ]
 }

package/plugins/ruflo-metaharness/scripts/_redblue.mjs

@@ -54,7 +54,9 @@
 //   (no real creds, no live targets, no shell, no eval, no arbitrary network).
 //   This wrapper does NOT relax those — it only forwards argv with shell:false.
 //   `--mock-judge` is the $0 CI path; the real model judge gates on
-//   $OPENROUTER_API_KEY which we never inject.
+//   $OPENROUTER_API_KEY. The wrapper never puts a key in argv. When that
+//   variable is unset but the plugin option `openrouterApiKey` (sensitive,
+//   ADR-506) is, the child gets it as OPENROUTER_API_KEY in its environment.
 
 import { spawnSync } from 'node:child_process';
 import { realpathSync } from 'node:fs';
@@ -67,6 +69,7 @@ import {
   makeDegradedEmitter,
   resolvePackageBin,
 } from './_invoke.mjs';
+import { readSecret } from './_secret.mjs';
 
 const DEFAULT_TIMEOUT_MS = 120_000;
 
@@ -78,6 +81,13 @@ const REDBLUE_PIN = `${REDBLUE_PKG}@${REDBLUE_PIN_VERSION}`;
 
 const REASON_PREFIX = 'metaharness-redblue';
 
+/** The child environment, with OPENROUTER_API_KEY filled from the plugin option when the variable is unset (ADR-506). */
+export function judgeEnv(env) {
+  if (env.OPENROUTER_API_KEY) return env;
+  const key = readSecret('openrouterApiKey', [], env);
+  return key ? { ...env, OPENROUTER_API_KEY: key } : env;
+}
+
 // Cache dir is versioned by the pin so bumping REDBLUE_PIN_VERSION
 // invalidates stale installs (handled inside ensureCachedInstall). The
 // resolved layout matches the pre-consolidation helper so existing caches
@@ -133,7 +143,7 @@ export function runRedblue(args, opts = {}) {
     encoding: 'utf-8',
     timeout: timeoutMs,
     cwd: opts.cwd,
-    env: { ...process.env, ...(opts.env || {}) },
+    env: judgeEnv({ ...process.env, ...(opts.env || {}) }),
     shell: false,
   });
   const durationMs = Date.now() - start;

package/plugins/ruflo-metaharness/scripts/_secret-env.cjs

@@ -0,0 +1,50 @@
+// Secret bridge (ADR-506): hand a sensitive `userConfig` option to the scripts a skill or command runs
+// through the Bash tool. Copied verbatim into each plugin that needs it (a test keeps the copies identical).
+//
+// Claude Code gives a plugin's option only to that plugin's hooks, as CLAUDE_PLUGIN_OPTION_<KEY>; the Bash
+// tool never sees it. A SessionStart hook, though, gets CLAUDE_ENV_FILE, and the lines it appends there are
+// loaded into every later Bash tool command of the session. So the plugin's SessionStart hook calls
+// bridge({ option: 'VARIABLE' }) and the script reads VARIABLE as it always did.
+//
+// Exposure: the value then sits in the session's environment file (under ~/.claude, owner-only) and in the
+// environment of the session's Bash commands, the same as exporting it in a shell profile. What the option
+// adds is masked entry and credential-store storage instead of a plain-text profile or .env file.
+//
+// Rules: a variable the user already exported wins and is never overwritten; the value is never printed,
+// logged or put in argv; a value with a newline or NUL is refused; any failure is silent and never blocks
+// the session.
+'use strict'
+const fs = require('fs')
+
+const optionEnvName = (name) => `CLAUDE_PLUGIN_OPTION_${String(name).replace(/[^A-Za-z0-9_]/g, '_').toUpperCase()}`
+const VARIABLE = /^[A-Za-z_][A-Za-z0-9_]*$/
+const quote = (value) => `'${value.replace(/'/g, `'\\''`)}'`
+
+/** The env-file lines for the options that are set and whose variable is not. Never contains a value that was not asked for. */
+function bridgeLines(map, env = process.env) {
+  const lines = []
+  for (const [option, variable] of Object.entries(map || {})) {
+    if (!VARIABLE.test(variable)) continue
+    if (typeof env[variable] === 'string' && env[variable].trim() !== '') continue
+    const value = env[optionEnvName(option)]
+    if (typeof value !== 'string' || value.trim() === '' || /[\n\r\0]/.test(value)) continue
+    lines.push(`export ${variable}=${quote(value)}`)
+  }
+  return lines
+}
+
+/** Append the lines to CLAUDE_ENV_FILE (owner-only). Returns the variable NAMES written, never values. */
+function bridge(map, env = process.env) {
+  try {
+    const file = env.CLAUDE_ENV_FILE
+    if (!file) return []
+    const lines = bridgeLines(map, env)
+    if (lines.length === 0) return []
+    fs.appendFileSync(file, lines.join('\n') + '\n', { mode: 0o600 })
+    return lines.map((line) => line.slice('export '.length, line.indexOf('=')))
+  } catch {
+    return []
+  }
+}
+
+module.exports = { bridge, bridgeLines, optionEnvName }

package/plugins/ruflo-metaharness/scripts/_secret.mjs

@@ -0,0 +1,35 @@
+// readSecret: one way to read a plugin credential (ADR-506). Copied verbatim into each plugin that
+// needs it, since a plugin may not import from another plugin's folder; keep the copies identical
+// (a test compares them).
+//
+// Order: CLAUDE_PLUGIN_OPTION_<NAME>, which Claude Code sets for a plugin's hooks when the
+// `userConfig` option (declared with "sensitive": true, kept in the credential store) is filled in;
+// then the environment variables the plugin always read, for standalone CLI use and backward
+// compatibility. The value is never logged, echoed, or placed in an error message.
+
+/** The environment variable Claude Code sets for a `userConfig` key (non-identifier characters to `_`, upper-cased). */
+export const optionEnvName = (name) => `CLAUDE_PLUGIN_OPTION_${String(name).replace(/[^A-Za-z0-9_]/g, '_').toUpperCase()}`
+
+/** The secret, or undefined when neither the option nor any fallback variable holds a non-empty value. Read at call time. */
+export function readSecret(name, envFallbacks = [], env = process.env) {
+  for (const variable of [optionEnvName(name), ...envFallbacks]) {
+    const value = env[variable]
+    if (typeof value === 'string' && value.trim() !== '') return value
+  }
+  return undefined
+}
+
+/** Where the secret was found (a variable NAME, never the value), or undefined. */
+export function secretSource(name, envFallbacks = [], env = process.env) {
+  for (const variable of [optionEnvName(name), ...envFallbacks]) {
+    const value = env[variable]
+    if (typeof value === 'string' && value.trim() !== '') return variable
+  }
+  return undefined
+}
+
+/** Error text for a missing secret: names the option and the variables, never a value. */
+export function secretHint(name, envFallbacks = []) {
+  const vars = envFallbacks.length ? ` or export ${envFallbacks.join(' / ')} in your shell` : ''
+  return `no ${name} configured: set the plugin option "${name}" with /plugin (it is stored in the credential store)${vars}. Never paste a key into a prompt or a file.`
+}

package/plugins/ruflo-metaharness/scripts/smoke.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# Structural smoke test for ruflo-metaharness v0.2.6 (ADR-150 Phase 1).
+# Structural smoke test for ruflo-metaharness v0.4.0 (ADR-150 Phase 1).
 set -u
 ROOT="$(cd "$(dirname "$0")/.." && pwd)"
 PASS=0
@@ -32,10 +32,10 @@ else
   bad "extraction-regex-rot: EXPECTED_TOOLS=$EXPECTED_TOOLS EXPECTED_SUBS=$EXPECTED_SUBS"
 fi
 
-step "1. plugin.json declares 0.2.6 with adr-150 keywords"
+step "1. plugin.json declares 0.4.0 with adr-150 keywords"
 v=$(grep -E '"version"' "$ROOT/.claude-plugin/plugin.json" | grep -oE '[0-9]+\.[0-9]+\.[0-9]+' | head -1)
-if [[ "$v" != "0.2.6" ]]; then
-  bad "expected 0.2.6, got '$v'"
+if [[ "$v" != "0.4.0" ]]; then
+  bad "expected 0.4.0, got '$v'"
 else
   miss=""
   for k in ruflo metaharness harness scorecard genome mcp-scan threat-model router adr-150 adr-148 adr-149 optional-dependency graceful-degradation subprocess phase-1-mvp; do
@@ -483,7 +483,7 @@ step "17z74a. plugin helper pins accept the CLI-declared ranges — no tool-time
 miss=""
 # 17z74 compares package.json files with each other; the helpers' own
 # *_PIN_VERSION constants (what findLocalPackageDir accepts) were never
-# compared with them and drifted (~0.3.0 / ~0.8.0 vs declared ^0.4.1 /
+# compared with them and drifted (~0.4.0 / ~0.8.0 vs declared ^0.4.1 /
 # ~0.10.2), so every metaharness_* call npm-installed an older release next
 # to the one ruflo ships. Hermetic runtime gate: stub packages at the declared
 # versions, published layout, npm/npx trap.
@@ -1287,7 +1287,7 @@ miss=""
 DISP="$ROOT/../../v3/@claude-flow/cli/src/commands/metaharness.ts"
 WRAPPER="$ROOT/../../v3/@claude-flow/cli/src/mcp-tools/metaharness-tools.ts"
 # All 13 dispatchable subcommands present (iter-73 had 10; @metaharness/redblue
-# added redblue.mjs; metaharness@0.3.0/darwin@0.8.0 added learn.mjs + gepa.mjs).
+# added redblue.mjs; metaharness@0.4.0/darwin@0.8.0 added learn.mjs + gepa.mjs).
 for sub in score genome mcp-scan threat-model oia-audit audit-list audit-trend similarity drift-from-history mint redblue learn gepa; do
   grep -q "${sub}" "$DISP" 2>/dev/null || miss="$miss subcommand-${sub}-not-listed"
 done
@@ -2199,7 +2199,7 @@ grep -q "result has 'degraded'" "$F" || miss="$miss no-degraded-assertion"
 grep -q "result has 'exitCode'" "$F" || miss="$miss no-exitcode-assertion"
 # All tool names enumerated (similarity iter 36, drift_from_history iter 54,
 # ADR-153 added bench/evolve/security_bench, @metaharness/redblue added redblue,
-# metaharness@0.3.0/darwin@0.8.0 added learn + gepa, ADR-322 added flywheel)
+# metaharness@0.4.0/darwin@0.8.0 added learn + gepa, ADR-322 added flywheel)
 for tool in metaharness_score metaharness_genome metaharness_mcp_scan metaharness_threat_model metaharness_oia_audit metaharness_audit_list metaharness_audit_trend metaharness_similarity metaharness_drift_from_history metaharness_bench metaharness_evolve metaharness_security_bench metaharness_redblue metaharness_learn metaharness_gepa metaharness_flywheel; do
   grep -q "${tool}" "$F" || miss="$miss missing-${tool}"
 done

package/plugins/ruflo-metaharness/skills/harness-bench/SKILL.md

@@ -1,8 +1,9 @@
 ---
 name: harness-bench
-description: Manage `@metaharness/darwin` bench suites — `bench create {repo}` scaffolds a JSON suite from a repo's test corpus; `bench verify {suite.json}` checks suite well-formedness. Bench suites are the fixed evaluation corpora that `harness-evolve --bench {suite.json}` scores variants against, decoupling evolution from the repo's natural tests. Degrades gracefully when @metaharness/darwin is absent.
+description: Create (`bench create {repo}`) or verify (`bench verify {suite.json}`) a @metaharness/darwin bench suite, the fixed corpus `harness-evolve --bench` scores variants against. Use when preparing an evolution benchmark. Degrades gracefully without darwin.
 argument-hint: "--op create --repo <path> [--out <path>]  |  --op verify --suite <path>"
 allowed-tools: Bash
+disable-model-invocation: true
 ---
 
 Surfaces `metaharness-darwin bench <create|verify>` — the supporting verb

package/plugins/ruflo-metaharness/skills/harness-drift-from-history/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: harness-drift-from-history
-description: One-command drift detection. Composes audit-list + oia-audit + audit-trend into a single primitive — finds the most recent audit in `metaharness-audit` namespace, runs a fresh audit against the current repo, diffs them via ADR-152 §3.1 similarity, and alerts when structural distance crosses `--threshold`. Iter 53 of ADR-150 deep integration.
+description: "One-command drift detection: run a fresh metaharness audit, diff it against the latest stored audit by ADR-152 similarity, and alert when the distance crosses `--threshold`. Use when checking whether a harness has drifted since its last audit."
 argument-hint: "[--path .] [--baseline-since 7d] [--threshold 0.95] [--dry-run] [--format json|table]"
 allowed-tools: Bash
 ---

package/plugins/ruflo-metaharness/skills/harness-evolve/SKILL.md

@@ -1,8 +1,9 @@
 ---
 name: harness-evolve
-description: Run `@metaharness/darwin evolve {repo}` to mutate a harness's seven policy surfaces (planner/contextBuilder/reviewer/retryPolicy/toolPolicy/memoryPolicy/scorePolicy), sandbox-score each variant, and promote only measured wins. The model is frozen; the harness evolves. Closes the loop ADR-150 opens (score+genome describe; evolve changes). Degrades gracefully when @metaharness/darwin is absent (ADR-150 + ADR-153 architectural constraints).
+description: "Run `@metaharness/darwin evolve {repo}`: mutate a harness's seven policy surfaces, sandbox-score each variant and promote only measured wins. Use when evolving a harness against its tests or a bench suite. Degrades gracefully without darwin (ADR-150/153)."
 argument-hint: "--repo <path> [--generations 3] [--children 3] [--concurrency 2] [--sandbox real|mock|agent] [--selection pareto|quality-diversity|...] [--mutator deterministic|ruvllm] [--diagnose] [--confirm]"
 allowed-tools: Bash
+disable-model-invocation: true
 ---
 
 Surfaces the upstream `metaharness-darwin evolve` CLI as a ruflo skill. The

package/plugins/ruflo-metaharness/skills/harness-gepa/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: harness-gepa
-description: "Inspect and audit GEPA genomes via the `@metaharness/darwin/gepa` library entry (darwin 0.8.0) — load/validate a genome (default is the shipped cand-6 promotion), render the system prompt a genome compiles to, or classify failure modes in a run transcript. The `gepaOptimize` loop itself is library-only (bring your own evaluator) and not surfaced here — use `harness-evolve` for sandbox-scored evolution. Degrades gracefully when @metaharness/darwin is absent."
+description: Load, validate or render a GEPA genome, or classify failure modes in a run transcript, via @metaharness/darwin/gepa. Use when inspecting or auditing a genome; use harness-evolve for sandbox-scored evolution. Degrades gracefully without darwin.
 argument-hint: "--op genome|validate|render|analyze [--path <genome.json>] [--transcript <t.json>] [--alert-on-invalid]"
 allowed-tools: Bash
 ---

package/plugins/ruflo-metaharness/skills/harness-learn/SKILL.md

@@ -1,8 +1,9 @@
 ---
 name: harness-learn
-description: Run a GEPA learning cycle via `metaharness learn` (upstream ADR-235, metaharness@0.3.0) — optimizes a harness genome against a SWE-bench-style slice manifest. $0 dry-run by default; `--run` is the explicit spend opt-in. Requires a metaharness repo checkout (`--repo` or $METAHARNESS_REPO) — without one it reports `checkout-required` with clone instructions. Degrades gracefully when metaharness is absent.
+description: Run a GEPA learning cycle via `metaharness learn`, optimizing a harness genome against a SWE-bench-style slice. Use when training a harness genome. $0 dry-run by default; `--run` opts into spend. Needs a metaharness checkout (--repo or $METAHARNESS_REPO).
 argument-hint: "--host <h> --model <m> --slice <manifest> [--repo <checkout>] [--run] [--alert-on-fail]"
 allowed-tools: Bash
+disable-model-invocation: true
 ---
 
 Surfaces `metaharness learn` — the upstream GEPA learning harness that

package/plugins/ruflo-metaharness/skills/harness-mint/SKILL.md

@@ -1,8 +1,9 @@
 ---
 name: harness-mint
-description: Scaffold a custom AI agent harness via `metaharness new {name} --template {id} --host {id}`. Defaults to DRY-RUN (no writes) unless --confirm is passed. Refuses to write to the calling repo root or anywhere inside it. Honors ADR-150 architectural constraint + ruflo's "destructive-action confirmation" pattern.
+description: Scaffold a custom AI agent harness via `metaharness new {name} --template {id} --host {id}`. Use when the user wants a new standalone harness. Dry-run unless --confirm is passed, and refuses to write inside the calling repo.
 argument-hint: "--name <id> --template <vertical:coding|minimal|…> [--host claude-code|codex|…] [--target /abs/path] [--confirm] [--format table|json]"
 allowed-tools: Bash
+disable-model-invocation: true
 ---
 
 The one write-capable skill in the plugin. Every other skill is

package/plugins/ruflo-metaharness/skills/harness-oia-audit/SKILL.md

@@ -3,6 +3,7 @@ name: harness-oia-audit
 description: Composite Phase-2 audit worker (ADR-150). Bundles harness oia-manifest + threat-model + mcp-scan into one timestamped audit record stored in the `metaharness-audit` memory namespace. Designed for cron-scheduled drift detection.
 argument-hint: "[--path .] [--dry-run] [--alert-on-worst clean|low|medium|high] [--format table|json]"
 allowed-tools: Bash
+disable-model-invocation: true
 ---
 
 The 13th worker (ADR-150 Phase 2) — runs three MetaHarness static

package/plugins/ruflo-metaharness/skills/harness-score/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: harness-score
-description: 5-dimension harness readiness scorecard from `metaharness score {path}`. Returns harnessFit / compileConfidence / taskCoverage / toolSafety / memoryUsefulness + estCostPerRunUsd + scaffoldReady. Pure-read; subprocess invocation; degrades gracefully when MetaHarness is absent (ADR-150 architectural constraint).
+description: 5-dimension harness readiness scorecard from `metaharness score {path}` (harnessFit, compileConfidence, taskCoverage, toolSafety, memoryUsefulness, plus cost per run). Use when rating a repo as an agent harness. Pure-read; degrades gracefully without MetaHarness.
 argument-hint: "[--path .] [--alert-on-fit-below 70] [--format table|json]"
 allowed-tools: Bash
 ---

package/plugins/ruflo-metaharness/skills/harness-security-bench/SKILL.md

@@ -1,8 +1,9 @@
 ---
 name: harness-security-bench
-description: Run `@metaharness/darwin security bench` (upstream "Darwin Shield" / ADR-155) — evolves a champion security-detection harness against a 10-vuln / 9-decoy corpus and grades it on TPR/FPR/patch-pass/repro/unsafe vs four baselines (B0 static, B1 LLM-single-pass, B2 fixed-agent, B3 Darwin-champion). Closest reference implementation for ruflo's own ADR-155 nightly self-learning security harness (PR #2417). Degrades gracefully when @metaharness/darwin is absent.
+description: "Run `@metaharness/darwin security bench` (Darwin Shield): evolve a security-detection harness on a 10-vuln/9-decoy corpus and grade TPR/FPR/patch/repro against four baselines. Use when benchmarking security-harness detection. Degrades gracefully without darwin."
 argument-hint: "[--population 2] [--cycles 1] [--seed N] [--alert-on-fail]"
 allowed-tools: Bash
+disable-model-invocation: true
 ---
 
 Surfaces the upstream `metaharness-darwin security bench` command. **This is

package/plugins/ruflo-metaharness/skills/harness-similarity/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: harness-similarity
-description: ADR-152 — weighted similarity between two harness fingerprints (genome + score JSON). Returns overall score in [0,1] plus per-component breakdown (cosine over 9 numerics, categorical agreement over 4 enums, jaccard over agent_topology). Unblocks ADR-151 §3.2 Recommender, §3.3 Drift Detection, §3.5 Plugin Compat. Pure-TS, no `@metaharness/*` dep — preserves ADR-150's four architectural constraints.
+description: "Weighted similarity (ADR-152) between two harness fingerprints (genome + score JSON): an overall score in [0,1] plus per-component breakdown. Use when comparing two harnesses or detecting drift between them. Pure TypeScript; needs no @metaharness package."
 argument-hint: "(--a a.json --b b.json | --a-key X --b-key Y) [--per-dimension] [--alert-below 0.5] [--format json|table]"
 allowed-tools: Bash
 ---