← ClaudeAtlas

confab-agentic-reliabilitylisted

Use when the user asks to audit this plugin repository's own skill, agent, or workflow definitions for agentic-loop reliability defects, wants to check for unbounded retries, missing escalation paths, unwired Find/Verify phases, or excessive tool grants, or asks 'is our own agent design safe'. Verification runs by default and only skips if the user explicitly says skip_verification. Findings fall into exactly four categories.
Anselmoo/werkstoff · ★ 1 · AI & Automation · score 67
Install: claude install-skill Anselmoo/werkstoff
Audit this repository's own `skills/`, `agents/`, `commands/`, and `workflows/` files for four specific agentic-loop reliability defect categories: `unbounded-retry`, `no-escalation-path`, `find-no-verify-wiring`, `excessive-tool-grant`. These four categories are fixed — never report a fifth category, and never rename one. ## Steps 1. Determine `repo_root` and the set of skill/agent/workflow files to scan (default: everything under `skills/`, `agents/`, `commands/`, `workflows/`). Determine `skip_verification`: **default `false`**, same rule as `confab-contract-drift` — only `true` on an explicit user request. 2. Dispatch the `agentic-reliability-auditor` agent in **Find mode** over those files. Ask for `{"findings": [...]}` (category must be one of the four above) plus, separately, any trivial-scope exceptions it noticed (a broad-looking tool grant on a genuinely trivial skill) as `{"exceptions": [{"evidence": "...", "reason": "..."}]}`. Write both to scratch files. 3. Unless `skip_verification` is `true`: dispatch the `agentic-reliability-auditor` agent again in **Verify mode**, independently re-checking each Find-phase finding against the actual file. Write `{"findings": [{"evidence": "...", "confirmed": true|false}, ...]}`. 4. Run: ``` python3 "${CLAUDE_PLUGIN_ROOT}/scripts/agentic_reliability.py" <repo_root> \ --find-json <path-from-step-2> \ [--verify-json <path-from-step-3>] \ [--skip-verification] \