audit-skill · git:20260831.b100768 · 2026-08-31 · sha256 5ce666998e21e7c6
audit-skill git:20260831.b100768B
Immutable. This exact content is served forever at /api/v1/blob/5ce666998e21e7c6.
--- name: audit-skill plugin: agent-scaffolders description: Audits and aligns individual agent skills and sub-agents against ecosystem evolution standards, verifying line budgets, evals schemas, and hub-and-spoke isolation. allowed-tools: Bash, Read, Write, Glob, Grep --- # audit-skill: Skill Alignment & Evolution Auditor Audits an individual skill against the 6 ecosystem evolution invariants and provides automated refactoring recommendations to achieve full compliance. --- ## The 6 Alignment Invariants 1. **Layer 1 Procedural Core:** `SKILL.md` target $\le 100$ lines. Offload domain depth to `references/` via Progressive Disclosure. 2. **Boolean Evals Schema:** `evals/evals.json` must be a root JSON list with `should_trigger: true/false`. 3. **Hub-and-Spoke Script Isolation (ADR-002/003):** All scripts reside in plugin root `scripts/` and are symlinked into `skills/<skill>/scripts/`. Zero un-symlinked real files. 4. **Frontmatter Standards:** `name` matches directory slug exactly; `description` is third-person active verb and $\le 1024$ chars. 5. **Contract & Fallback Links:** Must provide `acceptance-criteria.md` and `fallback-tree.md` in `references/`. 6. **Spoke Hygiene:** Zero session logs, raw wiki traces, or `.agent/` state in skill folders. --- ## Quick Execution Run the bundled alignment script directly against any skill directory: ```bash # Audit a single skill python3 plugins/agent-scaffolders/scripts/audit_skill.py plugins/<plugin>/skills/<skill-name> # Auto-repair fixable schema deviations (e.g. migrate evals.json) python3 plugins/agent-scaffolders/scripts/audit_skill.py plugins/<plugin>/skills/<skill-name> --fix # JSON output for automated pipelines python3 plugins/agent-scaffolders/scripts/audit_skill.py plugins/<plugin>/skills/<skill-name> --json ``` --- ## Alignment & Refactoring Procedure ### Phase 1: Diagnostic Assessment 1. Run `audit_skill.py <path>`. Note all errors and warnings. 2. If `SKILL.md` exceeds 100 lines: - Identify detailed background, multi-page tables, or complex workflows. - Extract them into `references/<topic>.md`. - Replace in `SKILL.md` with concise procedural steps and markdown links. ### Phase 2: Hub-and-Spoke Remediation (ADR-002/003) If real scripts are found inside `skills/<skill>/scripts/`: 1. Move the real script to `plugins/<plugin>/scripts/<script-name>.py`. 2. Register the symlink in `symlinks.json`. 3. Run `python3 .agents/skills/symlink-manager/scripts/symlink_manager.py restore`. 4. Run `symlink_manager.py diagnose` to confirm zero imposters. ### Phase 3: Contract Verification 1. Ensure `evals/evals.json` has at least 3 positive and 3 negative test cases using `should_trigger`. 2. Ensure `references/acceptance-criteria.md` and `references/fallback-tree.md` are linked. 3. Re-run `audit_skill.py` to confirm 100% PASS.