# Skill Auditor

> Audit skill candidates and classify each changed skill as unique or upgrade with severity findings. Use when creating/updating skills, preparing admission evidence, or producing audit JSON for skill-game scoring.

- Skill: `grtninja/skill-auditor` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add grtninja/skill-auditor`
- Raw SKILL.md: https://api.skillmd.com/api/skills/grtninja/skill-auditor/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: grtninja (https://skillmd.com/u/grtninja)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/grtninja/skill-auditor

---


# Skill Auditor

Use this skill to classify and gate new or updated skills.

## Workflow

1. Select changed skills (`--include-skill`) or scan the full candidate set.
2. Run the skill audit script to classify each skill as `unique` or `upgrade`.
3. Validate core quality checks:
   - frontmatter completeness (`name`, `description`),
   - `agents/openai.yaml` presence,
   - size/readability guardrails,
   - required arbiter evidence when enabled,
- meta-harness drift such as legacy `Documents\GitHub` roots or `:9000` treated as a model-authority surface.
4. Review findings by severity (`high`, `medium`, `low`) and fix failures.
5. Re-run until `high_count=0` for admission-ready output.

## Commands

Audit selected skills:

```bash
python3 scripts/skill_audit.py \
  --skills-root skill-candidates \
  --include-skill <skill-name> \
  --json-out /tmp/skill-audit.json
```

Audit with arbiter evidence required:

```bash
python3 scripts/skill_audit.py \
  --skills-root skill-candidates \
  --include-skill <skill-name> \
  --arbiter-report /tmp/skill-arbiter-evidence.json \
  --require-arbiter-evidence \
  --json-out /tmp/skill-audit.json
```

## Classification Rules

- `unique`: no strong near-peer overlap detected in current skill set.
- `upgrade`: near-peer overlap indicates refinement/extension of an existing lane.
- `high` findings block completion.
- `medium` findings should be addressed before admission when possible.
- Meta-harness-sensitive skills should also encode the canonical `G:\GitHub` root contract, direct LM Studio `:1234` model authority, explicit `:2337` hosted selection, non-model `:9000` support, and PC Control-first evidence rules where applicable.

## Scope Boundary

Use this skill only for skill-audit classification and findings generation.

Do not use this skill for runtime skill arbitration; use `$skill-arbiter-lockdown-admission`.

## References

- `references/audit-rubric.md`
- `scripts/skill_audit.py`

## Loopback

If findings remain unresolved:

1. Capture failing checks and evidence paths.
2. Route through `$skill-hub` for chain recalculation.
3. Resume only after an updated chain assigns deterministic fixes.