skills.rest
reddb-io
reddb-ioOFFICIAL
Explore all skills by this author →
GitHub

audit-skills

Scores SKILL.md files against house style with mechanical checks and an LLM judge.

20|2|Updated May 16, 2026
Search Tags:#prompt engineering#style guide#telemetry#documentation review#skill audit#quality scoring#LLM judge
One-click install
npx skills add https://github.com/reddb-io/red-skills --skill audit-skills-reddb-io
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: audit-skills
Source: https://github.com/reddb-io/red-skills/tree/main/plugins/dev/skills/engineering/audit-skills
Command: npx skills add https://github.com/reddb-io/red-skills --skill audit-skills-reddb-io

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Skill authors lack objective feedback on whether their SKILL.md files follow house style and writing-for-agents techniques, so weak skills ship unnoticed until they fail in practice. ## Core Features & Use Cases - Composite scoring: Combines mechanical checks (frontmatter, description budget, structure rules) with an LLM judge into a 0-100 worst-first scorecard. - Telemetry overlay: When the memory store is reachable, behavioral signals like abandoned or frequently-failing skills float measurable writing failures to the top. - Read-only by design: The audit never mutates skills, has no fix mode, and guards against prompt injection inside skill bodies. - Use Case: Run the audit before a release to identify which shipped skills have the weakest trigger descriptions or bloated bodies, then hand the worst offenders to authors for revision. ## Quick Start Ask the assistant to run /audit-skills and show the worst-first scorecard for all shipped skills.

Frequently Asked Questions about audit-skills

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I audit the quality of my SKILL.md files?▼

Invoke /audit-skills to run the read-only auditor, which scores every shipped SKILL.md against house style rules and prints a worst-first scorecard. Use the mechanical_only option to skip the LLM judge for a faster objective-only pass.

What does the skill audit score measure?▼

The composite 0-100 score combines 60% semantic judging of writing-for-agents techniques with 40% mechanical checks like frontmatter presence, description length budgets, and required section structure. Mechanical-only mode uses just the objective checks.

Can the skill auditor modify or fix my skills automatically?▼

No, the auditor is strictly read-only by construction. It has no fix flag, no git or gh access, and no backlog integration; it only reports scores and concrete suggestions for authors to apply manually.

Does the audit work without the memory store or LLM provider?▼

Yes, it degrades cleanly. Without the memory store, ranking falls back to composite score alone, and with mechanical_only enabled or a judge failure, scoring uses objective checks only with no provider call.

How does the auditor handle prompt injection inside skill bodies?▼

The judge prompt frames each skill body as untrusted data to score, never commands to obey. A skill attempting to manipulate its own score is treated as a quality defect rather than an instruction.

Related Skills

Explore More Software Engineering Skills→
AvaTar-ArTsAvaTar-ArTs

skill-development

Guides creation of Codex plugin skills with progressive disclosure and trigger phrases.

Community
Intermediate
mccleod1290mccleod1290
1

skill-optimizer

Audits and rewrites skill frontmatter to improve router invocation rates.

Community
Intermediate
ax-llmax-llm
2.9k

ax-java-agent-memory-skills

Tracks Ax Java agent memory recall and skill usage via host callbacks.

Official
Intermediate
ribatsheporibatshepo

docs-review

Detect style-guide violations in Markdown files and PR diffs with inline feedback.

Community
Intermediate