claude-skills-test / skill-audit-plugin
CLAUDE.md1 starsChanged 10 months agoMITView on GitHub
Portable skill activation experiment plugin for Claude Code. Audits and tests whether Claude correctly invokes project skills when given relevant queries. This plugin provides two skills: - /skill-audit:setup — Initialize the audit environment (check cclogviewer, discover project skills) - /skill-audit:audit — Run the full 4-phase activation experiment Skills are scored 0–3 on their description field: - +1 for activation directive ("ALWAYS invoke", "MUST use") - +1 for negative constraint ("Do not ... directly") - +1 for keyword triggers (5+ entries…
- Installs packages
In this repository3 other files it gives its agents.
- dockerfile-generatorSkill
- git-workflowSkill
- svelte5-runesSkill
The file
We split this CLAUDE.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 24
Always use env -u CLAUDECODE when running claude -p from within Claude Code to bypass nested session check
- Very specific: line 25
Use --output-format stream-json (not --output-format json) — the latter sends output to stderr
- Distinctive: line 27
Wait 2 seconds between trials to avoid rate limiting
≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
skill-audit Plugin
Portable skill activation experiment plugin for Claude Code. Audits and tests whether Claude correctly invokes project skills when given relevant queries.
Overview
This plugin provides two skills:
/skill-audit:setup— Initialize the audit environment (check cclogviewer, discover project skills)/skill-audit:audit— Run the full 4-phase activation experiment
Dependencies
- cclogviewer MCP — Required for session log analysis. Install with
go install github.com/anthropics/cclogviewer@latest
How It Works
- Phase 1 scans
.claude/skills/*/SKILL.mdand scores directive compliance - Phase 2 mines real queries from session logs via cclogviewer and generates contextual queries
- Phase 3 runs
claude -ptrials and verifies activation via cclogviewer - Phase 4 generates an audit report with activation rates, tool sequences, and recommendations
Key Conventions
- Always use
env -u CLAUDECODEwhen runningclaude -pfrom within Claude Code to bypass nested session check - Use
--output-format stream-json(not--output-format json) — the latter sends output to stderr - Parse stream-json line by line; the session_id is in the final
resulttype message - Wait 2 seconds between trials to avoid rate limiting
- Results are saved to
skill-audit-results/<timestamp>/
Directive Scoring
Skills are scored 0–3 on their description field:
- +1 for activation directive ("ALWAYS invoke", "MUST use")
- +1 for negative constraint ("Do not ... directly")
- +1 for keyword triggers (5+ entries in
keywords:field)
Agents
activation-tester.md— Sonnet-based agent that runs batches ofclaude -ptrials and returns session IDs
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

