skillcheck — does your skill actually trigger?

    1

    Check a skill before you install it, and find out which of your skills steals the requests meant for another.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    skillcheck — does your skill actually trigger?

    Example session with this skill installed

    My code-ref skill never fires in Claude Code anymore. Run a skillcheck scan and tell me why.

    • Read your context and instructions
    • Compiled the skillcheck

    Summary

    Ran python3 skillcheck.py scan (static checks, no model calls). code-ref isn't losing to another skill: it is switched off in your Claude settings, so it never reaches the skill list. Output below is from a real run, paths shortened.

    == claude: 17 skills
      !   EASY                         ~/.claude/skills/EASY
           [info] Agent Skills spec (agentskills.io) names are lowercase letters, digits and hyphens; other agents may reject this name
      OK  ai-radar                     ~/.claude/skills/ai-radar
      !   code-ref                     ~/.claude/skills/code-ref
           [info] settings.json skillOverrides is 'off': this skill is turned off
      OK  codex                        ~/.claude/skills/codex
      OK  deep-research                ~/.claude/skills/deep-research
      !   search                       ~/.claude/skills/search
           [info] settings.json skillOverrides is 'off': this skill is turned off
      OK  tracer                       ~/.claude/skills/tracer
      ...
    == codex: 8 skills
      OK  deep-research                ~/.agents/skills/deep-research
      ...
    
    SkillFindingWhat to do
    code-refturned off by skillOverrides in settings.jsonremove the override if you want it back (your call)
    searchturned off the same waysame
    EASYuppercase namefine in Claude Code; rename to easy if you share it with other agents

    Next steps

    • Turn code-ref back on, then check it actually wins its requests next to your other 16 skills: python3 skillcheck.py gen --skills code-ref and python3 skillcheck.py run --agent claude --suite suite.json --tag before --runs 3.
    • Before installing any new skill: python3 skillcheck.py preflight path/to/new-skill.
    turned off by in settings.json

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Most "my skill doesn't fire" problems are not a broken skill. Either another installed skill wins the request, or the description makes the model treat it as manual-only. Isolated evals (one skill in a clean environment) can't see either: in one measured setup, prompts that fired 21/21 in isolation missed 5 times with the user's other skills installed.

    What it does

    • scan: static checks, no model calls: frontmatter, name/folder mismatch, skills turned off, nested SKILL.md, listing budget.
    • preflight path/to/new-skill: before installing, see which of your installed skills it will compete with. --measure runs the same prompts with and without it in your real CLI.
    • run: sends prompts through your real claude -p / codex exec with everything installed and records which skill actually loaded ("taken by").
    • compare / html: before/after a description fix, as a one-file report.

    How it works One Python file, standard library only, MIT. Source and releases: https://github.com/Noem836/skillcheck (pinned install with sha256 in the release notes).

    Limits

    • Real measurement uses your own Claude Code / Codex subscription and takes about a minute per prompt run.
    • Claude runs get only the Skill tool; Codex runs without your config, MCP, web or plugins. Read any prompt suite you didn't write before running it.
    • Fixing descriptions is up to you. The paid Skill Trigger Doctor adds the measured rewrite rules and a guided before/after workflow.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.

    Requires Python 3 standard library. Testing takes ~1 min per prompt/run. Only supports Claude and Codex agent environments. Static analysis requires specific SKILL.md folder structures.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 20 days ago

    • Free with an Agensi account

    Accounts
    GitHub
    Software
    Python 3
    Claude Code CLI
    orCodex CLI

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access

    Allowed Hosts

    raw.githubusercontent.com

    Shell: runs python3 skillcheck.py and your claude/codex CLI. Reads your skill folders; writes suite.json, results and the HTML report in the current folder. Network: optional pinned download of skillcheck.py from GitHub, and a report upload only when you run the upload command yourself.

    Listed20 days ago

    What's inside

    Frequently Asked Questions

    This skill provides a Python-based testing framework that identifies why specific skills aren't triggering or are being "hijacked" by other tools. It replaces trial-and-error manual testing with data-driven reports on fire rates and description conflicts.

    Skillcheck is designed for developers using local agent environments like Claude Code and Codex. It requires a local Python 3 environment to run the scan and execution scripts.

    A single SKILL.md that tells your agent how to run skillcheck. The agent uses skillcheck.py from the plugin install, or downloads a pinned release from GitHub (https://github.com/Noem836/skillcheck) and you can check its sha256 against the release notes.

    Yes. Run the same prompt set before and after the edit (--tag before / --tag after), then `compare` shows fire rates and false fires side by side and `html` makes a one-file report. With a few runs per prompt this shows whether the change held up; it is not a statistical test.

    Each prompt run calls your real claude -p or codex exec and takes about a minute, so skills x prompts x runs adds up (5 skills x 6 prompts x 3 runs is roughly 90 minutes). Run it in the background. The scan and preflight static checks need no model calls and finish in seconds.

    New in Prompt & Skill Engineering

    Free