Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Skillset Lint: Fix Skills That Don't Trigger

    1

    It sees **all of them**, and picks one by its description.

    Free

    Security scanned
    Skillset Lint: Fix Skills That Don't Trigger

    Skillset Lint: Fix Skills That Don't Trigger

    Example session with this skill installed

    Lint the skills in this repo before I publish them, and check them against the skills I already have installed.

    • Read your context and instructions
    • Compiled the skillset lint

    Summary

    I linted the 20 skills in this repository (the real result for anthropics/skills). 2 will not load on strict clients, 3 will load but rarely trigger. Nothing risky was found in the scripts.

    Will not load

    SkillRuleProblemFix
    claude-apiSL005description is 1,068 characters; the spec limit is 1,024Trim to ~600 characters; move the model list into the body
    templateSL004name: template-skill does not match its folder templateRename the folder or the name

    Will load but trigger badly

    SkillRuleProblem
    theme-factorySL006says what it is, never when to use it
    webapp-testingSL006same
    templateSL006same

    Suggested description for webapp-testing:

    Tests local web applications with Playwright: verifies frontend behavior, debugs UI issues, captures screenshots and reads browser logs. Use when the user asks to test, check or debug a web app running locally.

    Cost and hygiene

    • claude-api: SKILL.md body is ~26,000 tokens, 5× the recommended size. Split reference sections into shared/*.md.
    • skill-creator: ~8,200 tokens.
    • The skill list adds ~2,600 tokens to every conversation; the longest descriptions are claude-api, academy-guide and discernment-nudge.

    Next steps

    Shall I (1) shorten the claude-api description, (2) rename the template folder, and (3) rewrite the three descriptions above? Then I will re-run the linter and show the before/after counts.

    SKILL.md

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

    About this skill

    Why your skill doesn't trigger

    Your agent never sees one skill. It sees all of them, and picks one by its description. So the bugs that hurt in practice live between skills:

    • Two skills read alike, and the agent loads the wrong one, or neither.
    • Two skills share a name in different folders, and which one wins depends on load order.
    • Every description is paid for in every conversation. 30 installed skills can cost 4,000+ tokens before you type a word.

    Skillset Lint checks each skill against the Agent Skills spec and best practices, then checks the set.

    What it checks

    Each skill (24 rules)

    • Frontmatter: valid YAML (with the exact line), name rules and folder match, description 1-1024 characters, compatibility, metadata, allowed-tools
    • Descriptions that say what but not when to use it, first-person descriptions, < > that some clients reject, reserved words
    • File references that do not exist, backslash paths, reference chains deeper than one level, scripts nothing mentions
    • SKILL.md over 500 lines or ~5,000 tokens, placeholder text
    • Security: curl | bash, hard-coded secrets (masked), hidden Unicode, credential files sent over the network, unguarded rm -rf

    The set (4 rules)

    • Duplicate names
    • Overlapping descriptions, with the shared words that cause it
    • How many tokens the skill list costs per conversation, and the longest descriptions

    How the agent uses it

    1. Runs the bundled zero-dependency Node script on a folder, a repo, or everything installed (--installed: Claude Code, plugins, Codex, OpenCode, project skill folders).
    2. Groups the results: will not load, will load but trigger badly, risky, cost and hygiene.
    3. Fixes what you approve: renames, rewritten descriptions (both sides of an overlap), fixed paths, split-out reference files. Then re-runs to show the before/after.

    Found in real repositories

    On anthropics/skills it found a description of 1,068 characters (over the 1,024 limit), a name that does not match its folder, and descriptions that never say when to use them. On obra/superpowers it flagged two skills whose descriptions overlap.

    Never executes the scripts it checks. JSON and SARIF output, plus a GitHub Action for CI.

    Open source (MIT): https://github.com/MBotirjonDev/skillset-lint

    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.

    Static analysis only; does not execute scripts. Requires Node.js to run the linting utility. Focuses on SKILL.md and metadata, not full logic debugging.

    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 9 days ago

    • Free with an Agensi account

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Listed9 days ago
    Updated2 days ago

    What's inside

    Frequently Asked Questions

    It checks every skill against the Agent Skills spec and then checks the whole set: duplicate names, descriptions so alike the agent picks the wrong skill, descriptions that never say when to use the skill, and file paths that do not exist. Your agent gets a ranked list and fixes what you approve, instead of you guessing why a skill does not trigger.

    Yes. Every installed skill's name and description is sent at the start of every conversation. The linter reports what your skill list costs in tokens, flags the longest descriptions, and flags SKILL.md bodies over ~5,000 tokens so you can move detail into reference files.

    Any agent that uses the open SKILL.md standard. The --installed option looks in the folders used by Claude Code (including plugins), Codex, OpenCode and the shared .agents folder, plus your project's skill folders. It needs Node.js 18+.

    No, this is a static analysis tool that scans code and markdown for syntax and security patterns; it never executes your scripts, ensuring your environment remains safe during the audit.

    The skill folder: SKILL.md (the workflow your agent follows) and two zero-dependency Node.js scripts, lint.cjs (28 rules: SL001-SL024 for each skill, SS001-SS004 for the set) and yaml.cjs (its frontmatter parser). The same code, plus a GitHub Action, is open source (MIT) on GitHub.

    It is built for SKILL.md files. Other prompt libraries without SKILL.md frontmatter are not checked.

    New in Prompt & Skill Engineering

    Free