Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Agent Skill Authoring Pack

    1

    Your skill silently never loads, or loads and never fires. Eight skills, a linter and a trigger tester fix SKILL.md for Claude Code, Codex, Gemini, Cursor.

    $24

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Agent Skill Authoring Pack

    Agent Skill Authoring Pack

    Example session with this skill installed

    My sql-migration-review skill never loads. Claude just answers migration questions itself. Here's the folder.

    • Read your context and instructions
    • Compiled the agent authoring

    The agent runs the pack's linter first

    $ node scripts/lint-skills.mjs skills
    error  sql-migration-review  description: plain value contains ": ", which YAML reads as a nested mapping; wrap the value in double quotes
    error  sql-migration-review  SKILL.md: broken link references/lock-levels.md
    warn   sql-migration-review  references/locks.md is never mentioned in SKILL.md; the agent will not know it exists
    warn   sql-migration-review  scripts/check-locks.mjs imports "pg", which is not a Node built-in; the user will not have it installed
    warn   sql-migration-review  SKILL.md hardcodes an agent skills directory; use paths relative to the skill folder so it works in every agent
    warn   sql-migration-review  version is not an Agent Skills field; loaders ignore it or reject the skill
    
    1 skill(s), 2 error(s), 4 warning(s)
    

    description: Reviews Postgres migrations. Use when: the user shares a migration is invalid YAML, so the skill never loaded in any agent. Quoting it makes the skill load, but a trigger test shows Claude still answers on its own:

    FAIL  0/2  expect sql-migration-review  fired none x2  "Is this safe to run on our 200M row orders table in prod? ALTER TAB..."
    

    Following skill-descriptions, the agent rewrites the frontmatter with the SQL keywords users paste and the situation that should route to the skill:

    ---
    name: sql-migration-review
    description: "Reviews Postgres schema changes for lock levels, table rewrites, NOT NULL and DEFAULT backfills, and index builds that block writes. Use whenever the user pastes ALTER TABLE, CREATE INDEX, ADD COLUMN or DROP COLUMN SQL, shares a migration file, or asks if a schema change is safe to run on a large or production table, even if the answer seems obvious."
    metadata:
      version: "1.0.0"
    ---
    

    It fixes the body (script path relative to the skill folder, link to references/locks.md, node:fs instead of pg) and re-runs both checks:

    $ node scripts/lint-skills.mjs skills
    1 skill(s), 0 error(s), 0 warning(s)
    
    $ node scripts/trigger-test.mjs cases.json --plugin-dir ./plug --runs 2 --model sonnet
    PASS  2/2  expect changelog-entry  fired changelog-entry x2  "I just fixed the timezone bug in the invoice export. Write the CHAN..."
    PASS  2/2  expect sql-migration-review  fired sql-migration-review x2  "Is this safe to run on our 200M row orders table in prod? ALTER TAB..."
    PASS  2/2  expect (none)  fired none x2  "What is the capital of Portugal?"
    
    3/3 cases passed (threshold 0.66, 2 run(s) each)
    

    Measured on Claude Code 2.1.296 with Sonnet: 0 of 2 before the rewrite, 2 of 2 after, and the unrelated prompt stayed quiet.

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

    About this skill

    Most skills fail before the agent reads a word of them. A colon in the description makes the YAML invalid and every agent drops the skill without a message. A description that reads well never fires, because the model answers migration or changelog questions on its own. Two skills share half their keywords and take turns stealing each other's requests. A Claude-only key gets the upload rejected by claude.ai. A fix ships, but the plugin version was not bumped, so nobody who installed it gets the fix. The marketplace scanner rejects the pack over one example line.

    This pack gives the agent the rules for each of those, checked on 2026-10-13 against the Agent Skills spec, the Claude Code, Codex, Gemini CLI, Cursor, Copilot and VS Code docs, and the Codex and Gemini CLI source:

    • skill-authoring: routes to the right skill, version table, release pre-flight checklist.
    • skill-loaders: where each agent looks, precedence, which frontmatter keys are portable, Claude Code extensions, Codex agents/openai.yaml and its 2% listing budget, Gemini CLI and Antigravity, diagnosing a skill that does not load.
    • skill-descriptions: what plus when, trigger words from real users, front-loading for truncation, negative triggers, collisions, measured before/after rewrites.
    • skill-content: lean SKILL.md, reference load conditions, compaction limits, instructions agents follow.
    • skill-scripts: zero-dependency scripts, deterministic output, exit codes, the Windows .cmd spawn trap, a tested template.
    • skill-evals: trigger test sets, headless claude -p facts (stream-json needs --verbose, --bare lists no skills), claude plugin eval suites with baselines and CI.
    • skill-publishing: semver and changelogs, plugin.json and marketplace.json that pass claude plugin validate, the Anthropic directory checklist, skills.sh, OpenAI and Cursor directories, the Agensi one-root-SKILL.md zip and what its scanner rejects, licensing free versus paid.
    • skill-listing: titles, pitches and demos that convert, with pricing data from the Agensi catalogue.

    Two Node scripts, no dependencies: lint-skills.mjs (spec rules, YAML traps, collisions against your installed skills, token cost, broken links, unmentioned files, npm imports) and trigger-test.mjs (runs prompts through claude -p and reports which skill fired). The linter's YAML handling matched a full YAML parser on 477 real installed skills.

    Works with Claude Code (skills folder or plugin), Codex, Gemini CLI, Cursor and Copilot. Plain Agent Skills format.

    Scripts need Node.js 20+. trigger-test.mjs drives Claude Code only; for Codex, Gemini CLI and Cursor the pack documents the headless commands but does not automate them. Agent versions move fast: facts were checked on 2026-10-13 against Claude Code 2.1.296, Codex CLI 0.162.1 and Gemini CLI 0.63.0.

    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 any payment.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Requires

    Python
    uv (Python tool runner)
    Node.js
    expo-doctor
    expo

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access

    File Scopes

    **/SKILL.md
    skills/**

    The linter reads the skill folders you point it at. trigger-test.mjs runs the Claude Code CLI (claude -p) on your test prompts. The agent edits SKILL.md and reference files in your own skills. No network calls from the scripts and no accounts needed.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    It will write a SKILL.md that reads well. Whether it loads and fires is a different question: an unquoted colon in the description silently drops the skill, keys differ per agent, and descriptions that do not name the words users type never trigger. The pack gives the rules plus a linter and a trigger tester that show the result instead of guessing.

    Claude Code (skills folder or plugin), Codex, Gemini CLI, Cursor and GitHub Copilot. The skill-loaders guide covers where each one looks, precedence and which frontmatter keys are portable.

    One entry skill with seven guides (loaders, descriptions, content, scripts, evals, publishing, listing) and two zero-dependency Node scripts: lint-skills.mjs and trigger-test.mjs.

    Yes. The linter takes a folder of skills, checks each one and flags description collisions between them and against the skills you already have installed.

    No. The linter only reads files. trigger-test.mjs runs your local Claude Code CLI, which uses your own Claude account as normal.

    New in Prompt & Skill Engineering