Agent Skills SKILL.md MIT
Skills for agents that have to get it right.
Procedures for the failures agents actually have: work that reads as finished and is not, tests that cannot fail, setup that reports success while broken, and commands to the physical world that were accepted but never happened. Each one ends in a verdict you can check: holds, broken or skipped.
Browse the skills Point an agent here
Works with Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Jules, and agents that drive hardware.
take-to-production on a notes app that looks finished
- holdsadd a notetyped, pressed Enter, note listed
- brokencome back and see itadded one note, reloaded: 1 before, 0 after
- brokena note that is HTML
<img src=x onerror="alert(document.domain)">ran script - holdsexport, ChromiumshowSaveFilePicker present
- brokenexport, Safari and Firefox engines
window.showSaveFilePickerabsent (emulated by deleting it): Export alerts "Something went wrong" - broken320 px phonedocument is 592 px wide at 320 px
- brokenvisible keyboard focusfocused input has outline-style none
- brokenno third-party requestslodash from cdn.jsdelivr.net, 0 uses in app.js
- brokenerror messages say what to do"Something went wrong"
- brokenempty stateno notes: the list renders nothing
- holdslight and darkcolor-scheme and prefers-color-scheme tokens
3 holds, 8 broken, 0 skipped of 11 items.
Point an agent here
Give it this line. Everything it needs is one fetch away, and nothing depends on which agent it is.
Read https://jules-prompts.wecanuseai.com/llms.txt and follow the skill that matches the task.Clients that support Agent Skills discovery can start from /.well-known/agent-skills/index.json, which lists every skill with a SHA-256 digest of its SKILL.md.
Read it yourself
Every skill is a page you can read, copy, download or install, and the same text an agent loads.
20 core skills, plus the older general-purpose ones, marked legacy.
Start here
Take a Project to Production Quality
To take a project in any state, software or physical, to production quality. It does its one job dependably and securely on every platform its people use, loses and leaks nothing, feels finished, carries nothing it does not need, and every area that decides this, from security and privacy to operations, is checked rather than assumed.
Act on the Physical World, and Prove It Happened
To make an agent's commands to hardware, devices, machines and real-world services safe to issue and provable afterwards, because a command that was accepted is not a valve that closed.
Install
One skill
mkdir -p .claude/skills/take-to-production
curl -fsSL https://jules-prompts.wecanuseai.com/.well-known/agent-skills/take-to-production/SKILL.md \
-o .claude/skills/take-to-production/SKILL.mdAll of them
git clone --depth 1 https://github.com/melbinjp/jules-prompts
cp -R jules-prompts/skills/* .claude/skills/As a Claude Code plugin
/plugin marketplace add melbinjp/jules-prompts
/plugin install jules-prompts@jules-promptsAs slash commands, over MCP
{ "mcpServers": { "jules-prompts": {
"command": "npx", "args": ["-y", "github:melbinjp/jules-prompts"] } } }Use .agents/skills/ for Codex and other agents that read AGENTS.md. Paste the standing rules into a project's AGENTS.md so they apply even when nobody picks a skill.
How you know a skill works
A skill nobody has seen fail is a claim. Each core skill has a fixture: a small project with planted defects and a list of them. Run an agent on it with the skill, and scripts/score_fixture.py says which defects its report named, which it missed and which it made up. The last line is always the denominator. The report at the top of this page is the one the flagship skill should produce on its fixture.
Questions
What is an Agent Skill?
A folder with a SKILL.md file in it. The file starts with a name and a one-line description of when to use it, and the procedure follows. An agent that supports skills reads the descriptions and loads the one that matches the task. Claude Code, Codex, GitHub Copilot, Cursor and Gemini CLI all read the format.
Which agents can use these?
Any agent that can read a file or fetch a URL. Install them as skills where the agent supports them, point it at llms.txt, add the MCP server to get them as slash commands, or paste the text. Nothing in them depends on one product's tool names.
Is this affiliated with Google or Jules?
No. The library started in 2025 as prompts for Jules, Google's coding agent, which is where the name comes from. The skills work with Jules and with every other agent.
What does it cost?
Nothing. Everything is MIT licensed, there is nothing to sign up for, and the site has no analytics, cookies or trackers.
Can an agent that controls hardware use them?
Yes. Act on the Physical World, and Prove It Happened is for any command that moves, heats, dispenses, spends or sends. It confirms the result through an independent measurement rather than the device's acknowledgement, and sends every failure to a safe state decided in advance.
Tools these skills pair with
A skill tells an agent what to do. These do part of the same work mechanically, so the agent spends its judgement on the cases that need it.
- docproof reads a repository's documentation and reports the claims its code contradicts: paths that no longer exist, commands that are not there, versions that moved. It is the automated half of Prove the docs, and it runs as a GitHub Action. MIT.
- rigout is an MCP server that lets an authorised agent use a real machine: commands, files, Docker, persistent terminals.
pip install rigout.
Writing
Notes from actually running this kind of tooling live at log.wecanuseai.com. Most recently, why a clean verdict over five per cent of a repository is not a clean repository.