Open-source Claude Code skills for building software that AI agents can actually use — by the team behind 2027.dev, where we run real agents through 100+ dev tools and score their experience.
Each skill is a self-contained folder under skills/. Drop one into your project and your agent will use it.
| Skill | What it does |
|---|---|
designing-agent-error-messages |
Audit and rewrite your error messages, exceptions, and API/SDK/CLI failures so an agent can self-correct from the message alone. Grounded in patterns from 3,500+ agent runs across 136 dev tools. |
agent-auth |
Log a real human into a dev tool's CLI by auto-opening the OAuth browser popup and letting the login complete itself — no token copy-paste, no browser automation. Sanity + Netlify recipes plus a verified login reference for 19 dev-tool CLIs. |
More coming — discoverability, agent-friendly docs, SDK design.
Copy whichever skill(s) you want from the table above — swap the skill name as needed (agent-auth, designing-agent-error-messages, …).
Project scope (this repo only):
git clone https://github.com/team2027/2027-skills.git /tmp/2027-skills
mkdir -p .claude/skills
cp -r /tmp/2027-skills/skills/agent-auth .claude/skills/
cp -r /tmp/2027-skills/skills/designing-agent-error-messages .claude/skills/Personal scope (all your projects):
git clone https://github.com/team2027/2027-skills.git /tmp/2027-skills
mkdir -p ~/.claude/skills
cp -r /tmp/2027-skills/skills/agent-auth ~/.claude/skills/
cp -r /tmp/2027-skills/skills/designing-agent-error-messages ~/.claude/skills/Then in Claude Code each skill loads automatically when it's relevant — or invoke one directly: /agent-auth, /designing-agent-error-messages.
These skills fix what you can grep for. To see where a real agent actually gets stranded in your onboarding — and how you rank against 135 other dev tools — check out the agent arena and submit a tool eval request.
MIT © 2027.dev