Contributions are welcome. The quality bar is the moat — production-ready, no stubs, no dead placeholders.
- Imperative voice. "Determine the target", not "You should determine the target."
- No emoji. Anywhere — including section headers. Use plain Markdown headings.
- No purple in any design output, token set, or UI.
- Progressive disclosure. Keep the entry file lean; push long templates, tables, and examples into
references/. Link to them. - No placeholders that don't work. Every example must run. Use
.env.examplefor keys — never commit real secrets. - Universal where possible. Note cross-tool portability (Agent SDK, MCP, other AI tools), even when Claude Code is the primary runtime.
| Type | Must include |
|---|---|
| Plugin | .claude-plugin/plugin.json, a README (what / when / install), and its skills or commands |
| Agent | A definition with a clear role, tool list, web-search wiring where relevant, and .env.example |
| Orchestrator | A lead agent + a named team of 5 sub-agents, plus documented human-in-the-loop and out-of-the-loop modes |
| Design style | A token set (W3C design-tokens JSON), a theme preview, and an export (Tailwind/CSS) |
- Fork the repo.
- Add your artifact in the right top-level directory, in its own folder.
- Include the README and the per-type requirements above.
- Open a pull request against
main.
By contributing you agree your work is released under the MIT License.