llm-evals

Agentic Plugin Marketplace

Production-ready agentic workflow building blocks: Plugins, agents, skills, and commands built for Claude Code and consumed natively by OpenAI Codex CLI, Cursor, OpenCode, Gemini CLI, and GitHub Copilot from a single Markdown source.

Author a capability once in Claude Code's plugin format, and ship it to every major agentic coding harness. A deterministic adapter layer translates the single canonical source into each harness's native packaging — so the instructions live in one place and never drift.

Status

Building in phases — see the roadmap for detail.

• ✅ Phase 0 — Foundations. Vision, architecture, plugin spec, taxonomy, distribution model.
• ✅ Phase 1 — Canonical authoring & Git marketplace. A 5-plugin seed catalog, a CI-gated validator, and the rosetta marketplace — verified by a live /plugin install in Claude Code.
• ✅ Phase 2 — Adapter layer & CLI. The rosetta CLI translates a canonical plugin into all five target harnesses with a per-component fidelity report: Cursor, Codex CLI, OpenCode, Gemini CLI, and GitHub Copilot — each golden-file tested.
• 🚧 Phase 3 — Catalog buildout. Populating the taxonomy with depth, starting by dogfooding — an authoring toolkit (plugin-scaffold, skill-writer, catalog-curator), then the Languages family and the AI & ML cluster. 14 plugins across 10+ categories so far (coverage).
• ⏳ Phases 4–5 — web registry/site, governance.

Documentation

Start with the docs index, or jump in:

• Vision — the problem, the bet, the principles.
• Architecture — how single-source → multi-harness works.
• Plugin spec — the canonical format you author in.
• Categories — the full catalog taxonomy.
• Distribution — Git marketplace, CLI installer, web registry.
• Roadmap — the phased plan.
• Contributing — author and submit a plugin.

See plugins/changelog/ for a fully-worked reference plugin.

License

MIT — see LICENSE.

Star History