Thanks for your interest. Strumentario is in the earliest public stage; the highest-leverage help right now is feedback on the vision, the initial CLI/server surface, and the eval strategy.
- Open an issue with a concrete use case, friction point, or design question.
- Try the Phase 1 server once it lands and report what broke in your agent host.
- Small, focused PRs once the foundation is tagged:
- Additional tools or scorers
- Improvements to the React MCP widget or AI SDK path
- Docs, architecture notes, or failure postmortems
- Scaffold recipes for common schemas
- Prefer depth on MCP + evals + one surface over broad feature lists.
- Keep secrets out of the tree (
.env.exampleonly). - Match the existing TypeScript and package structure.
- Every meaningful tool or scaffold change should ideally be accompanied by an eval case or a note in
docs/. - AGENTS.md / Cursor rules / skills in this repo describe how agents working on Strumentario itself should behave; keep them accurate.
# Once packages exist
npm install
npm run typecheck
npm test
# Eval suite (requires Braintrust keys)
npm run evalSee ROADMAP.md for the current phase and definition of done.
Be respectful. Assume good intent. Focus on the work.