A search engine and testing lab for Agent Skills: find a candidate Skill, try it in an isolated sandbox with your own prompts and data, and download a portable package.
| Path | What lives there |
|---|---|
apps/web/ |
React + TypeScript SPA (Vite) |
services/platform/ |
Go control plane: cmd/api HTTP server, cmd/worker queue consumer |
services/llm/ |
Python FastAPI service for LLM workloads (uv) |
contracts/openapi/ |
OpenAPI specs — the single source of truth for cross-language interfaces |
db/ |
PostgreSQL migrations, sqlc queries and config |
infra/compose/ |
Local and single-host deployment stack |
plans/, adr/ |
Product plan and architecture decision records (Traditional Chinese) |
spikes/ |
M0 exploration code; never imported by product code, never built in CI |
Prerequisites: Go, Node.js, uv, Docker, and Task.
task dev # start Postgres
task test # run every test suite
task lint # lint every serviceEach service also works with its own native toolchain — go test ./...,
uv run pytest, npm test — from its own directory.
Read AGENTS.md for the implementation rules, and ADR-019 for why the repository is laid out this way and what CI enforces.