← ClaudeAtlas

reposkein-setuplisted

Installs and verifies RepoSkein (the deterministic code-graph MCP server) in the current repository. Use when an agent wants RepoSkein's structural navigation but the tools are missing, error with "REPOSKEIN_REPO_PATH", or the repo has no .reposkein/ directory — i.e. set RepoSkein up and confirm it runs.
reposkein/reposkein · ★ 14 · AI & Automation · score 73
Install: claude install-skill reposkein/reposkein
# RepoSkein Setup & Health Check Goal: get RepoSkein actually serving in THIS repo, then confirm it before relying on it. RepoSkein has three moving parts — the **indexer binary**, the **`.reposkein/` graph**, and the **MCP server** registered in your agent. This skill walks all three and verifies each. ## 1. Check current state first Run the health check (it needs nothing but Node + the package): ```bash npx -y @reposkein/mcp doctor . ``` Read the report. `✓ indexer binary` and `✓ repo indexed` are the prerequisites. If both pass, skip to step 4 (verify the server). Otherwise continue. ## 2. Install + index (if doctor reported ✗) ```bash npx -y @reposkein/mcp init . # fetches the indexer, installs git hooks + the navigation skill reposkein-indexer index . # builds .reposkein/ (or ask the agent to call init_cpg_skeleton) git add .reposkein/meta.json .reposkein/config.toml .reposkein/.gitignore .reposkein/.gitattributes git commit -m "chore: add RepoSkein config" ``` `nodes.jsonl` and `edges.jsonl` stay out of git: they are derived from the working tree, so every clone rebuilds them in seconds (the MCP server does it on startup when they are missing), and committing them would conflict on every branch that touches code. `.reposkein/summaries/` appears once an agent writes its first summary — commit that directory, since nothing can regenerate it. Summaries are sharded into `<xx>.jsonl` by a hash of the node id, so two branches summarising different code