erichare/skillroute

curl -fsSL https://raw.githubusercontent.com/erichare/skillroute/main/scripts/install.sh | bash
uv run skillroute index --root examples/skills
uv run skillroute harness install claude-code

Usage

• •Index your skill library first: skillroute index --root <dir> (catalog defaults to ~/.skillroute/catalog.db).
• •Route requests with skillroute route "..." to get ranked skills with confidence, reasons, and evidence.
• •Install one of the 15 harnesses (claude-code, codex, cursor, opencode, gemini-cli, etc.) so the MCP server is wired into your agent.
• •Inspect past routing decisions with skillroute traces list and run golden-route evals with skillroute eval run.

Key features

• ✓Full-bundle indexing parses complete SKILL.md packages rather than just frontmatter.
• ✓Explainable routes: every ranked skill carries confidence, reasons, and evidence; uncertain routes return clarification questions instead of guesses.
• ✓Skill Atlas interactive UI (FastAPI + React Flow) showing the whole library as a graph with live route preview.
• ✓MCP server (TypeScript stdio) exposing skillroute.route, skillroute.search, and skillroute.inspect_skill.
• ✓Hybrid routing across backends: default local token retrieval, SQLite FTS5/BM25, or Astra DB / LangChain adapters.
• ✓Harness packs are declarative manifests, so adding a new tool client is usually data-only (15 harnesses supported, incl. a DeepSeek dsh plugin bundle).

Best for

Teams and builders with large skill libraries who want ranked, explainable, evals-backed skill selection across their coding agents.

Caveats

• ⚠Python 3.11+ required.
• ⚠Routing quality depends on the completeness of the indexed catalog.
• ⚠Astra DB backend is for when you outgrow local SQLite — optional, not the default.