SkillRanker sits around an agent's session and asks Jev to compare the skills available to it against the live context, returning which ones look like a fit for the next step, with local safeguards and inspectable feedback around the call.
It ships as a self-contained CLI with Claude Code hooks, structured JSON output, an abstention path when Jev is not confident, and local feedback, and can be tried offline first against a fixture case before connecting a real session.
It needs its own TypeSafe API key to run the ranking, since Jev is what actually makes the call.
What you get
- Full Rust source code under the MIT license with an added OpenAI and Anthropic rider
- The sr CLI, its Claude Code hooks, and an offline demo fixture
Jev
Rust
Node.js
JavaScript
Codex
TypeScript
Bun
Python
OpenRouter