ClaudeMap

SkillRanker

A Rust CLI that ranks an agent's installed skills for the current step using TypeSafe's Jev — harness-aware, inspectable, and advisory by design.

SDKs & Toolingskill-selectionrankingclirusthooks
Visit projectDicklesworthstone/skillranker

SkillRanker (sr) combines the recent conversation, the current request, workspace signals and the selected harness's visible skill inventory, then asks Jev in two passes: a broad compare, then a richer rerank of a shortlist — both with a real "none of these" option. Explicit skill requests resolve locally; libraries beyond 254 eligible skills are narrowed first by Quill lexical prefiltering (FrankenSearch). The result is advisory — the agent still decides — and everything is inspectable: --why-not traces exclusions with thresholds, and opt-in case capture replays offline without another Jev call. Integrates as a Claude Code prompt hook (sr hook claude) or an inline TUI.

What it does

  • Two-pass Jev ranking: a broad compare, then a shortlist rerank with richer excerpts — a real 'none of these' option guards against forcing a fit
  • Harness-aware: reads the selected harness's visible skill inventory and resolves explicit skill requests locally before ranking
  • Inspectable: --why-not exclusion traces with thresholds, opt-in case capture, offline replay with no extra Jev calls
  • Large libraries stay workable: Quill lexical prefiltering admits up to 254 candidates per pass

When to use it

Agents carrying a large skill library, where loading a plausible-but-wrong skill burns context and derails the step — the ranking stays advisory and never overrides explicit requests.

Getting started

git clone https://github.com/Dicklesworthstone/skillranker.git
cd skillranker
cargo install --locked --path . --bin sr

sr demo --case useful     # offline fixture, no network needed
sr rank --allow-network   # live ranking (needs TYPESAFE_API_KEY)
sr hook claude            # Claude Code prompt-hook integration

Things to know

Requires a TypeSafe API key — session context and skill excerpts are sent to Jev for evaluation. The license is MIT plus an OpenAI/Anthropic rider (GitHub reports NOASSERTION), not plain MIT. Built from source with a pinned Rust 2024 toolchain; no packaged binary. A failed hook recommendation falls back quietly and never blocks the agent. Advice only: the agent and the user decide what to load.

Repository

Dicklesworthstone/skillranker

Charted

Related on the map

  • Harness Bridge

    SDK / Tool

    A local launcher that points Claude Code and seven other coding harnesses at any OpenAI-, Anthropic- or Responses-compatible endpoint — without editing a single config file.

    • launcher
    • local-models
    • cli
    • menu-bar
    View

  • Claude for Siri

    SDK / Tool

    Answers macOS Spotlight and Siri prompts with your signed-in Claude Code account — an experimental macOS 27 model-delegation provider.

    • macos
    • siri
    • spotlight
    • model-delegation
    View

  • AI Coding Assistant Data Extractor

    SDK / Tool

    Extract your own local chat history from ten AI coding assistants into one normalized JSONL — for fine-tuning, analytics, or backing up years of conversations. Standard library only.

    • cli
    • data-export
    • jsonl
    • backup
    View