skillranker
Rust CLI powered by Jev from TypeSafe.
Overview
skillranker (github.com/Dicklesworthstone/skillranker) is Dicklesworthstone's Rust CLI (sr) built around TypeSafe Jev. It reads recent conversation, the current request, workspace signals, and the harness skill inventory, then asks Jev to compare candidates including a real none of these option. Large rosters can pre-filter with Quill from FrankenSearch before Jev runs; explicit skill requests resolve locally first. Hooks, JSON output, replay, and local feedback ship for production agents. A TypeSafe API key is mandatory: there is no bundled local model.
Problem: Large skill libraries look interchangeable in prose, so agents load the wrong procedure and burn context three turns later.
Built for: Claude Code and compatible harness users who want sr to rank skills from live session context with abstention when nothing fits.
First indexed on Jev Directory: 2026-09-18
Creator and team
- Name
- Dicklesworthstone
- Handle
- @Dicklesworthstone
How Jev is used
- Role in the product flow
- Two-stage skill fit ranking with abstention over the visible inventory
- Primitives
- ChoiceScore
- State in
- Session transcript slice, active user request, workspace metadata, and eligible skill records (full roster or Quill-narrowed shortlist per README).
- Decision out
- Ranked skill recommendations with none of these when Jev rejects the field; advisory output for the agent, not a forced load.
- Establish session context and eligible skills (local resolution for explicit picks)
- Optionally narrow very large libraries with Quill before Jev
- Jev broad comparison then richer excerpt evaluation on a shortlist
- Emit ranked skills, abstention, and inspectable JSON or hook payload
SkillRanker treats Jev as the evaluation engine, not the transport layer. Local code gathers candidates and safeguards; Jev answers which skill fits the next step and whether the set should be rejected entirely. Choice semantics cover picking among named skills plus none of these, which is how abstention stays typed instead of a hand-wavy maybe later. Score-style comparisons show up in the two-pass flow (broad compare, then deeper excerpt read). Operators run sr rank --allow-network with hooks wired in Claude Code when they want automatic nudges. Budget TypeSafe spend on busy sessions; README documents replay and local calibration tooling for when rankings drift.
Sourced performance claims
- Public GitHub repo Dicklesworthstone/skillranker had about one hundred thirteen stars when this listing was drafted.Source: GitHub star count September 2026
Features and stack
Features
- Rust sr CLI with JSON, hooks, and TUI surfaces
- Jev comparisons include none of these abstention
- Quill pre-filter for libraries above two hundred fifty four skills
- Requires TypeSafe API key (no substitute provider)
Stack
- Rust
- CLI
- Claude Code hooks
- TypeSafe System One
Pricing: Open source CLI; every rank call uses your TypeSafe account. Quill narrowing is local; Jev evaluations bill per README.
Links
FAQ
- Can SkillRanker run without TypeSafe?
- No. README states ranking requires your own TypeSafe account and API key; local retrieval only prepares candidates.
- What is none of these?
- Jev can reject the whole candidate set when nothing fits, so the agent is not forced to load a plausible-but-wrong skill.
- How do I try it offline first?
- Run sr demo --case useful for fixtures, then sr rank --allow-network when you are ready to connect a live session.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- System One model
The model family behind Jev: parallel typed questions, one forward pass, probabilities you can threshold.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.