SkillRanker
A Rust CLI ranks installed agent skills against live session context with Jev, then applies abstention and local feedback around the typed result.
Added to Jevfast
How it uses Jev
SkillRanker combines the current request, recent conversation and tool evidence, workspace signals, and visible skill inventory. Jev compares candidates broadly, then evaluates richer excerpts for a shortlist; a none-of-these option supports abstention. The result advises the agent, which retains control over what to load.
What Jev decides
Is the session stuck?Example answers · not a recorded Jev response · Source ↗
Question 1 · stuck
YOUR APP
INSTRUCTION
Is there evidence of repeated failure on the current task in this session?
STATE
Rendered session context and the admitted skill roster, including available skills for the next step.
JEV · NOUL
YesNo
App workflow
- Resolve explicit requests
Explicitly requested skills are resolved locally before probabilistic ranking.
- Prepare candidates
The tool discovers visible skills and session context; Quill narrows very large inventories locally.
- Rank and rerank
Jev compares the candidate roster and then evaluates richer excerpts from a shortlist, including abstention.
- Explain the recommendation
The CLI emits an advisory result with exclusions and evidence for inspection or replay.
Why it is interesting
It treats skill selection as a per-step routing decision and pairs Jev’s ranking with local identity checks, explanations, replay, and non-blocking fallback behavior.