Question 1

Is this a skill manager that will auto-load skills for me?

Accepted Answer

No. It recommends procedures and nothing more: it does not execute skills, grant permissions, or override the agent's governing instructions. The result is advisory, the agent still follows the user's instructions, and an explicit skill request is resolved locally before any probabilistic ranking runs.

Question 2

Do I need a TypeSafe key to try it at all?

Accepted Answer

Not to explore. Offline fixture demos, the readiness report, and replaying a recorded evaluation all work without a credential or network calls. Fresh ranking does require your own account for the selected provider — TypeSafe is the default and Cloudflare-hosted TypeSafe Jev is also supported. There is no local model in the box.

Question 3

What happens with a library of 400-plus skills?

Accepted Answer

Quill lexical prefiltering from FrankenSearch narrows the roster locally first, then Jev sees at most 254 eligible skills plus a none option per Choice; smaller rosters reach Jev in full. Ranking is two passes: a broad comparison, then a rerank of the shortlist with bounded descriptions and body excerpts.

Question 4

Why did it not suggest the skill I expected?

Accepted Answer

Run it with --why-not and the skill id to see where the candidate was excluded, with the thresholds and concrete recovery hints; that adds no inference calls. Both comparison passes also include a real 'none of these' option, so abstaining is a normal outcome rather than a failure, and relevance gates can drop a skill before ranking.

Question 5

What runtime and platforms does SkillRanker support?

Accepted Answer

It is a standalone Rust CLI (Rust 2024) installed on Linux or macOS, with output as JSON, agent hooks or an inline TUI. On macOS the README says to keep state on a local APFS or HFS+ filesystem with Unix permissions. The installer verifies a SHA256 checksum and installs sr into ~/.local/bin.

Question 6

How can I try it with no key and no network?

Accepted Answer

Run sr demo --case useful to inspect an offline labeled fixture without a key, network or private session. sr rank --context scratch/context.json --dry-run previews the redacted wide-pass request with no network or persistence, and sr doctor and sr capabilities report setup and available integrations.

Question 7

What happens when a library has more than 254 skills?

Accepted Answer

Quill, a lexical prefilter from FrankenSearch, narrows the candidates locally before Jev evaluates them, admitting up to 254 skills plus a none option to each Choice. Smaller rosters reach Jev in full, and explicit skill requests resolve locally before either stage.

SkillRanker — Dicklesworthstone

About this project

Highlights

Questions about this entry

Reactions & coverage

Other listings of this project

From the Jev guides

From the Jev AI model community