Claude Code skill for the bulk judgement calls — which of these files, which of these log lines, which of these tickets — that otherwise get read one by one.
View on GitHub →Checked, not fully reviewed
A Claude Code skill that hands bulk filter, classify and rank calls to Jev. Checked, not fully reviewed: skips secret-named files, not passwords in config.
Good for
Watch out for
Tested Oct 10, 2026 at 5d6fe5cd4f9d · Node 24 in Docker; its CLI (`qs.mjs`) driven against a local stand-in via QUICKSILVER_API_BASE, over a fixture with ten planted markers, inside and outside a git repository
Checked, not fully reviewed. We ran its CLI against a stand-in answering in TypeSafe’s published shape, over a fixture with ten planted markers, and counted every request. No Jev calls. The README’s token and speed figures are the author’s benchmark and we have not reproduced them.
Quicksilver installs as a Claude Code skill plus a zero-dependency Node CLI. When a task looks like
“read a lot to decide a little”, Claude runs filter, classify, rank or find over files,
directories, globs or stdin. Each item goes to Jev as its own request, and Claude gets back a short
list with confidences instead of the whole pile.
What leaves the machine is the item itself, so the file walker is the thing to check:
| planted in | sent |
|---|---|
.env, .env.local, .ssh/id_rsa, deploy/server.pem, config/secrets.json |
no — listed as “secret-like, never sent” |
a file named in .gitignore, inside a git repository |
no |
| the same file, in a directory that is not a git repository | yes |
config/database.yml (password: …) |
yes |
terraform.tfvars (db_password = …) |
yes |
config/aws.ini (AWS_SECRET_ACCESS_KEY=…) |
yes |
The skip list is name-based and good at what it covers, and it reports what it skipped, which few
tools do. What it does not cover is secrets in files with ordinary names, the same gap we found in
jevgrep. We reported it
with a short list of content patterns to check before sending. Until then, point it at source
directories rather than config/.
The rest is solid. The key is saved to ~/.quicksilver/config.json with mode 600. 408, 429 and
5xx including 529 are retried, honouring retry-after: one 529 was followed by a normal answer.
The API base is overridable, and it pins nothing by default (jev-latest), but --model takes a
version. One rough edge: setup reports a key as unverifiable when TypeSafe answers its check with
a single 529.
MIT, 116 stars, Node 18+.
See how it compares with other tools in Best Jev tools, tested hands-on.
Review updated Oct 10, 2026. Numbers quoted from the project are its author's own; we don't publish our own measurements of Jev.
kerpopule/hermes-jev-skills
Bundle of skills that hand an agent's small decisions to Jev: model routing, skill selection, retrieval filtering, compaction, and computer use, with a routing dashboard. Works with Hermes, Claude Code, and Codex.
wuyoscar/jev-skill
Nine installable agent skills — triage, routing, code review, document and UI work — with a catalogue of scenarios to copy.
jkudish/jev-mcp
Proof-of-concept MCP server with ready-made tools for fact checking, prompt-injection detection, and semantic ranking.
New Jev releases, pricing changes, and the best new projects, once a week. No spam; unsubscribe anytime.
Powered by Buttondown. See our privacy policy.