User repository on GitHub
Semantic code review with Jev, plain-English rules and Agent Skills.
- Forks
- 2
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Hunch — Kelbie
A CLI and agent skill that asks plain-English questions across a repository, sends each in-scope chunk to the provider, returns scored source locations to investigate, and runs recurring review rules on diffs.
- Stars
- 4
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- Kelbie
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Hunch — carldaws.
About this project
Ask a plain-English question across a repository without guessing which words the code uses. Hunch sends each in-scope chunk to Jev and returns scored source locations for a person or coding agent to investigate. It also runs recurring review rules on diffs.
Quoted from the project's README. Read the full README.
- Languages
- Python
Highlights
- find asks a behavior question rather than a keyword, so you need not guess the words the code uses; rg stays the right tool for exact symbols.
- Provider choice is per run, gateway, typesafe or semif, where semif runs an open model on your own machine and nothing leaves the box.
- check runs recurring rules locally, across whole files with --all, or on pull requests through GitHub Actions or the GitHub App.
- Reports separate result limits from search coverage and list omissions, handing back scored locations for a person or agent to investigate.
Watch out
The interpretation rules matter more than the tool. Search scores rank candidates and findings flag configured concerns; neither proves a bug or the absence of one. A complete run means the selected chunks were evaluated, not that every relevant behavior was found, which is why reports identify omissions and distinguish result limits from search coverage. Cost depends on input tokens, repeated context and how many questions you ask; the README says small model prices make broad scans practical while latency and recall still need measuring on your repository. Code is sent to the selected provider unless you use --provider semif. SemIf is an independent project whose accuracy is its own, so evaluate it on your own rules before it gates anything. Requirements are Node 22+, Git and a one-time provider auth login, and the README tells you to use rg for exact symbols and read the returned source before concluding. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
Because rg needs you to already know the words. Hunch takes the behavior in plain English, sends each in-scope chunk to the provider and returns scored source locations, which is what you want when one behavior has many implementations or several different names. The README is not anti-rg though: use rg for exact symbols and text, and read the returned source and follow its dependencies before drawing a conclusion.
No, and the README says so directly. Search scores rank candidates and findings flag configured concerns; neither proves a bug or proves its absence. A complete run means the selected chunks were evaluated, not that every relevant behavior was found, and reports exist to identify omissions and to separate result limits from search coverage, which are different failures.
Yes, with --provider semif, which runs SemIf, an open model locally, so the only thing involved is your own GPU. There is also --vercel, which reuses an existing Vercel CLI login instead of a key, and every command takes --provider gateway, typesafe or semif for that run. The tradeoff is accuracy: SemIf is an independent project, so run eval on your own rules before letting it gate a pull request.
Cost depends on input tokens, repeated context and the number of questions, with the README noting small model prices make broad scans practical while latency and recall still need measurement on your repo. For recurring use, init configures plain-English or typed questions, check runs them locally, check --all sweeps whole files, and pull requests are covered through GitHub Actions or the GitHub App, with policy read from the immutable base commit.
Run npx skills add Kelbie/hunch. It needs Node 22+ and Git, and provider authentication via npx -y @kelbie/hunch auth login once, after which it works in every directory.
Use --provider gateway, typesafe or semif. Semif runs the open SemIf model on your own machine with no key, and --vercel reuses a Vercel CLI login instead of a key.
Run hunch init to configure plain-English or typed questions, then check locally, with check --all across whole files, or on pull requests through GitHub Actions or the GitHub App.
More Jev resources like this
51GitHub stars
Local LinkedIn recruiting workspace on Browser Use's Jev Ultrafast: Jev screens titles, drives navigation, and assesses profile excerpts against your brief while a live browser stays visible; matches and evidence export as JSON.
Jev Agents & MCP#community#jevbest#catalog-import
Community
skeptrunedev/jev-recruiterPython
skeptrunedev/jev-recruiterPython33GitHub stars
Codex Skill plus stdlib-only Python CLI that recommends which installed skill to read next for a task, allowing no-skill or uncertain answers and never executing candidates.
Jev Agents & MCP#community#jevbest#catalog-import
Community
win4r/jev-skill-suggesterPython
win4r/jev-skill-suggesterPython32GitHub stars
Daily AI news pipeline that gathers RSS, arXiv, and social sources with staged agents and publishes a searchable web frontend.
Jev Agents & MCP#community#jevbest#catalog-import
Community
flyryan/ai-news-aggregatorPython
flyryan/ai-news-aggregatorPython72GitHub stars
MCP server over stdio that gives coding agents fourteen Jev judgment tools, turning evidence plus an enumerable question into auto, review, or escalate actions from calibrated probabilities.
Jev Agents & MCP#community#jevbest#catalog-import
Community
PyModel/jev-judge-mcpPython
PyModel/jev-judge-mcpPython21GitHub stars
Research runtime that pages a large logical tool and context space through a small bounded Jev decision surface with explicit approvals.
Jev Agents & MCP#community#jevbest#catalog-import
18GitHub stars
Python CLI and MCP server for keyless web search over DuckDuckGo or self-hosted SearXNG, with an optional local Laya sieve that filters, reranks and stance-checks evidence before it reaches an agent.
Jev Agents & MCP#community#jevbest#catalog-import
From the Jev community
Posts from builders shipping with Jev right now.