User repository on GitHub
Semantic linter for Go. Write rules as questions in plain English; the AST decides where to look, a model decides if the code is right.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Tools & CLIs
Semcheck
Go linter whose rules are plain-English questions: AST matchers select nodes and Jev answers one Noul per fragment above the rule's threshold. It runs standalone, under go vet, and as a golangci-lint plugin, with cached answers so runs only ask about changed code.
- Stars
- 1
- Language
- Go
- License
- MIT
- Published by
- Community
- Last verified
- Oct 2026
About this project
A linter for Go whose rules are questions in plain English. Static analysis decides where to look: deterministic matchers on the syntax tree select exact nodes, such as a log call or a function named GetSomething. A decision model decides whether the code is right: it answers a closed question about each node with a probability, and semcheck reports the ones above a threshold. Which model is up to you: see Decision models. Two rules, as they are written in .semcheck.yml:
Quoted from the project's README. Read the full README.
- Languages
- Go
- Capabilities
- Open sourceRequires API key
Highlights
- Two shipped rules: no-pii-in-logs asks whether matched log calls write personal data; name-matches-behavior flags Get/Is/Has functions that write state.
- All 12 sampled findings from the two shipped rules across caddy, cloudflared, and pocketbase were right; the README labels that 100% precision.
- First run on three projects: 6,412 questions took 203 seconds and 3.75 million tokens, about 16 cents for 300,000 lines of Go.
- Rules support call and func-prefix matchers, exported-func-doc, test-func, statement or function context, and per-rule min_confidence (0.9 default).
- Utility flags: -dry-run estimates cost, -record records every question, -stats reports cache hits and tokens, and SEMCHECK_JUDGE=fake:0.95 stubs the model.
Watch out
MIT license; the shipped provider sends code fragments and type notes to TypeSafe and requires TYPESAFE_API_KEY, and the README notes the model gives a probability, not a reason, with answers varying a few hundredths between requests.
Questions about this entry
Short answers drawn from the linked project or page.
Deterministic matchers decide where to look, so the model only sees selected nodes; answers are cached under a hash of question, code, and type notes, so a run only asks about changed code. On its own repo, a change touching half the files asked 30 of 201 questions.
TypeSafe Jev is the default provider and requires TYPESAFE_API_KEY; each question is answered as a Noul probability and compared to min_confidence. Any other model can be added behind the one-method Judge interface.
go install github.com/arturobermejo/semcheck/cmd/semcheck@latest, then semcheck ./... exits 3 when it has findings, or go vet -vettool=$(which semcheck) ./... runs it as a vet tool. It reads .semcheck.yml from the current directory or the closest parent, honours //nolint:semcheck, and can also ship as a golangci-lint module plugin built with golangci-lint custom.
Each rule pairs a deterministic matcher with one closed yes-or-no question: call globs resolved with types, func-prefix words, exported-func-doc or test-func, plus a context of statement or function and min_confidence (0.9 by default). The README advises asking about a contradiction visible in the fragment, being concrete, leaving out adverbs, then reading the numbers with -record before you set the threshold.
Five rules ran on caddy, cloudflared and pocketbase at fixed commits, and 96 stratified questions were labelled without seeing the answers. no-pii-in-logs and name-matches-behavior were right on every labelled sample, log-level-fits was 60% and ships at a raised 0.7 threshold, and doc-matches-code (33%) and test-name-matches (10%) did not ship because right and wrong findings carried the same probabilities. The labels were made by Claude, not a person.
A hosted judge receives the code fragment of every question plus the notes on its types; with the shipped provider that is the TypeSafe AI API, and semcheck sends nothing else and only when TYPESAFE_API_KEY is set. Answers are cached under a hash of question, code and type notes, with SEMCHECK_CACHE choosing another directory or off. A judge error warns and continues by default so an outage does not turn every build red; fail_on_judge_error: true makes it fail instead.
Yes. semcheck -dry-run ./... counts the questions and estimates their cost without asking anything, and -dry-run -record=q.jsonl writes every code fragment that would have been sent. For wiring tests, SEMCHECK_JUDGE=fake:0.95 stubs the model so it says yes to everything. Once running for real, -record appends every question with its answer as a line of JSON and -stats reports how many answers per package came from the cache and how many tokens were billed.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Thread on Selmar's Jev code-review calibration notes and ordering.py comparisons.
Lectem2 pointsCalibrating Jev as a Code ReviewerRead the thread on Hacker NewsHacker News: Jev classifies agent PR hunks as P0, P1, or P2 so reviewers see P0 first.
namanbhulawat47 pointsJev Based Code ReviewRead the thread on Hacker NewsX: Magic Jev Ball for code review
X: jev-review: a local-first score loop for coding agents
More Jev resources like this
2.6kGitHub stars
Cisco's security scanner for agent skills added a System One analyzer: it sends the full skill text to POST /v1/systemone as one advisory Noul question and records the probability, while being deliberately unable to emit a finding or change a severity.
Jev Tools & CLIs#community#security#agent-skills
Community
cisco-ai-defense/skill-scannerPython
cisco-ai-defense/skill-scannerPython56GitHub stars
grep, sort, cut and uniq that match by meaning: each line whose Jev Noul clears the threshold survives grev, sibling filters route by Choice and rank by Score, and the output is always your own input, never generated text.
Jev Tools & CLIs#community#cli#coreutils
Community
aurorainfra/grevGo
aurorainfra/grevGo11GitHub stars
Fuzzy linter hook for Claude Code and Codex: after the agent edits a file, it sends the added code to Jev, which answers Noul checks against team rule packs in about 0.3 seconds, letting the agent fix violations before code review.
Jev Tools & CLIs#community#linter#claude-code
Community
ckorhonen/jev-lintHTML
ckorhonen/jev-lintHTML3GitHub stars
Go CLI for Jev with noul, choice, score, eval, rank, and models commands: rank sends one Score question per item per weighted YAML dimension in a single request and sums weight times score in code, while threshold flags turn a failed gate into exit code 10 for scripts and CI.
Jev Tools & CLIs#community#cli#go
Community
saembit/jeff-cliGo
saembit/jeff-cliGo0GitHub stars
CLI quality gate that asks four typed questions about each change - three Score lenses for commit quality, doc alignment and breaking-API risk, plus a Noul secret audit - and fails a diff, commit message or doc set below the resulting 0-100 score. Runs from a terminal, a git pre-commit hook or a GitHub Action.
Jev Tools & CLIs#community#quality-gate#ci-cd
Community
sumant1122/jevciJavaScript
sumant1122/jevciJavaScript1.9kGitHub stars
A Go-based team workspace with optional Jev AI for agent auto-routing, task auto-fill, auto-assign, and automation condition nodes.
Jev Tools & CLIs#community#go#agents
Community
Paca-AI/pacaGo
Paca-AI/pacaGoFrom the Jev guides
Original write-ups on what this entry is for.
The request shape barely changes across surfaces; the surrounding contract does. When to call the API directly and when a gateway, a Worker binding, or a router entry earns its extra hop.
Jev Tools & CLIs#gateway#vercel#cloudflare
Read article
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
Jev labels a record that already exists. Code moves, flags, or holds it. How the document, mail, and transfer tools keep the action out of the model.
Jev Tools & CLIs#classification#email#security
Read article
From the Jev community
Posts from builders shipping with Jev right now.