Will it skip a test that my change actually breaks?

It can. Scoring is probabilistic, with touched-code and low-confidence-near-cutoff rules that run extra tests rather than fewer. Use the JSON log to audit scores and keep full runs for release gates.

Why should I use --exec instead of pasting the output myself?

Use --exec with a double dash so args go straight to spawn with no shell re-splitting. Stdout args look quotable but break under unquoted command substitution, which is why --exec exists.

Will a suspicious snap-file update fail my CI?

No. Snap-file verification is advisory and never fails CI. Changed snap files get plausible, review, or unknown with scores, and low-confidence or missing answers route to human review.

Why does Rust need a special flag while other languages do not?

Rust is opt-in because listing tests runs cargo test with list and triggers a build nobody asked for. Pass --format rust explicitly, and pass --format go to disambiguate mixed repos.

What does the JSON output give me?

The full scoring: every test with its score, confidence, and the reason it was kept or dropped, plus the runner argv and a spend block with calls, input and output tokens, milliseconds, retries, rate-limit hits, tokens per second, splits, and a dollar figure. It is the shape to read, log, or feed to another program.

How does it decide which tests to keep when it is unsure?

Two rules run more tests rather than fewer. A test whose own body is inside the diff is selected as touched without any question being asked and costs no tokens, and an uncertain verdict near the cutoff still runs because confidence routes rather than gates. Only a clear below-cutoff no is dropped.

Is the model allowed to read my repository files?

No. It gets one shared state, the diff, and one question per test in a single round trip; there is no agent loop and no file reading by the model. Source-based discovery uses git ls-files, so untracked and ignored files are not considered, and Playwright with --format playwright uses Playwright's own collected list instead.

CommunityJev Agents & MCP

Jev Test Filter

Diff-aware test selector that sends the git diff to Jev with one question per test, then prints the native filter flags and hands them to vitest, jest, node, bun, Playwright, cargo, or go test.

See highlights

Open github.com

Stars: 35
Language: TypeScript
License: MIT
Published by: Community
Author: mizchi
Last verified: Oct 2026

mizchi/jev-test-filter

User repository on GitHub

View on GitHub

Score every test against a git diff with Jev, and emit the filter arguments vitest, node:test, Playwright, cargo test and go test already understand

Forks: 2
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

jev-test-filter reads a git diff, asks a model how much that change can alter the outcome of every single test in the repository, and prints the filter arguments your test runner already understands. It hands those arguments to the runner you already use — vitest, jest, node --test, bun test, Playwright, cargo test, go test or rspec — so the run covers the tests the change could plausibly break instead of all of them.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification Jev for Ranking Jev for Moderation

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

One shared diff as state, one question per test, one round trip, with no agent loop and no model file reads.
Three shapes: --exec to run the runner, --json with per-test scores and spend, or quoted args on stdout.
Explicit exit codes separate printed args, nothing selected, bad flags, and tool failure without ambiguity.
Snap-file mode grades changed snap edits as plausible, review, or unknown without editing files or tests.

Watch out

Selection is advisory and probabilistic, so keep full-suite runs for releases. Truncated diffs, missing test sources, and image snap files yield unknown rather than approval, and one suspicious entry marks a whole snap file for review. 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.

More Jev resources like this

patdown

Node CLI that lints a file tree against fuzzy natural-language rules in one Markdown file, judging each matched file with a swappable TypeSafe-backed judge.

Jev Agents & MCP#community#jevbest#catalog-import

Community

tyler-dot-earth/patdownTypeScript

jev-guard

Claude Code plugin that scores pending tool calls and finished answers with Jev noul probabilities, denying the call with the correction attached when a rule threshold is met.

Jev Agents & MCP#community#awesomejev#catalog-import

Community

muratcakmak/jev-guardTypeScript

pi-jev-context

Pi context extension that folds repeated reads, filters long command logs through Jev, and retrieves saved originals as searchable verbatim chunks.

Jev Agents & MCP#community#jevbest#catalog-import

Community

Nyarlathoteppppp/pi-jev-contextTypeScript

jev-mcp

MCP server exposing Jev Choice, Score, Noul, batched ask, file-reading triage, and model listing as typed tools with full probability distributions.

Jev Agents & MCP#community#jevbest#catalog-import

Community

rashedInt32/jev-mcpTypeScript

typesafe-mod

A Claude Code function hook that sends AskUserQuestion dialogs to Jev and draws a probability per option above the dialog, so you still choose while seeing how confident the model is.

Jev Agents & MCP#community#jevbest#catalog-import

Community

BeLazy167/typesafe-modTypeScript

mastra-jev-moderation

One-file Mastra input processor that screens the last user message with Jev instead of an LLM verdict: a typed yes/no gate plus a log-only category label, in a single request that fails open behind a circuit breaker.

Jev Agents & MCP#community#jevbest#catalog-import

Community

CodeAlive-AI/mastra-jev-moderationTypeScript

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai