How is this different from running a full LLM review on every PR?

Jev triages every PR cheaply from diff, files, and commit messages, and the slower fallback model runs only for human-review PRs at elevated risk instead of on every PR.

What does the fallback review model default to?

Empty means auto-pick the cheapest available language model on your Gateway account at call time. Set fallback-review-model or repo config fallback_review_model to pin one.

Where do I put thresholds so repo policy wins over action inputs?

Put thresholds in .github/jev-gatekeeper.yml. Repo config wins over action inputs, which win over built-in conservative defaults.

Why does it use commit statuses instead of checks?

It uses the Statuses API rather than Checks for the same maintainer-visible outcome with less permission: statuses:write instead of checks:write.

Which four questions does Jev answer on each pull request?

should_review, risk, route, and touches_secrets. The action applies repo-configurable risk thresholds to those answers to compute the final route: auto-approve, human-review, or block.

What are the default risk thresholds for the gatekeeper?

risk-threshold-for-review defaults to cosmetic, risk-threshold-for-block to blocking, and fallback-review-risk-threshold to blocking. A repo config file takes precedence over action inputs, which take precedence over those defaults.

When does the detailed secondary review run?

Only for human-review PRs at elevated risk. By default it uses whichever language model is cheapest on your AI Gateway account at that moment, and it costs roughly 5-10x the latency and token volume of the Jev triage call.

CommunityJev Integrations & Adapters

Ci Gatekeeper Bot Jev

GitHub Action that triages pull requests with four Jev typed questions and routes each PR to auto-approve, human-review, or block with a commit status and explanatory comment.

See highlights

Open github.com

Stars: 1
Language: Shell
Published by: Community
Author: NemanjaManic
Last verified: Oct 2026

NemanjaManic/ci-gatekeeper-bot-jev

User repository on GitHub

View on GitHub

GitHub Action that uses Jev (TypeSafe AI via Vercel AI Gateway) to cheaply triage pull requests before expensive LLM/human review

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A GitHub Action that uses Jev (TypeSafe AI, via the Vercel AI Gateway) to cheaply and quickly triage pull requests before an expensive LLM or human review. See specs/001-jev-pr-triage/ for the full spec-kit design (spec, plan, research, data model, contracts, tasks) and .specify/memory/constitution.md for the project's governing principles. On every pullrequest event (opened, synchronize, reopened), the action: Fetches the PR's diff, changed files, and commit messages.

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

Tagscommunity github

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Shell

Highlights

Runs on pull_request opened, synchronize, and reopened using diff, files, and commit messages.
Asks should_review, risk, route, and touches_secrets, then applies configurable risk thresholds.
Posts or updates a PR comment without including raw diff content even on secret hits.
Elevated-risk human-review PRs can trigger a secondary review on the cheapest Gateway model.

Watch out

Needs github-token and ai-gateway-api-key secrets; Jev can return moderate risk even on trivial diffs under conservative defaults. 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

HA-SystemOne

Home Assistant integration exposing decision-model APIs through a native AI Task entity, typed jev actions, YAML decision sensors and an Assist router.