agent-gate-loop
GitHub Action where an AI agent turns an issue into a PR behind guards, checks, a reviewer and a Jev gate that escalates when unsure.
Overview
agent-gate-loop is a reusable GitHub Action that turns a labelled issue into a pull request. An AI agent (Claude) writes the change, then a four-part gate checks it before a human sees it: guards (protected paths, no deleted tests, size limit), your own typecheck, lint and test commands, a separate read-only Claude reviewer, and Jev.
Jev checks each reviewer finding and asks whether the change does what the issue asked, whether it touches unrelated code, and how good the tests are. If the gate fails, the agent gets one merged feedback note and tries again up to max_rounds; if Jev is unsure, the loop stops for a human. The action never merges. The README marks it as a prototype with known gaps, such as the agent being able to alter .git/config and check commands being able to read secrets, so it should only run on issues from trusted people.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does the change do what the issue asked? | yes · no |
| Yes / no | Does the change touch unrelated code? | yes · no |
| Score | How good are the tests? | — |
| Yes / no | Is this reviewer finding valid?(inferred) | yes · no |
How it works
- Issue labelled agent:run or /agent comment triggers the action
- Jev intake, then the agent writes a fix
- Guards, your checks and a read-only reviewer run
- Jev gate decides pass, fail or escalate
- On fail, feedback loops up to max_rounds; result is a PR plus report, never a merge
Strengths
- Layered gate before human review
- Escalates on uncertainty instead of guessing
- Never merges automatically
Caveats
- Prototype with documented security gaps
- Needs TypeSafe and Anthropic credentials
- Only for issues from trusted authors
Getting started
Copy examples/consumer.yml to .github/workflows/agent.yml and add TYPESAFE_API_KEY plus ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN secrets.At a glance
- Type
- Project
- Status
- prototype
- Language
- TypeScript
- Runtime
- GitHub Action
- Has code
- yes
Use cases and integrations
How curated lists describe it
Reusable GitHub Action: agent fix loop gated by checks, an AI reviewer, and TypeSafe Jev.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev check?
Reviewer findings, whether the change matches the issue, whether it touches unrelated code, and test quality.
Will it merge PRs?
No, it only opens a PR with a report.
Is it safe for public repos?
The README says to use it only on issues written by people you trust because of known gaps.
Related projects
Software-factory supervisor where Jev independently judges coding agents' completion, test sufficiency, progress and need for a human.
ProjectbetaPythonClaude Code plugin with rules and checks that make agents prove their work; an optional Jev judge triages review findings.
PluginbetaMarkdown + shellClaude Code Stop hook that blocks an unverified done: one four-question Jev call only when files changed and no check passed.
PluginbetaTypeScriptClaude Code plugin that scores a turn's redacted diff on eight Jev risk axes at Stop and asks Claude to reinvestigate once if high.
ProjectbetaGoClaude Code plugin where Claude plans phased work with checks and, at each stop, Jev judges whether the work is really done.
PluginbetaTypeScriptStop hook for coding agents where Jev checks each attempted stop against plain-language rules and sends the agent back if one is violated.
HookbetaPython