jev-review (thiago-ss)
Autonomous PR review bot with typed Jev questions, deterministic approval gates and escalation of uncertain reviews to trusted owners.
Overview
This Jev Review is a small autonomous pull-request review bot. It consumes structured PR metadata and patches, asks Jev finite typed questions, and applies deterministic approval gates; uncertain or risky reviews are routed to trusted owners with a structured explanation. Suggestions are advisory, and the bot never edits or merges code.
A pilot adds three typed perspectives (correctness, security and verification) to comment-only reviews, preserving disagreements, per-file hypotheses, missing evidence and CI links; the author notes these are correlated prompts, not independent reviewers. An interactive atlas and synthetic probes show false alarms as well as successful holds. Status: installed as a GitHub App on its own repository with evidence comments enabled and approval writes disabled; production approval requires a named repository, trusted configuration and representative held-out calibration evidence, which the synthetic examples do not provide. The CLI defaults to dry run, and a calibrate command reports readiness.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this patch correct / secure / adequately verified?(inferred) | yes · no |
| Choice | Should the PR be approved, commented on or escalated to an owner?(inferred) | approve · comment · escalate |
How it works
- Collect PR metadata and patches
- Ask Jev typed questions from three perspectives
- Apply deterministic approval gates
- Post evidence comments; escalate uncertain cases to owners
- Only approve when calibration evidence and configuration allow
Strengths
- Dry-run by default with explicit opt-ins
- Calibration readiness gate before approvals
- Standard-library Python runtime
Caveats
- Approval writes disabled pending real calibration evidence
- Synthetic probes show false alarms
- Perspectives are correlated, not independent
Getting started
python3 -m venv .venv && .venv/bin/python -m unittest discover -s tests -v, then .venv/bin/python -m jev_review review --input examples/review.jsonAt a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- CLI and GitHub App
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Autonomous Jev pull-request review with typed decisions, calibrated approval gates, and trusted-owner escalation.
hellogumbo/awesome-jev
Frequently asked questions
Will it approve PRs automatically?
Not yet; approval writes are disabled until representative calibration evidence exists.
Can it change code?
No. It never edits or merges code.
What happens on uncertainty?
The review is routed to trusted owners with a structured explanation.
Related projects
Claude Code PreToolUse layer where Jev answers YAML-defined questions about each tool call, with 19 hard rules and calibration runs.
ProjectbetaTypeScriptOpenAI-compatible endpoint that answers an agent's command-approval call with APPROVE, DENY or ESCALATE from a Jev Choice, failing closed.
ProxybetaPythonGitHub Action that asks Jev three yes/no questions about a pull request and approves it only when all clear a confidence threshold.
ProjectbetaTypeScriptCoding-agent tools on Jev with public benchmarks: a strict command gate, a code-review router Action, and use cases.
MCP serverbetaTypeScriptNatural-language linter combining ast-grep selectors with Jev to catch misleading names, stale comments and tests that miss their claims.
ProjectbetaTypeScriptCompiler that turns agent rules in CLAUDE.md and similar files into testable Jev programs: typed questions plus a coded verdict.
ProjectbetaTypeScript