User repository on GitHub
Autonomous Jev pull-request review with typed decisions, calibrated approval gates, and trusted-owner escalation
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Jev Review — thiago-ss
Autonomous PR review bot that consumes PR metadata and patches, asks Jev a finite set of typed questions across correctness, security and verification perspectives, then applies deterministic gates in code. Suggestions are advisory; it never edits or merges.
- Stars
- 0
- Language
- Python
- Published by
- Community
- Author
- thiago-ss
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Jev Review — devagrawal09.
About this project
A small autonomous PR review bot built around TypeSafe's Jev. It consumes structured PR metadata and patches, asks finite typed questions, then applies deterministic approval gates. Uncertain or risky reviews route to trusted owners with a structured explanation. Suggestions are advisory; the bot never edits or merges code.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Three typed perspectives - correctness, security and verification - added to comment-only reviews, with disagreements and per-file hypotheses preserved.
- Dry-run is the default, and review --execute only simulates active policy evaluation locally and never calls GitHub.
- Probability distributions and the vendor confidence statistic are kept separately, and model certainty alone never authorizes approval.
- The shipped calibration example is synthetic and reports ready: false; synthetic, stale, duplicated or mismatched evidence fails closed.
Watch out
Read the status line before anything else: the GitHub App is installed on the author's own repository with scheduled evidence comments enabled and approval writes disabled, and the README says production approval is gated on a named repository, trusted deployment configuration and representative held-out calibration evidence that the included synthetic examples do not satisfy. On evidence quality it is unusually careful - the ten synthetic bug, repair and metadata-injection probes are described as revealing false alarms as well as successful holds, and the ten-case real API run is said to match all synthetic approval labels while not establishing production accuracy or calibration. The three review perspectives are correlated prompts, not independent reviewers, so agreement between them is not corroboration. Operationally it is Python 3.9+ on the standard library, dry-run by default; only the github and poll subcommands can write externally, with JEV_COMMENTS=false opting out of evidence comments and JEV_EXECUTE=true enabling gated approvals and reviewer requests. Credentials, policy and calibration must come from the bot's default branch, never a PR branch, and the CLI does not automatically load .env. Not documented: a real held-out accuracy number, token cost per review, or latency. 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.
No, and confidence alone is explicitly insufficient. The README states model certainty alone never authorizes approval: Jev's probability distributions and a vendor confidence statistic are preserved separately, and approval runs through deterministic gates in code. Suggestions are advisory and the bot never edits or merges code. Approval writes are currently disabled on the installed app, and enabling them takes JEV_EXECUTE=true on top of the production readiness gate.
Less than the docs structure suggests, and the README says so in several places. The ten synthetic bug, repair and metadata-injection probes reveal false alarms as well as successful holds and do not establish production calibration. The shipped calibration example is intentionally synthetic and reports ready: false. The final ten-case run against the real API matched all synthetic approval labels, which the README immediately qualifies as not establishing production accuracy or calibration.
Locally it is Python 3.9+ and the standard library - create a venv, run the unittest suite under tests, then python -m jev_review review --input FILE. Supplying a review field exercises the pipeline offline; omitting it calls Jev, so you need TYPESAFE_API_KEY in the environment. The CLI reads environment variables and does not automatically load .env, which surprises people. For the App there is a setup guide in docs/github-app.md covering predefined permissions, installing on selected repositories, and running with short-lived scoped tokens.
Less than it sounds, and the README labels it: the correctness, security and verification passes are correlated prompts, not independent reviewers. What you get is a structured x-ray whose reports preserve disagreements, per-file hypotheses, missing evidence and exact CI links, rather than three votes to count. Values the prompts disagree on are kept in the report instead of being averaged away, which is the point of the exercise.
The GitHub App is installed on thiago-ss/jev-review and scheduled evidence comments are enabled, but approval writes are disabled. Production approval is gated on a named repository, trusted deployment configuration and representative held-out calibration evidence, and the included synthetic examples do not satisfy that gate.
It needs Python 3.9+ and the runtime uses the standard library. Create a venv, run the unit tests, then python -m jev_review --help. review --input FILE takes a JSON packet containing pr, and optional review, config and calibration; supplying a review exercises the pipeline offline, while omitting it calls Jev using TYPESAFE_API_KEY.
Jev returns probability distributions and a vendor confidence statistic, which are preserved separately, and model certainty alone never authorizes approval. Production evidence must match the concrete response model, prompt, schema and frozen policy; synthetic, stale, duplicated or mismatched evidence fails closed.
More Jev resources like this
51GitHub stars
Local LinkedIn recruiting workspace on Browser Use's Jev Ultrafast: Jev screens titles, drives navigation, and assesses profile excerpts against your brief while a live browser stays visible; matches and evidence export as JSON.
Jev Agents & MCP#community#jevbest#catalog-import
Community
skeptrunedev/jev-recruiterPython
skeptrunedev/jev-recruiterPython33GitHub stars
Codex Skill plus stdlib-only Python CLI that recommends which installed skill to read next for a task, allowing no-skill or uncertain answers and never executing candidates.
Jev Agents & MCP#community#jevbest#catalog-import
Community
win4r/jev-skill-suggesterPython
win4r/jev-skill-suggesterPython72GitHub stars
MCP server over stdio that gives coding agents fourteen Jev judgment tools, turning evidence plus an enumerable question into auto, review, or escalate actions from calibrated probabilities.
Jev Agents & MCP#community#jevbest#catalog-import
Community
PyModel/jev-judge-mcpPython
PyModel/jev-judge-mcpPython21GitHub stars
Research runtime that pages a large logical tool and context space through a small bounded Jev decision surface with explicit approvals.
Jev Agents & MCP#community#jevbest#catalog-import
23GitHub stars
Coding-agent plugin that routes prompt classification, subagent tiering, edit rule checks, and fast Jev-backed compaction through small typed judgments.
Jev Agents & MCP#community#jevbest#catalog-import
Community
0x7067/claude-jevTypeScript
0x7067/claude-jevTypeScript14GitHub stars
Opt-in Hermes plugin that routes each turn to one skill: a wide Jev Choice plus request checks gate a narrow fits round, injecting a single ignorable hint line.
Jev Agents & MCP#community#jevbest#catalog-import
From the Jev community
Posts from builders shipping with Jev right now.