DiffJury

Next.js app that runs one Jev call on a public GitHub PR to score risk, review depth, blockers and a verdict.

Projectprototype

Overview

DiffJury takes a public GitHub pull request URL and, with one click, fetches the PR dossier (title, body, diff, contributors) and runs a single Jev judgment. The result is shown as risk bars, noul probabilities and a verdict, with no generated essays. Large PRs are trimmed locally to fit Jev's token limits; small PRs keep full context.

The single call asks nine questions: overall merge risk, missing tests, documentation debt and blast radius as Scores; review depth (skim, standard or deep) and a verdict (approve with nits, request changes or block) as Choices; and whether design discussion, security review or a merge blocker is needed as Nouls. The directory describes it as a PR risk router that can decide how much human review a PR needs before a reviewer is assigned, doubling as a review coach. The TypeSafe key stays server-side.

What the decision model decides

Type	Question	Answers
Score	risk: overall merge risk	—
Choice	review_depth: how deeply should this PR be reviewed?	skim · standard · deep
Yes / no	needs_security: is a security review needed?	yes · no
Choice	verdict: what should happen to this PR?	approve_with_nits · request_changes · block

How it works

Paste a public PR URL
Server fetches the dossier and trims large diffs
One Jev call answers nine typed questions
Show risk bars, flags and verdict; route review effort accordingly

Strengths

One call per PR
No generated text to read
Key kept server-side

Caveats

Public PRs only
Large diffs are trimmed, which may hide issues
Not a replacement for human review

Getting started

npm install, copy .env.example to .env.local with TYPESAFE_API_KEY, then npm run dev and open http://localhost:3000.

At a glance

Type: Project
Status: prototype
Language: TypeScript
Runtime: Next.js web app
Has code: yes

Use cases and integrations

Classification & Routing306

Code review gating119

How curated lists describe it

Code review: routes each pull request by risk with Jev before a human reviewer is assigned, doubling as a review coach.
yibie/awesome-jev
DiffJury — TypeSafe Jev PR risk router + code review coach.
hellogumbo/awesome-jev

Frequently asked questions

What does DiffJury decide?

PR risk, review depth, whether design or security review is needed, merge blockers and an overall verdict.

Does it write review comments?

No, it returns scores, probabilities and a verdict only.

Does it work on private repos?

The README describes fetching public PRs only.

Related projects

agent-stack
Glue that turns one Linux machine into a team of coding agents, with Jev making typed calls like who builds a task and whether it may merge.
ProjectbetaTypeScript
git-jev-stage
Git tool that stages only the hunks matching a plain-language description, using Jev include/exclude/mixed probabilities per hunk.
ProjectbetaTypeScript
Jackalope
Desktop workspace for coding agents that runs tasks in parallel worktrees, listed as using Jev to pick the best agent per task.
ProjectbetaTypeScript, Rust
jev-playwright
Playwright reporter that uses Jev to select end-to-end tests relevant to a code change before browser execution in CI.
ProjectbetaTypeScript
jev-pr-labeler
Labels GitHub pull requests with Jev via OpenRouter: type, area, platform, attention and a size based on conceptual scope.
ProjectbetaPython
jev-triage
Probot GitHub app that classifies, scores, checks completeness and dedupes issues and PRs with Jev, routing uncertain ones to humans.
ProjectbetaTypeScript