Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-flash-review

Local MCP review engine where each rule becomes a Jev YES/NO/N/A Choice over agent-supplied diff chunks, with located evidence.

MCP serverbeta
View on GitHub ↗

Overview

jev-flash-review is rule-based code review for AI coding agents. The calling agent supplies a unified diff plus task context (business purpose, boundaries and invariants); the local engine never reads the repository. Every rule in the rules directory becomes a typed Choice (YES, NO or N/A with an applies-if condition), batched per chunk and rule set and run in parallel. Each rule takes its most severe outcome across chunks.

For violations, a second Choice over the diff's hunk markers picks the location ('select instead of generate') and code prints file and line; violations with no hunk are reported as absences (missing tests, docs, handling) or PR-level issues, and weak locations are adjudicated further. It ships three skills (review a PR, review anything, loop until clean) for Claude Code and Codex, with a bundled MCP registration. The dist bundle must be built locally after cloning.

What the decision model decides

TypeQuestionAnswers
ChoiceDoes this diff chunk violate this rule?YES · NO · N/A
ChoiceWhich hunk marker locates the violation?—

How it works

  1. Agent builds a diff plus task context
  2. Engine chunks the diff and asks a Choice per rule per chunk in parallel
  3. Merge to the most severe outcome per rule
  4. Locate evidence by choosing hunk markers and return structured verdicts

Strengths

  • Never reads the repository directly
  • Locations selected rather than generated
  • Loop-until-clean skill

Caveats

  • Must run npm run bundle after cloning
  • Rule quality determines review quality

Getting started

`/plugin marketplace add lucvalse/jev-flash-review` and `/plugin install jev-flash-review@jev-flash-review` in Claude Code, after `npm run bundle`.

At a glance

Type
MCP server
Status
beta
Language
TypeScript
Runtime
Local MCP server with agent skills
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Jev decide?

Whether each rule is violated in each diff chunk, and where the evidence is.

Does it read my repo?

No, the agent supplies the diff and context.

Which agents can use it?

Claude Code and Codex, through plugins and a bundled MCP config.

Related projects

  • lorenzini

    Claude Code skills that wait for AI pull-request reviewers and judge whether their reports actually permit a merge, never inferring a pass.

    Projectbeta
  • jev-codes

    CLI and agent commands auditing a git diff against YAML standards packs, scoring each hunk with Jev so agents fix only flags.

    Agent skillbetaTypeScript
  • jevkit

    Self-hosted Jev server plus CLIs, agent skills and a Pi tool for typed code questions and per-file diff review triage.

    Agent skillbetaPython
  • Jevmind

    Zero-dependency Python toolkit of nine agent skills whose typed decisions pass a code gate and are sealed in a hash-chained ledger.

    ProjectbetaPython
  • typesafe-jev-tools

    Claude Code hook that, when you write an LLM call or classifier, injects a three-way test: plain code, frontier model or Jev.

    HookbetaBash
  • decisionkit

    CLI, MCP server and Claude Code plugin for closed-form yes/no, choice, score and rank decisions via Jev, OpenAI Decisions or Claude.

    MCP serverbetaTypeScript