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

claude-jev

Claude Code plugin that adds Jev as an independent judge for review findings, debugging hypotheses, design options and search hits.

PluginbetaMIT
View on GitHub ↗

Overview

This claude-jev plugin by buchmark turns questions about your code into typed Jev queries and returns calibrated probabilities, as a second opinion an agent cannot argue past. When Claude reviews a diff, hunts a bug or picks a design, every judgment otherwise comes from Claude alone.

It exposes five MCP tools. Review findings go through Jev before you read them, so false positives are dropped with a number attached. Competing debugging hypotheses are ranked against the symptom. Implementation options are rated on three separate axes so trade-offs become visible when they disagree. Large sets of grep hits are filtered before any are read. The README explains Jev's question types, including that a Noul has no separate confidence value and a value near 0.5 means yes and no are about equally likely.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this review finding a real issue?yes · no
ChoiceWhich hypothesis best explains the symptom?candidate hypotheses
ScoreHow does this design option rate on each of three axes?—
Yes / noIs this search hit relevant?(inferred)yes · no

How it works

  1. Claude produces candidates (findings, hypotheses, options, hits)
  2. Plugin sends them to Jev via MCP tools
  3. Apply thresholds to keep, rank or drop
  4. Claude continues with the filtered set

Strengths

  • Independent judge for Claude's own claims
  • Trade-offs made explicit
  • Less context spent on irrelevant hits

Caveats

  • Needs a TypeSafe key
  • Quality depends on question design

Getting started

Install as a Claude Code plugin from the repository and set TYPESAFE_API_KEY.

At a glance

Type
Plugin
Status
beta
Runtime
Claude Code plugin + MCP server
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Adds Jev checks to Claude Code review findings, debugging hypotheses, design options and search results.

    logicrw/awesome-jev-projects

  • Claude Code plugin that scores review findings, debug hypotheses and design options with TypeSafe's Jev — calibrated probabilities instead of one more opinion.

    hellogumbo/awesome-jev

Frequently asked questions

What does it judge?

Review findings, debugging hypotheses, design options and search results.

Why a second judge?

So Claude's assertions get an independent, numeric check.

What does a Noul of 0.5 mean?

Yes and no are about equally likely; it is not a confidence value.

Related projects

  • jevex

    Single MCP tool that indexes a repo and has Jev rank a shortlist so a coding agent reads the right files and lines instead of grepping.

    MCP serverbetaPython
  • jgrep (kyu1204)

    Semantic grep for code, diffs and CSV rows that packs 16 chunks per Jev request and prints file:line hits with CI-friendly exit codes.

    Agent skillbetaTypeScript
  • Antigravity-mcp-semantic-search-with-TypeSafeAi

    MCP server that finds relevant code with Jev Choice and Noul checks instead of grep sweeps, and audits diffs against task intent.

    MCP serverprototypePython
  • jev-retrieval

    Grep-shaped Rust CLI for coding agents: BM25 recall plus Jev Noul window checks and a listwise Choice return path:line results with scores.

    CLIbetaRust
  • siftr

    CLI and MCP server giving coding agents Jev-powered semantic search, focused reads and list picking in about a second for about a cent.

    BenchmarkbetaPython
  • BGTS Context Engine

    Deterministic code-graph context for coding agents on Postgres, AGE and pgvector, with optional Jev or local re-selection of symbols.

    MCP serverbetaPython