jev-in-codex
Codex plugin for batch labelling JSONL records with Jev under custom policies, with Codex reviewing uncertain labels.
Overview
Jev in Codex is a Codex plugin for batch labelling with Jev. Its single tool, jev_label, turns a JSONL file of text records into a fully labelled file using custom question and criteria policies or a built-in feedback_theme preset, while Codex reviews uncertain decisions. Earlier tools for capability selection, context search and output triage were removed because tested workflows showed no useful improvement, and there is no lexical or local fallback.
The README reports that its 0.3 custom-policy benchmark on action-needed, sentiment, document routing and feedback tasks used 21.5 to 31.9 percent less total Codex input and 8.4 to 23.5 percent less time across four paired repetitions, with all records labelled correctly, and that a 0.2 benchmark on feedback measured 42.7 percent less input. These are 64-record synthetic workloads, and the author is explicit that they do not prove real-world accuracy. New capabilities must pass a reviewable benchmark against plain Codex before being exposed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which label from the user's taxonomy fits this record? | — |
How it works
- Prepare an approved JSONL file in the workspace
- Define a labelling policy or use the preset
- jev_label assigns one label per record
- Codex reviews uncertain decisions
Strengths
- Reduced Codex input and time on synthetic tests
- Explicit admission policy for new features
- Negative results published
Caveats
- Synthetic 64-record benchmarks
- No fallback if Jev is unavailable
Getting started
Install the plugin in Codex and call jev_label on an approved UTF-8 JSONL file; see docs/LABELLING.md.At a glance
- Type
- MCP server
- Status
- beta
- Language
- TypeScript
- Runtime
- Codex plugin (MCP)
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Jev-powered tool and skill selection, context search, and output triage for Codex via MCP
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
One label per record from your taxonomy.
Why were other tools removed?
Their tested workflows showed no useful improvement.
How much does it save?
21.5 to 31.9 percent of Codex input on synthetic benchmarks, per the README.
Related projects
Claude Code plugin that builds reusable Jev signatures for classification or scoring and runs CSV, Excel or text datasets through them.
PluginbetaPythonLive race labelling 1,000 app reviews with four typed Jev questions versus Gemini 3.8 Flash JSON output.
AppprototypeJavaScriptPipeline that ingests Google Maps restaurant reviews and classifies sentiment, five rating pillars and flags with Jev or an LLM.
ServiceprototypeTypeScriptpandas and Polars .jev accessor for classifying, sentiment-scoring and grading DataFrame rows with Jev probabilities.
DemobetaPython- jev-arena★ 122
Side-by-side tool comparing Jev and DeepSeek on labelling the same comments for relevance, sentiment and intent, with replay and reports.
ProjectbetaTypeScript Unix filter CLI that pipes files, lines or records to Jev for bulk yes/no, choice or score judgments so agents need not read everything.
ProjectbetaPython