Why not just add a policy-checker model call instead?

A second model call trades one unreviewable judgment for another. This emits narrow typed questions plus a reducer with thresholds you can read, diff, and unit-test, keeping facts like clocks and test exit codes out of the model entirely.

Will it convert my whole CLAUDE.md prompt into Jev?

Only the decision slice. Generation never compiles because Jev emits no text, and procedure is control flow for code. Every compilation returns the decision set, the residual prompt for the generative part, and the wiring between them.

What is the first end-to-end run on my own repo?

Install the jev-compiler package for the jevc command, run jevc scan to sort instruction files into decidable, procedure, and generation, then jevc compile with lift to get a lowering request for the agent you already have open.

How do I wire a compiled gate without breaking the agent?

Copy the generated .claude directory, register the gate in settings for PreToolUse, and replay offline with recorded fixtures before going live. The honest middle rule asks the human when the model is uncertain instead of guessing.

How do I install and run the first commands?

npm install -g jev-compiler installs the jevc command; run jevc check to replay the recorded corpus offline (58 fixtures), then jevc scan . to see what a project could enforce.

Which instruction files does scan read?

The files agent harnesses actually read: CLAUDE.md, AGENTS.md, GEMINI.md, .cursorrules, .github/copilot-instructions.md, .claude/skills/**, and .claude/agents/**, sorting each rule into decidable, procedure, or generation.

What does the Claude Code plugin add?

/plugin marketplace add doronp/jevc followed by /plugin install jevc@jevc adds one skill, /jevc:jevc, which runs the scan, lift, and compile for you and asks before it installs a gate.

CommunityJev Integrations & Adapters

Jevc

Compiler that turns ignored markdown rules into testable Jev gates: typed questions answered with probabilities plus a reducer computing allow, ask, or deny in code. Ships with offline replay fixtures and Claude Code hook wiring.

See highlights

Open github.com

Stars: 8
Language: TypeScript
License: APACHE-2.0
Published by: Community
Author: doronp
Last verified: Oct 2026

doronp/jevc

User repository on GitHub

View on GitHub

Compile agent policy prose into deterministic verdict programs: narrow evidence questions for the model, the verdict computed in code. Install: npm i -g jev-compiler

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

Turn the rules your agent keeps ignoring into gates you can test.

Quoted from the project's README. Read the full README.

Tagscommunity awesomejev catalog-import

Works forJev for Classification Jev for Tool calling

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Compiles rules into Jev typed questions plus an ordinary-code verdict reducer.
Scan sorts CLAUDE.md, AGENTS.md, skills, and agents into decidable versus other rules.
Offline corpus replay and sample-project gates run with no key or network.
Emits bouncer policies with source-line comments and residual prompts for the rest.

Watch out

Only decision text compiles; generation and procedure stay in code or the residual prompt, and live checks need a key. 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.

More Jev resources like this

system-one

Provider-neutral TypeScript runtime for System One decisions: write choice, noul, and score code once, then run it on TypeSafe Jev, local Reflex, any compatible /v1/systemone endpoint, or a custom provider.