A Claude Code plugin that verifies every tool call before it runs, asking Jev five parallel questions about the plan, scope, prior action and stated intent, then allowing, asking or blocking. Fresh installs ship in audit mode.
User repository on GitHub
Every tool call your agent makes, checked before it runs. A Claude Code plugin that uses TypeSafe AI's Jev to verify each pending tool call against the session plan, then allows it, asks you, or blocks it. Proof of concept
Repo stats from the GitHub API, cached Oct 2026.Homepage
A Claude Code plugin that verifies each agent action before it executes — not after the run finishes. It routes every tool call through a purpose-built verification model, TypeSafe AI's Jev, instead of a full LLM review, which is what makes checking every step affordable. That dialog is what mode: enforce looks like. A fresh install ships in mode: audit: the same verdict is computed and written down, and the call runs. You turn the gate on yourself, with /stepwarden enforce.
Quoted from the project's README. Read the full README.
Watch out
The README labels this a proof of concept built on two early-access dependencies, Claude Code function hooks and Jev, states that its default thresholds have not been tuned on production traffic, and says plainly it is not a substitute for deterministic security controls. A fresh install runs in audit mode: every call is verified and logged but nothing is blocked or interrupted until you run /stepwarden enforce. When a human cannot be asked, which covers a headless -p run, a dismissed dialog, or more than 50 questions in a session, onNoAnswer decides and it defaults to allow. The middle tier is deliberate: blocking on a merely probable signal produces false stops exactly where a gate matters most, so block is reserved for near-certainty and the wide band in between asks you instead. Both thresholds are inclusive, bad configuration values fall back to defaults with a warning rather than taking the gate down, and thresholds that would make ask unreachable are repaired and reported. It needs Claude Code 2.1.273 or newer with function hooks enabled, and the plugin has no runtime dependencies and ships no lockfile, so installing it never runs npm. Keys go to the OS keychain through the configure command, and Claude Code does not read a .env file on its own. Copy reviewed against the repository README on 2026-10-04, not from running it.
tyler-dot-earth/patdownTypeScript
kyu1204/oh-my-harnessTypeScript
zjunlp/JevLoopTypeScript
mizchi/jev-test-filterTypeScript
Nasrallah-AL/jev-cliTypeScript
Brainwires/jevwireTypeScriptPosts from builders shipping with Jev right now.