Jev in production › Agent guardrails and approvals

Jev Guard Harness

Local Go gateway checks each coding-agent tool call against deterministic policy, asks Jev for risk and confidence on ambiguous calls, returns allow, ask or block, and fails closed when Jev is unreachable; policy fixtures pass 9/9 (author).

Open on GitHub ↗

9/9measured, as published by the source
Use
Agent guardrails and approvals
Industry
Security
Form
Open-source tool
Stage
Beta
Plugs into
OpenCode, Kiro CLI, Claude Code
Listed
2026-09-24
Found via
github
Repository
dereknguyen269/jev-harness
Stars
6
Forks
1
Last push
2026-10-01
Language
Go
License
none stated

The README opens with

A local safety gateway for AI coding agents. Every tool call goes through jev-guard, which normalizes it, applies deterministic policy, consults Jev for ambiguous cases, and fails closed when Jev is unreachable. A React dashboard manages policy, users, approvals, and audit.

Jev never decides. Jev returns a structured judgment (risk/confidence/action); the Go policy engine turns it into the final allow / approvalrequired / block.

Interactive version (pan/zoom, trace, search): docs/jev-guard-how-it-works.html. Source: docs/jev-guard-flow.workflow.json.

Badge

For the project's own README, linking back here:

Listed in Jev in production

Also used for agent guardrails and approvals