Skip to content

Jev Gates

Claude Code plugin with nine Jev-judged gates over edits, prompts, shell commands, stops and commits. Each one escalates to a permission prompt, and none of them can approve an action.

Stars
0
Language
JavaScript
License
MIT
Published by
Community
Author
rashedInt32
Last verified
Oct 2026

About this project

Nine calibrated gates for Claude Code, judged by TypeSafe Jev. Each one escalates. None of them ever approves. A live run, nothing staged. Every probability and latency on screen is what Jev returned at render time. MP4 version.

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

Highlights

  • From 0.9.0 only the claims gate is enabled and only its Stop hook is wired, after a week of sessions found the intent guard cost about 650 ms per prompt.
  • Rule guard rides a PreToolUse hook over Edit, Write, MultiEdit and shell writes, reads CLAUDE.md rules, and caps at 64 with prohibitions first.
  • The claims gate checks each sentence of the reply twice: does it assert something done or observed, and does this turn's output support it.
  • Bash guard skips plainly read-only commands before spending a request, but chained rm, git branch -D, sort -o and awk system() do not qualify.

Watch out

Only the claims gate is on by default and only its Stop hook is wired; every other gate needs an environment variable, and the README's own session findings are the reason. The rule and scope guards interrupted far more often than they caught violations on real CLAUDE.md files that mix rules with prose, and the intent guard's edit asks landed mostly on scratch files. Hooks do not inherit your shell profile, so the key must sit in a file they can read, chmod 600, or in the env block of your Claude settings. The claims gate only considers replies from turns that ran tools, minus any sentence it already flagged, so an unchecked claim cannot vouch for itself, and this turn's outputs outrank earlier replies. The done gate catches silence rather than disagreement: declining out loud with a reason counts as addressed. The prompt check blocks nothing and never rewrites your prompt; a gap is two checks at or below the threshold or one at half of it, and a single borderline score varied run to run so it is ignored. No gate can approve anything, only escalate. The only evaluation figures given are 45 real prompts where it flagged the same 3 of 17 change requests across two runs and skipped 6 follow-ups; per-gate cost is described as about half a second and a fraction of a cent. Our README copy is truncated mid-sentence inside the claims-gate evidence section, so the proof gate, the commit guard, threshold defaults and the instructions for turning the other gates on are not documented here. 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

4GitHub stars
A DeepSeek Harness plugin that replaces DSH's size-based context compaction with Jev judgments: tool results that are still needed are never trimmed however large, and spent ones get a code-written receipt instead of a model-written summary.
Jev Agents & MCP#community#awesomejev#catalog-import
1GitHub stars
A DeepSeek Harness plugin that exposes System One decisions as agent tools and adds an opt-in auto-guard, with a labeled benchmark it re-runs against the live API.
Jev Agents & MCP#community#jevbest#catalog-import
14GitHub stars
Pi extension that pairs a local coding model with hosted Jev judgments for skill hints, file retrieval, and evidence ranking, with caching and failure bounds on every call.
Jev Agents & MCP#community#jevbest#catalog-import
20GitHub stars
TypeScript recipe library with 248 focused functions that call Jev for routing, reranking, gating and labelling, returning structured ready-or-review decisions for Node backends.
Jev Agents & MCP#community#awesomejev#catalog-import
7GitHub stars
MCP server exposing one evaluate tool for Jev Choice, Noul, and Score questions, plus a separate eval harness that compares Jev against LLMs on shared labeled sets.
Jev Agents & MCP#community#jevbest#catalog-import
6GitHub stars
Skill and CLI that helps Codex, Claude Code, and other agents add bounded Jev judgments for routing, ranking, classification, and next-step choice.
Jev Agents & MCP#community#awesomejev#catalog-import
Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai