claude-jev
Claude Code plugin sending small judgments to Jev: rule checks that block edits, verbatim compaction, subagent sizing and routing hints.
Overview
claude-jev aims to stop paying frontier-model prices for small judgments in Claude Code. It sends quick questions to Jev: what kind of prompt is this, does it need tools, does this edit break a rule in CLAUDE.md, which turns still matter when context fills up.
Edits that break instruction files are blocked with a file:line citation; the README reports re-scoring 247 accepted edits blocked 23 (9.3%). Compaction keeps the exact rows that matter instead of writing a summary, taking about a second, and planted user constraints survived every time in its tests. Each subagent spawn gets a model tier, and briefs missing paths, acceptance criteria, verification or commit policy are sent back once. Each prompt gets a one-line routing hint. Every hook fails open on errors, missing keys or timeouts, with compaction falling back to Claude Code's own summary. It needs only python3, works with TypeSafe or OpenRouter keys, and also installs in Pi.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this edit break a rule in CLAUDE.md or AGENTS.md? | yes · no |
| Choice | What kind of prompt is this, and does it need tools? | — |
| Yes / no | Does this context row still matter? (compaction)(inferred) | yes · no |
| Choice | Which model tier should this subagent get? | — |
How it works
- Hooks intercept prompts, edits, subagent spawns and compaction
- Jev answers the relevant typed questions
- Block rule-breaking edits with citations; add routing hints
- Compaction keeps verbatim rows Jev marks as relevant
- Fail open on any error
Strengths
- Verbatim compaction in about a second
- Edits blocked with file:line citations
- Fails open everywhere
Caveats
- Blocking rate measured on the author's own repos
- Compaction needs a recent Claude Code version
Getting started
claude plugin marketplace add 0x7067/claude-jev && claude plugin install claude-jev@claude-jev, then set TYPESAFE_API_KEY or OPENROUTER_API_KEY.At a glance
- Type
- Plugin
- Status
- beta
- Language
- Python
- Runtime
- Claude Code plugin (also Pi)
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Claude Code plugin: Jev for rule checks, verbatim compaction, and prompt routing
logicrw/awesome-jev-projectsClaude Code plugin: Jev for rule checks, verbatim compaction, and prompt routing.
hellogumbo/awesome-jevClaude Code plugin that sends the small judgements — what kind of prompt is this, does it need tools, does this edit break a rule — to a decision model instead of a frontier one.
MrJev/awesome-jev
Frequently asked questions
What happens without a key?
The hooks switch off silently.
How does compaction differ?
Jev keeps exact relevant rows instead of writing a summary.
Can it block my edits?
Yes, edits that break instruction-file rules are blocked with a citation.
Related projects
Three Italian-language Claude Code prompts that build Jev-powered model routing, context compaction and code review into your project.
ProjectprototypeCoding-agent hook that asks Jev one yes/no per project rule on every edit or turn and makes the agent fix any violation.
HookbetaTypeScriptGlue that turns one Linux machine into a team of coding agents, with Jev making typed calls like who builds a task and whether it may merge.
ProjectbetaTypeScriptAdvisor for coding agents: git and Stop hooks flag test tampering, secrets and skipped hooks, with optional Jev review findings.
HookbetaClaude Code mod that picks reasoning effort per prompt with a Haiku judge, optionally Jev, and shows cache and context state.
ProjectbetaTypeScriptDesktop workspace for coding agents that runs tasks in parallel worktrees, listed as using Jev to pick the best agent per task.
ProjectbetaTypeScript, Rust