jev-preflight
Claude Code plugin that scores a turn's redacted diff on eight Jev risk axes at Stop and asks Claude to reinvestigate once if high.
Overview
jev-preflight is a Claude Code plugin that catches risky code changes before Claude finishes a turn. On prompt submit it creates a private Git baseline; at Stop it selects and redacts the turn's diff and sends it to Jev to score eight risk axes in one request: behaviour regression, authorisation, input validation, data integrity, error handling, compatibility, lifecycle and regression tests. Below the threshold Claude finishes; at or above it in assist mode, Claude gets at most one additional investigation and should change code only if it finds evidence.
The README is clear that scores are signals for attention, not proof of defects, that the default 0.85 threshold is uncalibrated, and that the plugin does not replace tests, linters, SAST or secret scanning. Without a key the hook fails open. It ships as Go binaries for macOS, Linux and Windows. Status is v0.1.0 public beta, verified with synthetic code, and it is independent of TypeSafe and Anthropic.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How much risk does this turn's diff carry on each of eight axes (behaviour regression, authorisation, input validation, data integrity, error handling, compatibility, lifecycle, regression tests)? | — |
How it works
- Create a private Git baseline on prompt submit
- Claude makes changes
- At Stop, redact the diff and send it to Jev
- If any axis passes the threshold, Claude reinvestigates once
- Otherwise Claude finishes
Strengths
- Bounded: one request and at most one extra investigation
- Diff is redacted before sending
- Fails open without a key
Caveats
- Default threshold is uncalibrated
- Public beta verified only on synthetic code
- Not a merge blocker or substitute for tests
Getting started
claude plugin marketplace add muse0509/jev-preflight, then claude plugin install jev-preflight@jev-preflight and enable it.At a glance
- Type
- Project
- Status
- beta
- Language
- Go
- Runtime
- Claude Code plugin
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A bounded Jev risk check for Claude Code: eight risk axes, one request, one optional reinspection.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Risk scores on eight axes for the code changed in a Claude Code turn.
Will it block my work?
No. At most it asks Claude to investigate once more before finishing.
Is the threshold tuned?
No. The README says the default 0.85 is uncalibrated.
Related projects
Stop hook for Claude Code and Codex that refuses 'done' until a check passes after the last edit; Jev advises but never causes a block.
CLIbetaTypeScriptCoding-agent hook that asks Jev one yes/no per project rule on every edit or turn and makes the agent fix any violation.
HookbetaTypeScriptClaude Code plugin with rules and checks that make agents prove their work; an optional Jev judge triages review findings.
PluginbetaMarkdown + shellClaude Code plugin with nine Jev-judged gates for rules, scope, intent, completion and claims that escalate but never approve.
HookbetaTypeScriptCompiler that turns agent rules in CLAUDE.md and similar files into testable Jev programs: typed questions plus a coded verdict.
ProjectbetaTypeScriptRisk gate for AI coding agents: Jev decides at five layers and Claude Code hooks enforce deny or ask deterministically.
ProjectbetaJavaScript