What it is
Security hook package that asks Jev three typed questions (risk, user_requested, from_untrusted) before tool runs and screens results for prompt injection afterward.
What it does: Deny / ask / allow with session context across many coding-agent hosts; also scans skills/plugins for unexpected behavior.
Why it belongs in the showcase: Clear tool-call guardrail + untrusted-input screen pattern with a public npm distribution.
Caveats: Provider key required; host permission APIs differ; measure thresholds on your traffic.
Evidence
Public MIT README + npm jev-guard; hooks for Claude Code, Codex, Copilot, Gemini, Cursor, pi, OpenCode, ACP.
Caveats
- Needs TypeSafe / AI Gateway / OpenRouter key
- Author-reported latency/price — calibrate thresholds yourself
- Host ask/deny semantics differ (some warn-only)
Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.