Jev Sentinel open-sources per-action agent monitor that scored 53,870 HF payloads, flagging 98.3%
schwentker · x · 2026-10-01
Developer shapor has open-sourced Jev Sentinel, a per-action monitor for AI agents covered by TechCrunch. Before any agent action runs, Jev scores it against the task's scope and returns allow / ask-a-human / block with a plain-English reason — catching what LLMs 'shouldn't read'.
- Ships with full result data from 53,870 Hugging Face incident payloads, 98.3% flagged
- Minimal setup: Python 3.10+ (stdlib only), a Jev API key, and Claude Code or Codex
- Protect a project by describing its job in one sentence via make install-hook, which wires a hook into Claude Code to validate every proposed action
- Built at JEVATHON (SF, 2026-09-26) as a hackathon project, not production code
Context: TechCrunch previously called OpenAI's "Decisions API" a Jev clone, underscoring the value of fast, cheap intelligence.
More from coding & agent
- Survey aims to map how professional software engineers actually use coding agents — thegautamkamath · 2026-10-01
- OpenAI engineer runs all her engineering work through her dot agent — jxnlco · 2026-10-01
- 17 open-source GitHub repos to learn AI engineering from scratch — ZabihullahAtal · 2026-10-01
- Human engineers still far faster than agents, if the product is technical enough — remilouf · 2026-10-01
- Harness Learning: Adapting Agent Scaffolds at Test Time Without Touching Model Weights — burny_tech · 2026-10-01
- The orchestration layer, not the rail, is where agentic payment decisions happen — mikeflache · 2026-10-01