Open-Source ORIMUSU Splits Agent Intent from Permissions with Three-Way Verdicts
GROUND_ink · reddit · 2026-11-18
The founder of ORIMUSU shares an open-source (Apache-2.0) local Node.js CLI that separates what an agent decides to do from what it's permitted to do. Every proposed operation is checked against explicit permissions and evidence, yielding ADMITTED, DENIED, or UNRESOLVED—with UNRESOLVED explicitly not treated as a silent allow. It's an early Technical Preview: default mode only observes, blocking mode runs against a local mock broker, and real credential isolation isn't demonstrated yet. He asks practitioners how they handle permissions today and what such a check would need to prove before being allowed to block agents.
More from coding & agent
- AI models a scene from scratch in Blender with no assets, author says barely intervened — op7418 · 2026-11-18
- Kilo Desktop Goes GA: Run Many Coding Agents and Review Diffs in One Window — anacondainc · 2026-11-18
- Researcher dissects artex: an MCP-powered pentest toolkit, not an AI, despite FBI claims — weskambale · 2026-11-18
- The legendary CLI coding agents battle royale: 6 agents, one instruction — kill the others — SIGKITTEN · 2026-11-18
- Security researcher: patching vulnerabilities one by one is a losing game in the AI coding era — agstrait · 2026-11-18
- Founder vibe-codes a dashboard to track all his AI agents across machines — Ben_Reinhardt · 2026-11-18