Jev, LangGraph, and Tenuo combine to build a safe dependency upgrade agent
Jev in practice: typed decisions, scoped authority
Tenuo Engineering built safe-upgrade, an open-source agent that uses Jev to turn repository evidence into typed probabilistic decisions, LangGraph to carry those decisions through a durable workflow, and Tenuo to grant each action a task-scoped warrant. The design rule: keep judgment, control flow, and authority separate. The pattern extends beyond dependency upgrades to any CI agent that modifies code or uses repository credentials.
Keep the component that recommends an action separate from the component that defines its authority.
- gyre007
Shockingly vibed blog post. Actually hurt my brain reading this and I had to stop.
- fuzzumms
Interesting way to guarantee some sort of safety for automated decisioning
- dj0_
The task bound idea is interesting. I’m starting to hear more of this theoretically on AI enterprise architecture. Has anyone actually implemented it successfully?