What it does
Sensitive paths stop locally; other tool calls receive a risk score that can require interactive approval, with configurable fail-closed behavior.
Agent tooling
TypeSafe AI for OpenClaw
Add Jev-backed tool approval, chat triage, transcript curation, and model routing to OpenClaw.
Bookmark: TypeSafe AI for OpenClaw
Keep this in your collection.
Leave a noteWhat would you try with this?
: TypeSafe AI for OpenClaw
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
A developer can add this tool to an OpenClaw project to check planned actions. The app asks an outside AI service called TypeSafe to score the task. If TypeSafe gives the action a high risk score, the local app pauses. It then waits for a person to approve the step.
The source shows how to set a risk limit in the project settings. A developer must provide a private access key to connect the software to TypeSafe. The local project stops actions marked dangerous. It cannot guarantee that it catches every unsafe step.