Permissions Management for AI Agents
Most IT systems have ways to authorize and log sensitive actions. This project asks how to design versions of those systems for AI agents, given that agents are changing so rapidly and that the controls may need to be built without support from the agent's developers.
The goal is a fork of a popular open-source agent, such as gemini-cli, with authorization and AI control techniques built in, along with evidence that they don't degrade the agent's performance and will be easy to maintain as the agent evolves.