Agent Chaperone

Review before execution

Agent Chaperone screens tool calls before they run and tool results before the agent reads them. It connects through an MCP proxy or hooks for a client’s built-in tools, applying configurable deterministic and model-assisted policies.

Policies, approvals, and logs

Shadow mode records decisions so developers can review how a policy behaves before enabling enforcement. Approval handling and local decision logs make the review process inspectable. The project includes configuration tooling and guides for integrating it with existing clients.

I contributed the initial implementation and subsequent development, fixes, and releases. Screening is an additional review layer; operating-system isolation and the client’s permission controls remain necessary. Model-assisted screening sends the reviewed content to the configured backend.