Claude agent confirmations

claude-agents provides the managed-agent tool-confirmation backend. It is off by default on a fresh installation and requires governance. Its descriptor is olivares.claude-agents, with API routes under /v1/m/claude-agents/.

The module reads a session’s thread events and lets a human approve or reject a pending tool confirmation. It uses governance’s approval state machine; enabling this module does not start an agent or approve its tools automatically. Decisions are audited against a redacted fingerprint of the session and tool use, rather than storing the raw tool payload in the decision record.

Event reads require governance:approval:read. Confirmations require governance:approval:admin and a stable human user identity; a system token cannot confirm a tool. With no thread-event source, the event list is empty; an error from a wired source is reported as an upstream failure.

Use Edition & modules in the console, or the CLI:

olivares modules ls
olivares modules on claude-agents
olivares modules off claude-agents

See governance for the approval model and live operation & sessions for session runtime.