🛠 agent-pd: An Audit Tool for Claude Code
agent-pd has been released — a specialized tool for auditing and monitoring the actions of Claude Code agents. It functions as a "flight recorder," logging tool usage and changes in access permissions, including the actions of hidden sub-agents. The system uses deterministic detectors to identify violations without consuming LLM tokens.
🌍 The emergence of such tools addresses the "black box" problem, where the actions of autonomous AI agents become unpredictable or dangerous to the system.
👤 Claude Code developers gain a transparent way to track whether an agent is attempting to step outside the project scope or execute dangerous commands.