ClearFrame – an open-source AI agent protocol with auditability and goal monitoring

ClearFrame is an open-source protocol and runtime designed to address security and transparency issues in AI agent systems, such as untrusted content handling, plaintext API keys, lack of audit logs, and undetected goal drift. It introduces features like reader/actor isolation, a GoalManifest for alignment scoring, a Reasoning Transparency Layer for structured decision logging, and an HMAC-chained audit log for tamper-evidence. The project includes a control plane for real-time monitoring and is available as a Python library with CLI tools. It targets users concerned with prompt injection, credential leakage, and regulatory compliance.