AgentTrace is a defensive digital-forensics and incident-response (DFIR) tool. It is designed to analyze evidence you are legally authorized to investigate (your own AI-agent deployments, or incidents you are engaged to investigate with proper authority). It has no capability to access remote systems, accounts, or data, and it runs fully offline.
It must not be used for attack, exploitation, unauthorized access, surveillance, or covert monitoring.
If you discover a security vulnerability in AgentTrace itself (for example, a path-traversal in bundle extraction, a way to forge a chain-of-custody signature, or a parsing crash on malformed evidence), please report it responsibly:
- Do not open a public issue for undisclosed vulnerabilities.
- Contact the maintainers privately (see repository contact/security advisory).
- Include reproduction steps and, if possible, a minimal proof of concept using synthetic data only.
We aim to acknowledge reports promptly and coordinate a fix and disclosure.
Because this tool processes forensic evidence:
- Raw sensitive content is never stored in normalized events — only redacted
summaries plus content hashes (see
redact_summaryinmodel.py). - Integrity and chain-of-custody are verifiable via hash chains and signed manifests; report any weakness in these mechanisms as a security issue.
- Treat all ingested evidence as untrusted input.