AI & Automation

AgentTrail Scrub AI transcript secret cleaner

AI coding assistants like Claude Code, Cursor, and Aider frequently store raw conversation logs and workspace caches that accidentally capture plaintext API keys, passwords, and environment variables on disk. AgentTrail Scrub audits these local agent transcripts and sanitizes sensitive credentials before they are inadvertently exposed or shared. ## Key features - **Targeted Agent Discovery**: Automatically detects transcript stores and session histories across Claude Code, Cursor workspaces, Gemini CLI, Aider, Continue.dev, and Codex directories. - **Entropy-Assisted Detection**: Validates Anthropic, OpenAI, GitHub, AWS, and JWT tokens using regular expressions coupled with Shannon entropy thresholds to minimize false positives. - **Safe In-Place Redaction**: Replaces identified raw secrets with masked placeholders while preserving structured file copies in a backup directory. - **Overlap Suppression and Safeguards**: Prevents duplicate reporting on intersecting match spans, ignores allowlisted tokens, and skips files exceeding 100 MB. ## Example ```bash python agenttrail_scrub.py --help python agenttrail_scrub.py --scan python agenttrail_scrub.py --redact ``` ## Limitations AgentTrail Scrub only scans and redacts files on your local filesystem; it does not automatically revoke or rotate compromised keys with external API providers.

python agenttrail_scrub_ai_transcript.py --help