aigaze 0.1.0

Audit AI agent actions from transcript files

aigaze

Audit AI agent actions from transcript files.

Install

pip install aigaze

Usage

# Auto-discover transcript files on your machine
aigaze discover

# Scan a specific transcript file or directory
aigaze scan /path/to/transcripts/

# Auto-discover + scan all in one step
aigaze scan --auto

# Output as JSON
aigaze scan --auto --format json

# Write report to file
aigaze scan --auto -o report.json --format json

What it does

AIGaze reads AI agent transcript files (e.g. VS Code Copilot Chat .jsonl logs) and scans them for security-relevant actions:

• R1: Credential file access (.ssh/, .env, .aws/credentials, etc.)
• R2: Dangerous terminal commands (curl|sh, rm -rf /, etc.)
• R3: Hardcoded secrets in AI-generated code (API keys, tokens)
• R4: Out-of-workspace file access
• R5: External URL access outside allowlist

Supported AI Tools

• VS Code Copilot Chat
• Cursor (planned)
• Claude Code (planned)

License

AGPL-3.0-only