Add pre-open-source security review

[joshconverse]

Summary

• Adds a comprehensive security_review.md documenting a full security audit of the codebase before open-sourcing
• Covers all 8 categories from the Vibecoder Security Review Checklist: secrets, auth, data privacy, test vs prod, file handling, dependencies, hygiene, and injection/RCE
• Reviewed all 31 Python files, 2 shell scripts, 1 HTML file, config files, and test examples

Key Findings

Severity	Count	Top Issue
CRITICAL	1	Unrestricted bash tool allows LLM agents to execute arbitrary shell commands (agent/tools.py:226)
HIGH	3	Unrestricted file read/write by LLM agents; predictable /tmp/ paths vulnerable to symlink attacks
MEDIUM	4	Hardcoded Layr-Labs org reference in graph.py:79; observability server serves $HOME with wildcard CORS
LOW	5	Debug flag stuck on, error message leakage, file handle leaks
INFO	4	Passing checks (no hardcoded secrets, no unsafe deserialization, clean CI/CD)

Test plan

• Review each finding for accuracy and agreed severity
• Prioritize remediation of CRITICAL and HIGH findings before public release
• Run trufflehog / gitleaks on full git history (recommended in report)
• Run pip-audit or uv audit on locked dependencies
• Note: GitHub Dependabot already flagged 1 moderate vulnerability on the default branch

🤖 Generated with Claude Code