# Security Audit

> Review security of command execution, tool permissions, and API key handling. Use when user mentions "security review", "audit", "check security", "vulnerabilities", or before deploying to production.

- Skill: `aiskillstore/security-audit-2` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add aiskillstore/security-audit-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/aiskillstore/security-audit-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security
- Author: aiskillstore (https://skillmd.com/u/aiskillstore)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/aiskillstore/security-audit-2

---


# Security Audit

## Instructions
1. **Command Execution Review** (`backend/main.py`):
   - Check `run_terminal_command()` for shell injection vulnerabilities
   - Verify timeout is enforced (should be 15 seconds)
   - Look for dangerous command patterns

2. **Tool Permission Review**:
   - Verify Chat mode only allows: `read_file`, `web_search`
   - Check Agent mode tool restrictions
   - Look for permission bypass vulnerabilities

3. **Secrets Management**:
   - Ensure `.env` is in `.gitignore`
   - Check no API keys are hardcoded
   - Verify `python-dotenv` usage for environment variables

4. **WebSocket Security**:
   - Check for authentication on `/ws` endpoint
   - Review message validation
   - Look for injection points in user input

5. **Frontend Security**:
   - Check for XSS in markdown rendering
   - Review image upload handling (base64 encoding)
   - Verify no sensitive data in client-side code

6. Generate report with:
   - Critical issues (immediate action required)
   - Warnings (should fix before production)
   - Recommendations (best practices)

## Examples
- "Run a security audit"
- "Check for vulnerabilities"
- "Review security before deploy"

## Guardrails
- This is a READ-ONLY audit; do not modify files
- Report findings without exploiting vulnerabilities
- Recommend fixes but get user approval before implementing
- Never log or expose discovered secrets