Security & Governance¶
Code Scalpel is built around a core principle: you can't govern AI agents with suggestions. Security and governance are enforced at the infrastructure level, not the policy level.
Architecture: Local-First, Zero Exfiltration¶
Your code never leaves your machine. Code Scalpel runs as a local MCP server — all AST parsing, dependency tracking, and security analysis happens on your hardware. No telemetry. No cloud calls. No data exfiltration.
What We Detect¶
Code Scalpel's taint-based security analysis tracks user-controlled values from source to sink:
| Category | Examples |
|---|---|
| SQL Injection | sqlite3.execute(f"SELECT * FROM {user_input}") |
| XSS | innerHTML = user_value |
| Command Injection | os.system(cmd), subprocess.call(user_cmd) |
| Path Traversal | open(f"/data/{user_path}") |
| Insecure Deserialization | pickle.loads(data) |
| Hardcoded Secrets | API keys, passwords in source code |
What We Don't Detect¶
We document limitations honestly rather than overstating capabilities:
- Runtime behavior — static analysis only; dynamically constructed payloads may be missed
- Business logic flaws — we detect code patterns, not semantic intent
- Encrypted or obfuscated code — AST parsing requires readable source
Policy Integrity¶
Enterprise tier includes cryptographic verification of policy manifests via HMAC-SHA256. Policies cannot be tampered with silently — the verify_policy_integrity tool checks integrity before every policy evaluation.
Fail-Closed Behavior¶
When the policy engine encounters an error or ambiguity, it defaults to deny. Agents cannot bypass governance by triggering exceptions.
Vulnerability Reporting¶
Found a security issue in Code Scalpel itself? Please report it responsibly:
- Email: security@codescalpel.dev
- GitHub Security Advisories: Report a vulnerability
We aim to acknowledge reports within 48 hours and patch critical issues within 7 days.
Full Security & Governance Reference¶
For the complete Governance & Audit documentation, including compliance evidence generation and enterprise deployment security, see the Security Analysis Guide and the Enterprise Deployment Guide.