|| Version | Supported | ||---------|-----------| || 0.7.x | ✅ Active development | || 0.6.x | ✅ Supported | || < 0.6 | ❌ Not supported |

This project uses GitHub's private vulnerability reporting feature (preferred). To report a vulnerability:

1. Go to https://github.com/iknowkungfubar/autoresearch-stack/security/advisories
2. Click "Report a vulnerability"
3. Describe the issue, including steps to reproduce

Alternatively, email turin@autoresearch.io with:

• Subject: [SECURITY] Vulnerability in autoresearch-stack
• Description of the vulnerability
• Steps to reproduce
• Affected versions

This is a local research tool with no API endpoints, no user accounts, and no stored PII. It runs entirely on the user's machine and connects only to external LLM APIs that the user explicitly configures.

• API keys are read from environment variables only — never from config files
• The config.to_dict() method automatically masks API keys in output
• The CI pipeline scans for hardcoded secrets on every commit

The sandbox module (sandbox.py) uses AST-based validation to prevent dangerous code execution:

• Blocks imports of os, sys, subprocess, socket, ctypes, builtins
• Blocks calls to eval(), exec(), __import__(), compile(), open(), breakpoint(), getattr()
• Blocks attribute access chains on blocked modules (e.g., os.environ.get)
• String-matching bypasses (e.g., import os, importos) are caught by AST parsing

• All dependencies are audited with pip-audit in CI
• Runtime dependencies are minimal (numpy, pyyaml, requests)

We appreciate responsible disclosure. Contributors who report valid security issues will be credited here (with permission).