Only the latest minor release of agentic-architectures receives security updates.

Please do NOT open a public GitHub issue for security vulnerabilities.

Instead, email fareedhassankhan12@gmail.com with:

• A clear description of the vulnerability
• Steps to reproduce (proof-of-concept code is ideal)
• The affected version(s)
• Your name + GitHub handle (if you'd like to be credited in the fix)

You'll get an acknowledgement within 3 business days. A patched release is typically published within 14 days of confirmation, depending on severity.

Specific things we treat as security issues:

• Prompt-injection bypasses of safety gates — particularly in BrowserAgent, ComputerUse, DryRun, SWEAgent, ConstitutionalAI. If you can make the deterministic-picker safety check approve an action it shouldn't, that's a security issue.
• Sandbox escapes — SWEAgent's _safe_path() rejecting ../ traversal, BrowserAgent's _check_safety() rejecting blocked domains, Voyager's subprocess isolation.
• Secret leakage — anywhere in the library that might log, persist, or send credentials (API keys, vector-store contents) to an unintended destination.
• Dependency vulnerabilities — if you find a known CVE in one of our pinned deps and we haven't bumped yet.

Out of scope:

• LLM-output quality issues (hallucinations, mis-routing) — those are bugs, not security issues. Use the bug report template.
• Issues in upstream packages we depend on — file those upstream and tag us if it affects this repo.

We follow standard 90-day coordinated disclosure. If you need more time we'll discuss; if we need more time to patch we'll let you know.

Thanks for keeping the community safe.