Responsible disclosure policy
We take the security of AgentInterdict seriously. Because this is an enforcement tool whose entire purpose is trust, we particularly value reports that find gaps between the security model we publish and what the code actually does. We appreciate responsible disclosure.
Our commitment
We will treat every report seriously, respond promptly, and work with you toward a coordinated fix and disclosure. We would rather hear about a vulnerability from a researcher than discover it after it has been exploited.
What to report
We welcome reports of vulnerabilities in AgentInterdict itself:
- Defects in the runtime enforcement logic
- Bypasses of origin-bound authority or derivation-amplification controls
- Flaws in memory handling (e.g. credentials persisting when they should be rejected or redacted)
- Weaknesses in tamper detection or the integrity-seal mechanism
- Any discrepancy between the documented security model and observed behaviour
Out of scope
- Third-party dependencies. Issues in libraries AgentInterdict uses should be reported to the respective projects.
- The model itself. Behaviours of the underlying language model are not a vulnerability in this tool.
- Host OS vulnerabilities. The operating system and host environment are outside this project.
- Already-public issues. Issues already disclosed publicly and not yet fixed.
How to report
Report privately, not publicly. The maintainer is reachable through the AgentInterdict GitHub repository:
- Use the repository's Security tab to create a private advisory, or
- Contact the maintainer directly through GitHub with the details below.
Please include:
- A clear description of the vulnerability.
- Steps to reproduce it.
- The impact you believe it has.
- A suggested fix, if you have one.
What we commit to
- Acknowledge your report within 72 hours.
- Triage and assess it within 7 days.
- Work toward a coordinated disclosure with you once a fix is ready.
- Credit you as the reporter (only with your permission).
Please do not
- Test against production Stripe-live infrastructure or other users' data.
- Attempt denial-of-service or destructive testing against live systems.
- Report publicly before we have coordinated.
We intend to respond to good-faith researchers with cooperation, not legal action.