Until AgentLoop reaches 1.0, security fixes are made on main and released in
the newest version. Older pre-1.0 versions do not receive guaranteed backports.
| Version | Security fixes |
|---|---|
| Latest release | Yes |
main |
Yes, before the next release |
| Older releases | No guaranteed backports |
Do not open a public issue, discussion, or pull request for a suspected vulnerability.
Use GitHub private vulnerability reporting. If that form is unavailable, use the private contact information on the maintainer's GitHub profile to request a secure reporting channel. Do not send exploit details until a private channel is established.
Include as much of the following as you can:
- the affected version, commit, and component;
- the impact and a realistic attack scenario;
- minimal reproduction steps or a proof of concept;
- any known mitigations; and
- whether you have disclosed the issue elsewhere.
Use synthetic traces and placeholder credentials. A report must not include customer prompts, production trace data, API keys, access tokens, or other people's personal information.
The project aims to acknowledge a complete report within three business days and provide an initial assessment within seven business days. These are response targets, not a service-level agreement. The reporter will receive updates when the assessment changes or a fix is ready.
The maintainer and reporter will coordinate disclosure. The project may publish a GitHub security advisory with affected versions, mitigations, and credit after a fix is available. Please allow a reasonable remediation window before public disclosure; 90 days is the usual upper bound unless active exploitation or user safety requires a different timeline.
Reports about the Python package, API server, dashboard, official container, and official integrations are in scope. Vulnerabilities that exist only in a third-party dependency should normally be reported to that dependency first, though AgentLoop still welcomes reports about exploitable dependency use or a missing mitigation.
Good-faith research should avoid privacy violations, service disruption, data destruction, persistence, and access beyond what is needed to demonstrate the problem. Stop testing and report immediately if you encounter real user data. The project will not recommend legal action against research that follows these guidelines and makes a good-faith effort to avoid harm.