Sagarithm Kit is designed to govern and secure the behavior of AI coding agents operating within production codebases. Security is a primary engineering discipline of this framework.
| Version | Supported |
|---|---|
0.x (Pre-release / Foundation) |
Best-effort security patches |
When integrating Sagarithm Kit into software repositories, consider the following security boundaries:
- Prompt Injection & Instruction Override: Canonical skills and policies must be structured to resist instruction hijacking.
- Command & Tool Execution: Agent adapters must not generate configurations that grant unrestricted shell execution or unauthorized environment access.
- Secrets & Credentials: Sagarithm Kit specifications must never contain credentials, API tokens, or sensitive environment variables. Policies explicitly block agents from committing secrets.
- Supply Chain & Dependencies: Policies must require auditing and version-locking before any new third-party dependency is introduced by an agent.
If you discover a security vulnerability within Sagarithm Kit or its generated configurations, please report it responsibly:
- Do not open a public GitHub issue.
- Email your findings directly to
security@sagarithm.com(or create a private GitHub Security Advisory). - Provide a detailed summary, proof of concept, and affected version(s).
- We acknowledge security reports within 48 hours and aim to coordinate a disclosure timeline upon verification.