Skip to content

Security: phamhungptithcm/SatsunicAIAgentKit

Security

SECURITY.md

Security Policy

Reporting A Vulnerability

Please report security concerns privately to the maintainers instead of opening a public issue. Include the affected version, reproduction steps, impact, and any suggested fix.

If a public contact address has not been configured yet, create one before public launch and add it here.

Security Design

AI Agent Kit bootstrap and its default governed workflows are local-only.

Bootstrap must not:

  • stage, commit, push, branch, merge, rebase, or tag changes
  • create or update remote pull requests, merge requests, tickets, or deployments
  • access production systems
  • print, write, or collect secrets
  • modify application source code during bootstrap

The single optional remote-ticket exception is product github-sync --apply. It is denied unless the caller supplies the current exact issue-plan approval hash plus a short-lived Ed25519 action from a repository-trusted MEMBER with the product.github.write capability and an operator or team-lead role. The action is bound to the repository identity, product ID, operation, target GitHub repository, and plan hash; its nonce is durably consumed before any remote call. A human repository owner must provision the public trust policy before delegation and keep its private key outside agent-visible context. Preview remains read-only, and all other bootstrap, Git, release, deployment, and production mutations remain outside the CLI's authority.

Changes that weaken these constraints should be treated as security-sensitive.

There aren't any published security advisories