chatgpt-team is mostly instructions and local bootstrap helpers, but mistakes can still expose credentials or encourage unsafe permissions.

Use GitHub's private security-reporting mechanism for this repository when available for issues involving:

• credential or secret exposure;
• bootstrap behavior that can overwrite or expose user state;
• instructions that materially weaken sandbox/approval boundaries;
• a reproducible path that causes authenticated browser state or sensitive evidence to be committed;
• another issue where public disclosure before a fix would create unnecessary risk.

Ordinary documentation errors, stale product claims, and non-sensitive workflow bugs can use normal issues.

This project does not provide a security boundary by itself. Agent instructions are behavioral guidance; real enforcement comes from operating-system permissions, Codex sandbox/approval controls, Git/GitHub permissions, and the security controls of the project being worked on.

Never include real credentials, private tokens, cookies, authenticated browser storage, or private application data in a security report unless the reporting channel explicitly supports that sensitivity and the minimum necessary evidence cannot be provided safely otherwise.