safety-model.md

Safety Model

Codex Learning Loop is conservative by default.

Safety rules:

• Work only inside ~/.codex during maintenance.
• Do not scan the whole home folder.
• Do not scan project repositories.
• Do not process binary or huge files.
• Do not process files that appear to contain secrets.
• Do not enable automation unless requested.
• Write reports and logs for every run.
• Preserve user learning files during uninstall unless --purge is used.

The helper detects obvious credential-like strings, but users should still avoid writing secrets into learning notes.