Working habits
Settings can’t cover everything, so these habits cover the rest:
- Read untrusted content first: instruction-like text in an issue, pull request, email or web page is the attack.
- Keep production credentials out of your environment: an agent can’t misuse a token it can’t see.
- Never run
--yoloor full access on your own machine, and never across a home directory. - Use one workspace per session, and git worktrees for parallel or background work.
- Treat auto-review as help, not a boundary: OpenAI says it “can still make mistakes”, so keep a person approving sensitive work.
- Put project rules in
AGENTS.md: they shape the code Codex writes, but only the profile and rules enforce anything.