Preferences

This configures how Codex behaves around this project's written artifacts — not the project's business/product content itself, and not conversational behavior. It doesn't touch code, master docs, or .agents/skills/. Safe to re-run anytime to change an answer.

Always project-scoped — this isn't a question about how you want Codex configured across every project on your machine, just this one. The answers live in docs/PROJECT_PREFERENCES.md, versioned alongside the rest of the project's docs, so the whole team gets the same behavior, not just whoever ran this command.

1. Check current state

Before asking anything, read docs/PROJECT_PREFERENCES.md if it exists.

Report what's already set before asking about it. Only ask about what's missing, or what the human explicitly says they want to revisit — never re-run the full interview from scratch just because the skill was invoked again.

2. Ask

Documentation language

"What language should docs/ and .codex/work/ artifacts be written in — context.md, architecture.md, plan.md, business-context files, ADRs, and so on?"

Be explicit this is not about conversation: Codex already mirrors whatever language the human writes in, session to session, with no configuration needed. This setting only controls the language of files that get committed to the repo.

3. Write

• docs/PROJECT_PREFERENCES.md: create it if it doesn't exist, or update the relevant section if it does (don't touch sections the human didn't just answer). Keep it short and plain:

  # Project Preferences
  
  ## Documentation language
  Write `docs/` and `.codex/work/` artifacts in: <language>. Conversational
  responses aren't affected — Codex already mirrors whatever language the
  human writes in.

4. Confirm

## Preferences set
- Documentation language: [English (default) / <language>]

## Where these live
- docs/PROJECT_PREFERENCES.md — the record every `warm-up` skill invocation reads

Re-run this skill (meta-preferences) anytime to change an answer.