AI Codex Workflow

Purpose

Repository-specific guidance for using Codex effectively without adding process overhead.

Default collaboration model

  • Use Codex as navigator/reviewer by default.
  • Keep implementation delegated only for explicit, scoped requests.
  • Keep responses concise, technical, and evidence-based.
  1. Start from docs/Architecture/Overview.md.
  2. Identify target entry points.
  3. Propose or execute minimal scoped changes.
  4. Run relevant verification commands.
  5. Update docs when behavior/operations change.

Skills currently used in this repo

  • mcaf-architecture-overview
  • mcaf-feature-spec
  • mcaf-testing
  • mcaf-formatting
  • small-change-fast-path
  • trace-entrypoints
  • write-tests-first
  • create-plan
  • skill-creator
  • skill-installer
  • worklog-doc-sync

Local environment notes

  • Start Codex from repo root so local AGENTS.md and .codex/skills/ are loaded.
  • Keep user-local Codex config in ~/.codex/config.toml only.
  • Do not commit personal Codex profile/config files.

Practical prompt patterns

  • "Trace entry points for and list exact files to edit."
  • "Write tests first for , no implementation."
  • "Review this diff with P0/P1/P2 findings only."