use-codex
Use Codex
If you are Codex, you do not need this skill — stop reading and proceed however you otherwise would. Delegating to yourself gains nothing; this skill exists for other agents delegating scoped work to the Codex CLI (codex exec, non-interactive). It runs on a separate token budget (ChatGPT plan / OpenAI API) with token-economical models, and provides an independent second perspective.
Should this go to Codex?
Delegate when the task is scoped and the answer/diff is all that matters:
- Repo exploration & Q&A ("where is X handled?", "trace this flow")
- Implementation where the architecture is already decided and unknowns reduced
- Mechanical edits at scale (renames, migration sweeps)
- Second-opinion / adversarial review of a plan or diff
- A bug stuck after 2+ failed attempts (different model, different angle)
Keep in Claude: ambiguous/underspecified work, architecture & decomposition, long multi-turn efforts needing session memory, user-facing synthesis. Don't delegate one-liners answerable from context already in hand (each Codex call has ~30k tokens of fixed overhead on its side).
Full rationale, evidence, and cost/limits: references/task-fit.md. If the decision is "don't delegate", stop here — no need to read further.