Route by task type and make each agent review the other

Goal: Use Claude Code and Codex together, routing each task by type
Setup: agents used in source: Claude, Codex, Cursor.
Before you start: A terminal and an account for each tool; Claude Pro plan; ChatGPT login or OpenAI API key for Codex.
Steps:
1. Install and sign in to both tools: The author's video covers install and sign-in (01:05–02:27).
2. Classify each task by type, first match wins: Review, locale text, audits, research, English prose, broken, mechanical edit, feature, script, data.
3. Map types to tool and model: Audits/research/content → Claude Code; bugs/features/scripts/data → Codex; lighter models for lighter jobs.
4. Work in a separate project folder: Away from any script that sends email or writes to an outside service.
5. Lock down permissions: Claude Code in manual mode with an allowlist, push/commit/merge/deploy/delete disallowed; Codex workspace-write or read-only.
6. Fall back to the other tool on usage limits: At most four attempts; never moves to paid credits without approval.
7. Have the other tool review substantial work: The reviewer reports findings and does not edit.
Rules: ask me before anything irreversible (sending, posting, paying, deleting); don’t invent data, results or tools I haven’t given you; tell me what you couldn’t verify.
Based on: Mike Holp — https://www.ai-automation-station.com/guides/claude-code-vs-codex.html (summarised from source; the source has no explicit step list)

Steps are our summary of the source (it has no explicit step list). Includes a “Based on” credit line.