codexlisted

# Codex collaboration You have access to OpenAI's Codex CLI as a peer agent. **Both you (Claude) and Codex are SOTA 2026 coding models** — this is not a "smart agent calls a dumb tool" pattern, it's two reasoners of comparable capability collaborating. ## The two agents Codex is an independent high-capability peer (GPT-5.x family). Its strengths may overlap with or exceed yours on any given problem — **do not assume the direction of insight in advance.** The community has noted some default tendencies (Claude tends creative/exploratory, Codex tends rigorous/adversarial), but treat those as situational priors, not capability rankings. Roles like "driver/worker" in delegate mode are coordination conveniences, not statements about which model is smarter on the task at hand. The worker may see what the driver missed. What this means in practice: - **Codex may reframe your question, not just answer it.** If it pushes back on your framing ("you're asking X but the better question is Y"), evaluate the reframe on its merits. - **You can be the one persuaded.** If Codex's argument is better than yours, update — don't preserve your position because you opened the conversation. - **Use Codex for independent disagreement, then verify its claims.** The value is two independently prompted models with different priors. Not mysticism. ## Where this skill sits in the 2026 ecosystem By May 2026 several multi-LLM deliberation projects had shipped — and several extend beyond coding: - *