Codex 多模型分工：Sol 规划、Luna 干活、Astra 监督
把 GPT-6 Astra 挂成只读顾问，只在定方案前、错误反复出现时、宣布完成前开口；作者说 Jev 工程是同一招下沉一层，无需思考的分叉交给 Jev 半秒内决定。
Codex tip: once GPT-6.1 Sol is your main model, stop leaving GPT-6 Astra sitting idle and stop burning Sol tokens on tasks GPT-6 Luna can swarm put it on call as a dedicated supervisor keep it on standby in .codex/agents/advisor.toml GPT-6.1 Sol plans and ships the code GPT-6 Luna swarms the routine work at medium effort GPT-6 Astra reads the full session, every tool call included, and only speaks up at three points: → before a plan: is this the right approach? → when the same error comes back: am I digging in the wrong place? → before "done": what did I miss? GPT-6 Astra reviews. GPT-6 Luna executes. GPT-6.1 Sol ships Jev engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) go to Jev in under half a second, and the big model only sees the ones that split Plan on high. Delegate on medium. Keep Astra on call. - the full tree > GPT-6.1 Sol on high runs the main session > explorer reads the code > worker edits and runs tests > researcher pulls the docs > all three on GPT-6 Luna at medium effort > GPT-6 Astra on call as the supervisor paste the tree and this prompt into Codex CLI ↓ "Rebuild my Codex setup around this tree: 1. Check ~/.codex/agents and .codex/agents for subagents that already fit explorer, worker and researcher. > Draft new ones only for missing roles > Give each model = 'gpt-6-luna', model_reasoning_effort = 'medium' > Create .codex/agents/advisor.toml with model = 'gpt-6-astra', sandbox_mode = 'read-only' 2. In ~/.codex/config.toml, set: [agents] default_subagent_model = 'gpt-6-luna' default_subagent_reasoning_effort = 'medium' max_concurrent_threads_per_session = 8 3. Set main session model to 'gpt-6.1-sol' with model_reasoning_effort = 'high'. Check for environment variables that disable multi-agent tools (agents.enabled) or override reasoning effort. Report them, change nothing 4. Add one rule to AGENTS.md: consult the advisor agent before approving a large plan, when an error repeats, and before marking a long task done Show me every change as a diff first. No edits until I say go." ↳ http://learn.chatgpt.com/docs/agent-configuration/subagents
这条是引用着另一条写的：x.com/mirku21/status/2098836008468992471