Codex 代理树：Sol 写码、Astra 把关、Jev 管小决策

elune @elune0x原推发布于 2026.10.03

在 X 上打开原推

Codex tip: once GPT-6.1 Sol is your main model, you don't need Astra on every turn. Keep Astra available as an architect agent. GPT-6.1 Sol continues writing the code. Bring in Astra at three points: → before a plan: is this the right approach? → when the same error keeps returning: am I looking in the wrong place? → before saying "done": what did I miss? Astra reviews. Sol ships. Jev engineering follows the same idea one layer down: simple decisions (which file, which tool, retry or stop) go to Jev in under half a second. The larger models only handle the decisions that need deeper thought. - the full tree > GPT-6.1 Sol on high runs the main session > explorer reads the code on Luna > worker edits and runs tests on Sol > researcher checks the docs on Luna > all three on medium > Astra stays on call as the architect > auto_review checks every approval Paste this tree and prompt into Codex ↓ "Rebuild my Codex setup around this tree: 1. Check ~/.codex/agents and .codex/agents for existing agents that fit the explorer, worker and researcher roles. > Draft new TOML files only for roles that are missing > Set explorer and researcher to gpt-6-luna, worker to gpt-6.1-sol, all with model_reasoning_effort medium > Add an architect agent on gpt-6-astra with model_reasoning_effort high, focused only on reviewing plans, repeated errors and completed work > Skip any agents that pin a different model and list them 2. In ~/.codex/config.toml, set model to gpt-6.1-sol, model_reasoning_effort to high and approvals_reviewer to auto_review 3. Look for anything that could override these settings (active profiles, flags in my shell aliases, agents.default_subagent_model). Report what you find without changing it 4. Add one rule to AGENTS.md: spawn the architect before a large plan, when an error repeats, and before calling a long task done Show me every proposed change as a diff first. Don't edit anything until I say go." ↳ http://developers.openai.com/codex/subagents

这条是引用着另一条写的：「Dots 不只是更聪明的聊天：代理在提示之间继续干活」