Codex tip: let Astra manage while Sol writes the code

Archive @ArchiveExplorerOriginally posted 2026.10.02

Open original on X

Codex tip: once your dot is running, stop letting Astra write the code put your dot on as the manager your dot runs on GPT-6 Astra and has its own cloud computer but every Codex task it starts burns your Codex limits like you started it so the dot plans, watches and decides Codex tasks on your connected computer do the coding: → before a task: which repo, which lane, what "done" means → when a task stalls: retry, escalate, or ask you → before "done": did the tests actually pass? Astra manages. Sol ships Jev engineering is the same move one layer down: inside each Codex task the small forks (which lane, retry or stop, is it really done) go to Jev in under half a second, and Sol only sees the ones that need a thinker one trap: a dot's tasks are orchestrated in the cloud, so hooks from your local config and plugins don't fire. hook-based Jev routers go silent under a dot. skills still load - the full tree > dot on GPT-6 Astra plans, watches, approves > local Codex task on GPT-6.1 Sol, medium > explorer and researcher on GPT-6 Luna > Astra again only on the escalate lane > Jev skills in ~/.codex/skills pick the lane and call retry / escalate / complete > custom rule: "Ask before taking action" on push and merge paste the tree and this prompt into Codex ↓ "Set up this machine for coding work my dot hands off: 1. Run git clone https://github.com/kerpopule/hermes-jev-skills ~/hermes-jev-skills && python3 ~/hermes-jev-skills/install.py --check > Show me what it would install. Run it without --check only after I say go > Then run jev setup-key (I paste the key on the page myself, never in chat) and jev doctor 2. Check ~/.codex/agents and .codex/agents for agents that already fit explorer, worker and researcher > Draft new TOML files only for missing roles > explorer and researcher on gpt-6-luna, worker on gpt-6.1-sol, all with model_reasoning_effort medium > Skip any that pin a different model and list them 3. In ~/.codex/config.toml set model to gpt-6.1-sol and model_reasoning_effort to medium 4. List every hook in my Codex config and plugins my workflow depends on. Under a dot they won't fire. Report, change nothing 5. Add one rule to AGENTS.md: before starting, run jev lane classify --task "<task>"; after each cycle run jev lane step with this repo's tests and a --scope; switch to gpt-6-astra only when step says escalate; never call it done unless step returns complete Show me every change as a diff first. No edits until I say go." then tell your dot ↓ "For coding in <repo>, don't code on your cloud computer. Create a local Codex task on my connected computer and follow the repo's AGENTS.md. Report back the lane Jev picked, the test result and the diff. Bring me a decision only if a task escalates twice or touches security. Confirm." ↳ https://learn.chatgpt.com/docs/dots/tasks-and-memory

This post quotes another post: x.com/archiveexplorer/status/2103590048230695408