{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreiekn5k54hoo5p3fh7c7cqh5fnc7fotnqmlchzhkzi6mpg2tiefwsq",
"uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3mmms4bdpwbh2"
},
"path": "/t/model-aware-task-delegation-for-subagents/1381671#post_2",
"publishedAt": "2026-05-24T19:26:06.000Z",
"site": "https://community.openai.com",
"tags": [
"Codex Subagents",
"Subagent model and reasoning guidance",
"Codex Configuration Reference",
"Configuration Precedence"
],
"textContent": "Hi and welcome back!\n\nInteresting idea! I believe this is partially doable already and requires some custom settings.\n\nThe official Codex docs do not describe automatic parsing of inline task metadata like this:\n\n\n [model: gpt-5.4-mini | reasoning: medium]\n\n\nWhat they do support today is configuring custom subagent types with `model` and `model_reasoning_effort`.\n\nYou can create project-scoped custom agents here:\n\n\n .codex/agents/\n\n\nOr personal custom agents here:\n\n\n ~/.codex/agents/\n\n\nEach custom agent TOML file must include `name`, `description`, and `developer_instructions`. It can also include normal Codex config keys such as `model`, `model_reasoning_effort`, and `sandbox_mode`.\n\nExample:\n\n\n # .codex/agents/scaffolder.toml\n name = \"scaffolder\"\n description = \"Fast implementation agent for boilerplate, project setup, and straightforward file creation.\"\n model = \"gpt-5.4-mini\"\n model_reasoning_effort = \"medium\"\n\n developer_instructions = \"\"\"\n Implement scoped, straightforward tasks quickly.\n Keep changes small and follow existing project conventions.\n \"\"\"\n\n\n\n # .codex/agents/architect.toml\n name = \"architect\"\n description = \"Deep reasoning agent for architecture, state machines, and critical logic.\"\n model = \"gpt-5.5\"\n model_reasoning_effort = \"high\"\n\n developer_instructions = \"\"\"\n Handle ambiguous or high-impact design work.\n Trace tradeoffs, edge cases, and integration risks before recommending changes.\n \"\"\"\n\n\nThen ask Codex to use those agent names explicitly:\n\n\n Spawn scaffolder for T001 and architect for T020. Use each task's notes as the worker prompt.\n\n\nUseful global subagent settings can go in `.codex/config.toml` or `~/.codex/config.toml`:\n\n\n [agents]\n max_threads = 6\n max_depth = 1\n\n\nThe docs say omitted custom-agent fields inherit from the parent session, so leaving out `model` or `model_reasoning_effort` gives you the fallback behavior you want.\n\nHere are the links to the relevant docs:\n\n * Codex Subagents\n * Subagent model and reasoning guidance\n * Codex Configuration Reference\n * Configuration Precedence\n\n",
"title": "Model-aware task delegation for subagents"
}