--- name: codex-cli description: Use Codex CLI as a subagent. Lets the main agent prompt Codex CLI from the terminal in headless mode with the `codex` command, and send the goal, the decisions and the scope with the task. Use when the user asks to delegate work to Codex CLI, when a task needs a second independent agent, or when a plan needs a fresh perspective. allowed-tools: - Bash - Read - Write - Grep - Glob --- # Codex CLI (subagent / task delegation) ## What is it This skill teaches your agent to use an agent from another vendor as a subagent. Almost every coding agent has a headless mode. Headless mode runs the agent from the terminal with one command and returns the answer to stdout. This skill gives the command pattern for one such agent, and the protocol to transfer enough context with the task. You stay the main agent. You keep the plan, the decisions and the conversation with the user. The subagent does one bounded task and reports back. ## When to delegate - **The user asks for it.** The user names another agent, or asks for a second opinion. - **Large or multi-step work.** The task covers several files, phases or checkpoints. - **Code review with a fresh mind.** Another agent reviews the work with a different lens and no memory of the decisions that produced it. - **Planning that needs different ideas.** A second agent proposes options that you did not consider. - **An internal council.** Several subagents answer the same question, and you compare the answers. - **Deep research.** The other CLI has web search, web extraction or browser automation. - **Background investigation.** A long task runs in the background while you continue in the foreground. ## When not to delegate - **Small or single-step tasks.** One or two edits, or a short explanation, are faster in this thread. - **Tight feedback loops.** The user wants quick back-and-forth in one conversation. - **Secrets or policy-sensitive work.** Do not pipe credentials into a subagent. Redact first. - **Context you already hold.** Repeating the whole plan in a prompt adds no value. - **Low return on investment.** If writing the handoff costs more than doing the edit, do the edit. - **Work that needs high precision and full judgement.** A subagent without your context will break a task that depends on a detail only this session knows. ## Delegation and context transfer protocol An isolated subagent saves tokens, but it splits the story. The subagent does not see this conversation. It has no memory of the goal you agreed with the user, the decisions you already made, or the paths you agreed not to touch. A poor handoff causes misread tasks, wrong assumptions about the stack or the style, and wasted edits. Treat every call as **stateless delegation**: the subagent keeps nothing between calls, so each command must carry its own state. A good handoff happens when you keep these practices in mind: - **Shared state handoff.** The prompt carries the original goal, the decisions already made, and the explicit scope, to bridge the subagent's missing session history. - **Contextual preservation.** The high-volume work stays in the isolated process, so the bulk never enters this conversation and no "split story" forms from missing history. - **Result reconciliation.** What comes back is folded into this thread afterwards, so there is a single source of truth. Write the prompt as a transfer of shared state, not as a title. Include all six fields: | Include | Why | |---|---| | **Goal** | The same objective as the user, not only the immediate micro-task. | | **Decisions** | Framework, patterns, naming, "use X not Y" — anything that would otherwise be guessed wrong. | | **Scope** | The absolute paths and modules to touch, and the areas to leave alone. A relative path may not resolve where you expect. | | **Constraints** | Performance, accessibility, compatibility, review gates, "no new dependencies". | | **Verification** | The exact command the subagent must run and pass before it returns. | | **Output** | For example: "report only, no edits", "apply edits with a minimal diff", "list the files changed". | Prefer sequential delegations with explicit carry-over. Parallel runs diverge unless every run gets the same briefing. **Run the subagent in the mode that does the work without asking for approval.** Headless mode has nobody to answer a permission prompt. Any mode that stops to ask will stall, or return an empty answer with a success exit code. Each CLI names this mode differently — take the flag from the vendor card below. **Never call a subagent in plan mode.** Many CLIs have a plan or read-only mode. That mode makes the other agent write a plan *for itself*, which is not what you asked for: you want its work or its answer, and you keep the planning. Plan mode also waits for someone to approve that plan, so the run comes back with nothing. If you want no file changes, keep the auto-approving mode and write "report only, no edits" in the **Output** field. Do not use plan mode to make a run read-only. After the subagent returns: - **Report to the user.** Give a short summary. Do not paste long logs unless the user asks. - **Reconcile the context.** Record the decisions it made, the files it changed, and the open risks, so this session stays the single source of truth. ## CLI failure modes Every vendor has its own terminal rules. A headless run fails quietly more often than it fails loudly. - **Wrap the call in an external timeout.** A headless CLI can stall before its own timeout starts. - **Exit code 0 is not success.** If stdout is empty, treat the run as failed and read stderr. A permission the CLI could not ask for, and a prompt that never arrived, both look like success. - **Never carry a flag habit from one CLI to another.** The same short flag means different things in different tools. In OpenCode, `-p` is `--password`: passing a prompt to it empties the message and the run waits on stdin forever. Confirm every flag against the CLI's own `--help`. - **An unrecognized-flag error means this skill is stale, not that the task is impossible.** Read `--help`, continue with the flags that exist, and tell the user which line in this file is wrong. ## Vendor card ### Binary and prompt form - **Binary:** `codex` - **Prompt form:** **positional**, under the `exec` subcommand (alias `codex e`). `codex exec "prompt"` runs non-interactively and exits when the task completes; the prompt string is a plain argument to `exec`, not a flag's value. A bare `codex "prompt"` (no `exec`) opens the interactive TUI and never returns. Stdin handoff also works: `cat instructions.txt | codex exec -`. ### Headless and output flags | Need | Flag | |---|---| | Run once and exit | `exec "prompt"` (alias `e`) | | NDJSON event stream instead of formatted text | `--json` | | Write the final assistant message to a file | `--output-last-message`, `-o ` | | Layer a config profile | `--profile`, `-p ` | | Skip loading the standard config file | `--ignore-user-config` | ### Approvals and permissions - **Sandbox policy** — `--sandbox`, `-s ` — bounds what model-generated shell commands can touch. - **Approval policy** — `--ask-for-approval`, `-a ` — the CLI's own docs page highlights `on-request` and `never`; the source additionally defines `on-failure` and `untrusted` as accepted values. `--ask-for-approval never` suppresses prompts but does **not** by itself grant network access — that's a separate `sandbox_workspace_write.network_access` config setting. - **Full bypass** — **`--dangerously-bypass-approvals-and-sandbox`** (alias **`--yolo`**) — sets the sandbox to `danger-full-access` and the approval policy to `never` together. This is the flag to use for unattended delegation; only run it in an already-hardened environment (container/VM/CI). - **`codex exec` defaults to `AskForApproval::Never`** even without any flag — headless mode does not wait for approval by default. - **What happens with nobody to answer:** the official docs state plainly that when an approval is needed but unavailable in non-interactive mode, **"Codex exits with an error."** It does not stall. ### Models and how to list them No dedicated `exec`-time flag to print the model catalog is confirmed; the docs reference `codex debug models` for inspecting available models, but this is not documented in the same depth as the `--model` flag itself, so verify it against the installed binary. Select the model for a run with `--model`, `-m `; reasoning effort/tier is a config-level concern, not part of the model string. Interactively, `/model` lets you change the model and effort mid-session. If the user names a model, use it. Otherwise check `codex debug models` (or the docs below) and [Artificial Analysis](https://artificialanalysis.ai/) for current identifiers and pricing before picking one — Codex model names and retirements change often enough that a pinned one can fail outright. ### Command pattern ```bash codex exec "GOAL: [goal] | DECISIONS: [decisions] | SCOPE: [paths] | CONSTRAINTS: [constraints] | VERIFICATION: [test_command] | OUTPUT: [format]" \ --dangerously-bypass-approvals-and-sandbox --model 2>&1 ``` ### Prompt examples - **Implement:** `codex exec "GOAL: [goal] | DECISIONS: [decisions] | SCOPE: [paths] | CONSTRAINTS: [constraints] | VERIFICATION: [test_command] | OUTPUT: [format]" --yolo` - **Logic check:** `codex exec "GOAL: Analyze logic for edge cases and race conditions | SCOPE: [paths] | VERIFICATION: [check_command] | OUTPUT: detailed report" --model --sandbox read-only --ask-for-approval never` - **Audit:** `codex exec "GOAL: Security audit of the authentication layer | SCOPE: [paths] | CONSTRAINTS: report only, no edits | OUTPUT: audit report" --sandbox read-only --ask-for-approval never` ### Docs and reference - Flags, sandbox/approval values, delegation checklist: [reference.md](reference.md) - Vendor documentation: and - **Checked against the official OpenAI documentation on 2026-09-19. Not run against an installed binary — confirm with `codex --help` before trusting a flag.** Note on `--full-auto`: an earlier version of this card recommended it. It is now **deprecated** — it still works for one release cycle and prints a warning, but it is scheduled for hard removal, so a command relying on it will start failing. The bypass flag is `--dangerously-bypass-approvals-and-sandbox` (alias `--yolo`).