name: codex-cli-ff
description: This skill should be used when the user asks to "run codex autonomously", "fire-and-forget codex", "codex-cli-ff", "execute codex in background", or wants to run OpenAI Codex CLI without interactive prompts. Autonomous Codex execution with aggressive defaults via forked subagent.
argument-hint: [--model ] [--effort ] [--sandbox ]
disable-model-invocation: true
context: fork
agent: codex-runner
allowed-tools: Bash(codex *)
Codex CLI Fire-and-Forget
Environment Check
- Codex version: !
codex --version 2>/dev/null || echo "CODEX_NOT_INSTALLED"
Task
Execute the following Codex CLI task autonomously with complete output capture.
Argument Parsing
Parse $ARGUMENTS to extract optional flags and the task description:
Optional flags (override defaults if present):
--model <MODEL> — Override default model (see references/models.md for available models)--effort <LEVEL> — Override reasoning effort (low, medium, high, xhigh)--sandbox <MODE> — Override sandbox mode (read-only, workspace-write, danger-full-access)
Defaults (used when flags are not specified):
- Model:
gpt-5.3-codex
- Effort:
xhigh
- Sandbox:
workspace-write
- Auto:
--full-auto (always enabled)
Everything after the flags is the task description.
Command Assembly
Assemble the codex command following this pattern:
codex exec \
-m <model> \
--config model_reasoning_effort="<effort>" \
--sandbox <sandbox> \
--full-auto \
"<task description>"
Execution
- If the environment check above shows
CODEX_NOT_INSTALLED, stop immediately and report the error
- Assemble the command from parsed arguments and defaults
- Execute the command via Bash
- Capture and return the complete stdout and stderr — do not summarize or truncate
Input
$ARGUMENTS
Additional Resources
For model catalog, sandbox modes, and CLI flag reference, consult:
references/models.md — Complete model table, effort levels, sandbox modes, and command examples
