cx
Spawn Codex CLI sub-agents non-interactively with fixed model/effort and required flags (web search, skills, yolo). Use when dispatching a single agent per bead via codex exec with a positional prompt string and locked defaults.
When & Why to Use This Skill
The cx Claude skill is a specialized automation utility designed to spawn Codex CLI sub-agents in a non-interactive, headless environment. It streamlines agentic workflows by enforcing fixed high-reasoning models and pre-configured flags such as web search and 'YOLO' mode, bypassing manual approval bottlenecks. This skill is ideal for developers and power users who need to orchestrate complex, multi-step tasks by dispatching autonomous sub-agents to handle specific 'beads' or task units efficiently.
Use Cases
- Automated Task Delegation: Seamlessly dispatch specialized sub-agents to handle individual components of a larger project, enabling parallel processing and autonomous task execution.
- CI/CD Pipeline Integration: Incorporate autonomous agentic reasoning into automated build and deployment scripts where interactive prompts are not possible, utilizing the non-interactive 'codex exec' mode.
- Rapid Technical Research and Implementation: Deploy sub-agents with pre-enabled web search and skill access to perform deep-dive technical investigations and immediate code updates without manual configuration overhead.
- Batch Processing of Agentic Tasks: Use the skill to iterate through a list of tasks (beads), spawning a dedicated agent for each to ensure consistent performance with locked reasoning defaults.
| name | cx |
|---|---|
| description | Spawn Codex CLI sub-agents non-interactively with fixed model/effort and required flags (web search, skills, yolo). Use when dispatching a single agent per bead via codex exec with a positional prompt string and locked defaults. |
cx
Workflow
- Accept a positional prompt string (required).
- Run
$HOME/.codex/skills/cx/scripts/cx-exec.sh "$PROMPT". - Do not override defaults; update the script if defaults change.
Defaults
- Model: gpt-5.2-codex
- Reasoning effort: high (via
-c model_reasoning_effort="high") - Flags:
--search,--enable web_search_request,--enable skills,--yolo - Sandbox/approvals: bypassed (via
--dangerously-bypass-approvals-and-sandbox) - Mode:
codex exec(non-interactive)
Script
$HOME/.codex/skills/cx/scripts/cx-exec.sh
Example
$HOME/.codex/skills/cx/scripts/cx-exec.sh "Work bead 123. Use skill fix. Source the next task per that skill's workflow."