This skill enables Codex (or GPT-powered agents) to delegate scoped subtasks to a local Claude Code instance. GPT reasons much more accurately but has no eye for design and is so smart it forgets humans need step-by-step pedagogical explanations; Claude is less reliable and less intelligent, yet produces polished UI, accessible writing, and human-friendly documentation. This skill combines both.
Codex investigates, reasons, plans, codes. Claude Code executes presentation-layer work:
| Delegated to Claude | Why |
|---|---|
| Front-end UI | Natural aesthetics |
| Documentation | Human-friendly writing |
| Conclusion rewriting | Turns concise findings into readable docs |
| Demos / prototypes | Visually appealing output |
Codex retains ownership of investigation, complex reasoning, and accuracy review.
| Command | Purpose |
|---|---|
launch |
Start a scoped task |
status |
View task status and logs |
resume |
Continue a task with follow-up instructions |
MIT
