core-codex-delegator_skill

Overview

This skill routes code-generation and scaffolding tasks to the OpenAI Codex CLI for rapid prototyping and artifact creation. It decides when to delegate based on task type, existing files, and safety checks, and produces safe, reviewable Codex commands. The skill always flags Codex outputs for manual review.

How this skill works

It validates the request, checks Codex CLI availability and authentication, and consults a simple delegation matrix to choose between Codex and the local assistant. For eligible tasks it returns a ready-to-run codex exec command plus a short rationale and a review_required flag. If pre-checks fail or the task is unsuitable, it explains why the local assistant should handle it.

When to use it

• Requesting new boilerplate, scaffolds, or project initialization from scratch
• Generating multi-file templates or repetitive code patterns (CRUD, data models)
• Creating test suites from natural language specifications
• Rapid prototyping from pseudocode, diagrams, or clear NL specs
• Large-scale repetitive code generation across files

Best practices

• Ensure task_description is specific and actionable before delegating
• Confirm task_type is correctly set to generation for delegation
• Verify Codex CLI is installed (which codex) and user is authenticated
• Keep review_required = true and perform manual security and correctness checks on generated code
• Avoid delegating tasks with security, business, or architectural context

Example use cases

• Generate an Express.js + TypeScript REST API scaffold with CRUD endpoints
• Create unit and integration test suites from an API specification
• Scaffold a CLI tool with cross-platform entry points and packaging files
• Produce repetitive data model and CRUD boilerplate across multiple services
• Translate pseudocode or a sequence diagram into runnable module stubs