codex-cli

The codex-cli skill provides a sophisticated orchestration layer for developers using the OpenAI Codex CLI within Claude Code. It is designed to act as a bridge between high-level logic and low-level code execution, focusing on intelligent prompt engineering, context injection, and resource-efficient multi-agent workflows. By standardizing how tasks are delegated, this tool minimizes unnecessary token consumption and ensures that the AI executes complex coding requirements with high precision and logical consistency.

• Advanced Context Injection: Automatically collects local code snippets, compiler errors, file paths, and dependency relationships to prime the Codex model before execution, bypassing initial exploration phases.

• Persistent Session Management: Enables intelligent session reuse by maintaining state across related coding tasks, allowing the model to recall previous analysis and implementation history effectively.

• Parallel Task Orchestration: Safely manages multi-process execution for independent sub-tasks, such as concurrent performance audits, parallel security scanning, or batch refactoring across distinct modules.

• Structured Prompt Engineering: Utilizes a formal prompt formula (Action + Scope + Requirements + Format + Constraints) to ensure consistent, machine-readable output in formats like JSON or Markdown.

• Workflow Automation: Supports complex multi-stage pipelines where initial analysis informs subsequent fix and verification cycles, reducing manual intervention in the development loop.

• Intended for developers working in complex repositories who need to automate repetitive coding, debugging, and analysis tasks via command-line interfaces.

• Typical inputs include local file paths, compiler error logs, and specific refactoring requirements; outputs consist of optimized code blocks, structured JSON analysis reports, or fixed file diffs.

• Practical constraints: Ensure proper grouping of parallel tasks (e.g., read-only operations are highly parallelizable, while write-intensive operations on the same files must remain serial).

• Always leverage the resume functionality for multi-round interactions to maintain contextual continuity and reduce the cognitive load on the agent.