调用 OpenAI Codex 获取第二意见,批判性评估后三分类展示给用户
The 'codex-advisor' skill provides users with an automated mechanism to request critical code reviews or second opinions by leveraging OpenAI's Codex MCP (Machine Code Processing). By doing so, it helps software developers improve their code quality and decision-making processes. The skill integrates seamlessly into development workflows, offering unbiased assessments of code, highlighting potential issues like bugs, security risks, and performance concerns, all without the need for manual intervention. This enables teams to make informed decisions while optimizing their coding processes.
The skill operates by first determining the user's intent through contextual analysis, such as file references or git diffs, and then collects relevant context. It proceeds by building a request to the Codex MCP, which is then processed to provide actionable feedback. This feedback is critically evaluated, categorized, and presented to the user for decision-making. It also includes a robust tagging system, enabling users to distinguish between actionable items, discussions, style suggestions, and potential over-engineering. 'Codex-advisor' can be used for continuous code review, ensuring quality at every stage of development.
This skill is primarily targeted at software developers, quality assurance teams, and project managers who need to maintain high-quality code standards. It is particularly useful for teams working on large codebases, where manual code review can become time-consuming, or for developers looking for alternative perspectives on their code quality. By automating the review process and offering clear, actionable insights, this skill supports both individual developers and collaborative team efforts.
The skill infers the user's intent based on the context, such as file references, git diffs, or questions. It automatically identifies whether a code review, change review, second opinion, or overall evaluation is needed.
If Codex fails to provide feedback or encounters an error, the system will display a failure message, including details about the possible causes, such as network issues or configuration problems, and suggest retrying or refining the request.
The skill primarily requires access to the project directory, relevant files or git diffs, and any specific questions or code snippets the user wants reviewed. Additional context may be gathered during the process.
Yes, the skill can work with any codebase as long as the context and necessary files (e.g., source code or git history) are available. It does not depend on any particular programming language.
No, the skill does not automatically implement suggestions. It presents all feedback for the user to decide, with each suggestion tagged for easier decision-making. The final choice remains with the user.
Quick Setup:
.claude/skills/Repository
cuipengfei/prompts