메뉴
Runs automated code reviews using configurable reviewer agents. Use during or after implementation to check code quality, security, and best practices.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Executes an implementation plan by writing code. Use when ready to implement a plan that has been brainstormed, planned, and approved.
Creates concrete implementation plans from brainstormed ideas. Use when ready to turn an idea into actionable steps — after px-brainstorm, before px-implement.
Fetches Figma designs via MCP and implements them as React components. Use when asked to build, implement, or code a UI from a Figma link or design.
Automates iOS and Android simulators/emulators for mobile app development and testing. Use when the user needs to interact with a mobile simulator, test a mobile app, tap buttons, swipe, type text, take screenshots, or automate any mobile device interaction.
Facilitates idea exploration through conversation. Use when starting a new project, feature, or exploring solutions — before any planning or implementation.
Reviews completed work to document learnings, mistakes, and patterns for future cycles. Use after implementation is done or when reflecting on past work.
| name | px-review |
| description | Runs automated code reviews using configurable reviewer agents. Use during or after implementation to check code quality, security, and best practices. |
| argument-hint | file paths or directories to review, or 'staged' for staged git changes |
Run automated code reviews by discovering and launching reviewer sub-agents in parallel. Each reviewer focuses on a specific concern and reports findings in a consistent format. Findings are presented to the user — nothing is auto-fixed.
If $ARGUMENTS is provided, use it as the scope:
staged → review staged git changes (git diff --cached)If no arguments, ask the user what to review.
Scan ../agents/reviewers/ for all reviewer agent definitions. Each .md file in that directory is a reviewer to run.
If the directory is empty or doesn't exist, inform the user and stop.
Selecting reviewers: By default, run only the core reviewers: security, code-quality, and simplicity. If the user specifies additional reviewers (or all), run those too. This keeps the default cost low while still catching the most important issues. Present the list of selected reviewers to the user before launching them.
Launch all discovered reviewers as parallel sub-agents using Task. Each reviewer receives:
git diff output so reviewers focus on what changed. Include enough surrounding context lines (git diff -U8) for reviewers to understand the change, but no more.If a reviewer genuinely needs broader file context (e.g., to check an import at the top of a file), it can read the file itself — don't pre-load it.
Collect results from all reviewers and present a unified summary to the user, organized by severity:
For each finding, include:
If no findings, report a clean review.
After presenting findings, ask the user which (if any) they want to fix. Only make changes the user explicitly approves.
Each reviewer in ../agents/reviewers/ must follow the output format defined in ../reviewer-output-format.md.