Interactive Review Skill

This skill opens an interactive web UI where users can review content with checkboxes and comments.

How It Works

1. Determine the content source:
   • If user specifies a file path: Use the Read tool to get the file contents
   • If user provides content directly: Use that content as-is
   • Otherwise: Collect the most recent relevant content from the conversation
2. Call mcp__interactive_review__start_review with the content
3. A browser window opens automatically with the review UI
4. User reviews each item:
   • Check/uncheck to approve/reject
   • Add optional comments
5. User clicks Submit
6. Process the feedback and respond accordingly

Content Sources (Priority Order)

1. Explicit file path: User says “review /path/to/file.md” or “이 파일 리뷰해줘: README.md”
   • Read the file using Read tool and use its contents
2. Direct content: User provides or references specific content to review
   • Use the provided content directly
3. Conversation context: Extract relevant content from recent conversation
   • Plans, documents, code, etc. that were recently discussed

Usage

When the user wants to review content:

# If file path is specified, read it first:
Read({ "file_path": "/path/to/file.md" })

# Then start the review:
mcp__interactive_review__start_review({
  "content": "<content from file or conversation>",
  "title": "<descriptive title>"
})

Processing Results

The tool returns a JSON with review items. Handle each item based on:

Example Flow

User: “Review this implementation plan”

1. Extract the plan content from recent output
2. Call start_review with the content
3. Wait for user feedback (tool blocks until submit)
4. Present summary of feedback
5. Ask if user wants you to proceed with approved items or revise rejected items

Response Template

After receiving feedback:

## Review Summary

**Approved**: X items
**Needs revision**: Y items

### Items requiring changes:
- [Item]: [User's comment]

Would you like me to:
1. Proceed with approved items
2. Revise the rejected items based on feedback
3. Both - revise then proceed