Your Workflow

Step 1: Fetch and Display PR Comments

Execute the following bash script to fetch and display PR comments:

bash

#!/bin/bash
set -euo pipefail

# jq query to group comments by thread
JQ_QUERY='
# Group by thread ID (use own ID if in_reply_to_id is null)
def thread_id: if .in_reply_to_id then .in_reply_to_id else .id end;

# Group by thread
group_by(thread_id) |
map({
  thread_id: .[0] | thread_id,
  file: .[0].path,
  line: .[0].line,
  position: .[0].position,
  comments: map({
    id: .id,
    user: .user.login,
    body: .body,
    created_at: .created_at,
    is_reply: (.in_reply_to_id != null)
  }) | sort_by(.created_at)
}) |
sort_by(.file, .line)
'

# Fetch PR comments and format with jq
gh api "repos/:owner/:repo/pulls/$(gh pr view --json number -q .number)/comments" | jq -r "$JQ_QUERY" | jq -r '
.[] |
"📁 \(.file):\(.line) (position: \(.position))",
"🧵 スレッドID: \(.thread_id)",
(.comments[] |
  if .is_reply then
    "  └─ 💬 \(.user): \(.body)"
  else
    "  🟢 \(.user): \(.body)"
  end
),
""
'

Step 2: Create Todo List

After displaying comments, create a todo list with the following structure:

•One todo item for each unique file location with comments
•Mark all items as "pending" initially
•Use descriptive names like "Address reviews in [file_path]:[line]"

Step 3: Analyze and Address Each Comment

For each todo item:

•Read the file: Use the Read tool to examine the code at the specified line
•Understand the context: Read surrounding code to understand the implementation
•Analyze the review comment: Determine what change is being requested
•Propose a solution: Based on the comment, suggest specific code changes
•Ask for confirmation: Show the user what you plan to change and ask if they approve
•Implement the fix: Once approved, use the Edit tool to make the changes
•Mark todo as completed: Update the todo list to track progress
•Move to next comment: Continue until all comments are addressed

Step 4: Summary and Next Steps

After addressing all comments:

•Summarize what was changed
•Suggest running tests if applicable
•Recommend the user to reply to review threads on GitHub
•Optionally offer to commit the changes

Output Format (Step 1)

The output will show:

•📁 File path and line number
•🧵 Thread ID
•🟢 Original comment (thread starter)
•💬 Reply comments (indented)

Example Workflow

code

[Fetching PR comments...]

📁 backend/src/api/routes/capsules.py:45 (position: 123)
🧵 スレッドID: 1234567890
  🟢 reviewer1: This function should handle edge cases better
  └─ 💬 author: Good point, I'll add validation

📁 frontend/src/components/Header.tsx:12 (position: 456)
🧵 スレッドID: 1234567891
  🟢 reviewer2: Consider using useMemo here for performance

---

[Creating todo list...]
✅ Todo list created with 2 items

[Starting to address comments...]

📍 Working on: backend/src/api/routes/capsules.py:45
[Reading file...]
[Analyzing comment: "This function should handle edge cases better"]

現在のコード:
```python
def create_capsule(capsule_data):
    return db.save(capsule_data)

レビューコメント: "This function should handle edge cases better"

提案される変更:

•capsule_data のバリデーションを追加
•エラーハンドリングを実装

この変更を実装してよろしいですか？

[User approves]

[Implementing changes...] ✅ Changes implemented ✅ Todo item completed

[Moving to next comment...]

code


## Prerequisites

- The current branch must have an associated pull request
- GitHub CLI (`gh`) must be authenticated
- `jq` must be installed

## Important Notes

- Always read the file before making changes to understand the full context
- Ask for user confirmation before implementing significant changes
- Track progress using the TodoWrite tool
- If a comment is unclear, ask the user for clarification
- Consider the impact of changes on related code and tests
- Maintain the existing code style and conventions

## Error Handling

- If there are no PR comments, inform the user and exit gracefully
- If a file cannot be read, skip that comment and note it in the summary
- If GitHub CLI fails, display the error and suggest troubleshooting steps