Agent Reviews

What it does

This skill, Agent Reviews, automates the process of reviewing and responding to feedback from pull request (PR) review bots. It operates in a two-phase workflow: first, it automatically fixes existing issues identified by these bots; then, it periodically checks for new comments. The tool provides detailed information about each comment, including its full content, code context, and replies, allowing the agent to accurately assess and address bot findings.

When to use it

• When working on pull requests that have automated review bots providing feedback.
• To streamline the process of addressing and resolving issues flagged by PR review bots.
• To reduce manual effort in reviewing and responding to bot comments, especially when dealing with a large number of reviews.
• In development workflows where quick iteration and efficient bug fixing are important.

Key capabilities

• Identifies Current Pull Request: Automatically detects the current pull request being worked on.
• Fetches Bot Comments: Retrieves all unanswered bot comments, providing full details including comment body, code diffs, and replies.
• Evaluates Findings: Classifies bot findings as True Positive (real bug), False Positive (incorrect flag), or Uncertain (requires user input).
• Automated Fixing: Automatically fixes identified bugs based on the evaluation of bot comments.
• User Interaction: Uses AskUserQuestion to prompt for clarification when uncertain about a finding.

Example prompts

Since this skill operates automatically, there are no direct user prompts. However, the agent might present questions like:

• "This comment flags a potential type mismatch. Are you certain this is intentional?" (When Uncertain)
• "I've identified an issue with missing error handling. Should I apply the suggested fix?" (Before automatically fixing a True Positive)

Tips & gotchas

• The skill relies on gh (GitHub CLI) being installed and configured to access your GitHub account.
• It’s important to understand that the agent's evaluation of bot findings is based on predefined criteria; manual review may still be necessary in complex cases.
• The accuracy of automated fixes depends on the clarity and correctness of the bot comments.