pr-fix-comments_skill

Overview

This skill reads unresolved GitHub PR review comments from a curated list of trusted authors and either addresses, resolves, or flags them for human review. It automates fetching review threads, applying product-principles-guided decisions, making code fixes, replying where needed, and explicitly resolving threads. The skill enforces a strict workflow for traceable task tracking and final PR summarization.

How this skill works

The skill locates the target PR (by argument or current branch), retrieves unresolved review threads via the GitHub GraphQL API, and filters threads to only process those whose first comment author is in the trusted list. For each trusted thread it categorizes the feedback (valid issue, not valid, resolved by product principles, or ambiguous), performs code changes when required, posts explanatory replies when appropriate, and explicitly resolves threads using the GraphQL resolveReviewThread mutation. It tracks progress with TaskCreate/TaskUpdate, runs linting and commits changes, pushes the branch, and posts a top-level PR summary comment.

When to use it

• You want to automatically address routine suggested fixes from project collaborators and trusted bots.
• You need consistent application of documented product principles to subjective review comments.
• You want to reduce manual PR housekeeping (resolving threads, replying to reviewers) for high-trust contributors.
• You need an auditable summary of which comments were fixed, explained, or flagged for humans.

Best practices

• Keep the trusted authors list up to date so only intended accounts are auto-processed.
• Maintain clear, actionable rules in rules/product-principles.md to minimize ambiguous flags.
• Run the skill on feature branches where commits can be safely amended and pushed.
• Ensure CI and linting pass locally; the skill runs dyad:lint and requires a clean commit.
• Verify that replies to 'not valid' comments cite the exact principle number/name when applicable.

Example use cases

• A collaborator leaves multiple style or bug-fix suggestions across files; the skill applies fixes and resolves threads.
• A trusted product reviewer leaves a UX trade-off comment that maps to a product principle; the skill enforces the principle and closes the thread.
• A bot reports minor typing or formatting issues; the skill commits automated corrections and posts the resolution.
• A comment is unclear or requires further discussion; the skill flags it and leaves a clear, principle-cited request for human attention.