pr-feedback-classifier_skill

Overview

This skill fetches and classifies pull request review feedback for the current branch and returns deterministic, structured JSON with thread and comment IDs. It isolates review threads and discussion comments, tags each item by actionability and complexity, and groups actionable items into ordered batches for deterministic resolution.

How this skill works

The skill locates the current branch and associated PR, fetches review threads and discussion comments (optionally including resolved threads), and runs a comment classification model on each comment. It determines actionability (actionable vs informational), assigns complexity levels for actionable items, truncates original text to the first 200 characters, and emits JSON that includes thread IDs, comment IDs, batch ordering, and metadata for automation.

When to use it

• Before starting a PR fix pass to know which comments require code changes.
• When preparing an automated agent or CI to address review feedback deterministically.
• To audit open review threads and estimate effort by complexity.
• When triaging large PRs with mixed bot messages and human review comments.
• To generate targeted reply or resolve actions for discussion comments.

Best practices

• Run from the feature branch so the skill can find the correct PR and thread IDs.
• Include resolved threads only when you need full context; omit them for prioritized action lists.
• Use the produced batch ordering to decide which items can be auto-proceeded versus manual review.
• Keep reviewer comments concise so the classification model can correctly detect actionability.
• Verify 'thread_id' and 'comment_id' values before automated resolution commands.

Example use cases

• Generate a prioritized JSON plan for an agent to apply quick single-line fixes automatically.
• Produce a report for maintainers showing cross-cutting and complex items that need design work.
• Triage PRs to count informational messages (CI, bots) vs actionable review requests.
• Feed structured feedback into downstream tools that post replies or resolve threads.