| name | evaluate-pr-review-comments |
| description | **WORKFLOW SKILL** — Evaluates review comments on a GitHub Pull Request, classifying each by validity, category, impact, and risk, then recommends whether to apply, consider, or skip. WHEN: "evaluate PR comments", "review PR feedback", "triage review comments", "evaluate PR reviews", "PR comment evaluation", "should I apply this review comment". INVOKES: GitHub PR read tools, file reading tools. FOR SINGLE OPERATIONS: read individual PR review comments via GitHub MCP tools directly. |
| argument-hint | Optionally provide the PR number to review (e.g., 42) |
Evaluate PR Review Comments
Evaluate review comments on a Pull Request, classifying each for validity and
recommending actions. Produces a structured assessment report with per-comment
analysis and an overall summary.
Prerequisites
- Required MCP tools:
github (pull request read access)
- Required permissions: Read access to the repository codebase
- Enable MCP: Run
/mcp add github if not enabled
MCP Tools Used
| Step | Tool | Method | Purpose |
|---|
| 1 | github | get_pull_request | Load PR details |
| 2 | github | get_pull_request_diff | Retrieve full diff |
| 2 | github | get_pull_request_files | List changed files |
| 2 | github | get_pull_request_reviews | Retrieve submitted reviews |
| 2 | github | get_pull_request_review_comments | Retrieve threaded review comments |
| 2 | github | get_pull_request_comments | Retrieve non-review comments |
CLI Fallback (if MCP unavailable):
gh pr view <number> --json title,body,author
gh pr diff <number>
gh pr view <number> --json files
gh api repos/{owner}/{repo}/pulls/<number>/reviews
gh api repos/{owner}/{repo}/pulls/<number>/comments
Process
Step 1 — Determine the PR
- If a PR number was provided as an argument, use it.
- Otherwise, check the current Git branch and find an open PR for it
using the GitHub repository tool.
- If no PR can be determined, ask the user:
"Which PR number should I review comments for?"
Step 2 — Retrieve PR Context
Use the GitHub Pull Request read tool to gather:
- PR details (method:
get) — title, description, author, intent.
- Diff (method:
get_diff) — full changeset.
- Changed files (method:
get_files) — files modified in the PR.
- Reviews (method:
get_reviews) — submitted reviews and verdicts.
- Review comments (method:
get_review_comments) — threaded comments
with resolution status. Paginate with perPage: 100 using the after
cursor until all comments are retrieved.
- PR comments (method:
get_comments) — non-review discussion comments.
Step 3 — Analyze Each Review Comment
For each comment or thread, perform three sub-steps:
3a. Understand the Comment
- What is the reviewer asking or suggesting?
- Does it reference a specific line, file, or pattern?
- Is it a question, suggestion, nitpick, or blocking concern?
3b. Validate Against the Codebase
- Read the relevant file(s) for full context around the commented code.
- Assess whether the observation is factually correct.
- Check alignment with project coding conventions, patterns, and architecture.
3c. Classify and Assess
For each comment, determine the following attributes:
| Attribute | Values |
|---|
| Validity | Valid / Partially Valid / Invalid |
| Category | Bug, Security, Performance, Style, Readability, Architecture, Nitpick, Question, or Praise |
| Recommendation | Apply / Consider / Skip |
| Difficulty | Trivial / Easy / Moderate / Hard |
| Impact | Breaking / Significant / Minor / Cosmetic |
| Risk | Regression risk if applied: Low / Medium / High |
Step 4 — Generate the Assessment Report
Produce the report using the template in assets/report-template.md. The
report contains four sections:
- Summary — totals, breakdowns by recommendation and category, overall
review quality.
- Detailed Assessment Table — one row per comment with all classification
attributes.
- Detailed Analysis — for each Apply or Consider comment: quote, context,
analysis, recommendation, and implementation notes.
- Comments Recommended to Skip — for each Skip comment: reason for
skipping (e.g., stylistic preference, incorrect observation, already
addressed).
Edge Cases
- No review comments found: Report that no comments exist and stop.
- Resolved threads: Include in the assessment with resolved status noted.
- Ambiguous comments: Flag as ambiguous, provide best interpretation, and
note the uncertainty.
- Large PRs: If the diff is very large, focus on changed files referenced
by review comments rather than reading the entire diff.
Validation
After generating the report, verify:
Guidelines
- Be objective — base assessments on code quality, conventions, and
correctness, not on authorship.
- Consider project context — reference existing codebase patterns when
evaluating suggestions.
- Prioritize safety — always recommend applying security and correctness
concerns.
- Flag ambiguity — if a comment is unclear, note this and provide the best
interpretation.
- Include resolved threads — note resolved status but include in the
assessment for completeness.