SkillVaultskills Browse all 1,000+ skills

Productivity · Version 1.3.0 · Reviewed 2026-08-02

Review Comment Classifier

Turn engineering context into a reliable artifact for review thread triage and comment intent classification with evidence, explicit trade-offs, and a verification plan.

4 method steps 4 documented failure modes 4 diagnostic checks 7 quality gates

Classifies review comments by mechanism, severity, action, evidence, and whether the current revision already resolves them. It grounds the decision in review threads, cited lines, revision history, repository rules, and the current diff and explicitly prevents keyword classification losing the requested outcome or treating an outdated line reference as an unresolved defect.

₹199 one-time

Get this skill archive

Install in your AI coding tool

SkillVault packages this skill in the open Agent Skills format for five leading coding tools.

What this skill helps you do

  • Review thread triage
  • Comment intent classification
  • Stale finding detection

How Review Comment Classifier works

You provide

Repository access, entry points, and the target change

It inspects

Real execution path for review thread triage, not naming

It decides

A comment intent classification explanation with safe change seams

You verify

Claims confirmed against code paths or a test

What it checks first

Review Comment Classifier classifies review comments by mechanism, severity, action, evidence, and whether the current revision already resolves them. It grounds the decision in review threads, cited lines, revision history, repository rules, and the current diff and explicitly prevents keyword classification losing the requested outcome or treating an outdated line reference as an unresolved defect. Use it when the work involves Review thread triage, Comment intent classification, Stale finding detection.

  1. The entry points and the data flow between them, which is the fastest way to build an accurate mental model.
  2. Where behavior is actually decided, rather than where it appears to be configured.
  3. Which parts change frequently, since those carry the most current knowledge and the most risk.
  4. The seams where a change can be made safely without a wide blast radius.

Failure modes it recognizes

  • A mental model built from naming conventions that no longer match behavior.
  • Hidden coupling through global state, events, or reflection that static reading misses.
  • Dead code that appears authoritative and misleads the reader.
  • Documentation that describes an intended design the code no longer implements.

Answers it will reject

  • Explaining what code does line by line rather than what it is responsible for and why.
  • Trusting comments and documentation over the executing code path.
  • Recommending a refactor before the current behavior is understood and covered by tests.

Decision rules it applies

  • Trace one real request end to end before generalizing about the architecture.
  • Verify a claim about behavior against the code path or a test, and label unverified claims.
  • Identify the smallest safe change point rather than the theoretically correct structure.

Evidence it asks for

  • Follow a concrete input through the system and name each file and function it reaches.
  • Use call hierarchies and references rather than text search alone to establish coupling.
  • Confirm behavior with an executable test before changing it.

The method inside

  1. Map the artifact, actors, boundaries, and invariants relevant to review thread triage.
  2. Trace concrete failure or abuse paths for comment intent classification; do not report checklist items without a mechanism.
  3. Prioritize stale finding detection findings by impact, likelihood, confidence, and cost of correction.
  4. Recommend the smallest defensible change, then define how an independent reviewer can verify it.

Deliverables

  • Review thread triage assessment
  • Comment intent classification decision and action plan
  • Stale finding detection verification checklist

Evidence requirements

  • Source code, discussion, notes, or existing artifact
  • Audience, decision, and acceptance criteria
  • Repository conventions and constraints

Quality gates

  • Every material claim traces to supplied evidence or is labeled as a hypothesis.
  • The response follows the declared deliverable contract.
  • No execution, access, measurement, or verification is invented.
  • Secrets and personal data are redacted rather than repeated.
  • The user receives a concrete independent verification step.
  • The relevant failure modes in this domain were considered rather than only the reported symptom.
  • No listed anti-pattern was recommended as a solution.

Example task

Input

Apply the review comment classifier to our current review thread triage work. We need a concrete decision, bounded changes, and evidence that the result is correct.

Expected output

Start with review threads, cited lines, revision history, repository rules, and the current diff. The highest-risk failure is keyword classification losing the requested outcome or treating an outdated line reference as an unresolved defect. Classify from the full thread and current behavior, separating questions, blockers, suggestions, and accepted trade-offs. Verify the result by sampling each class against human labels and checking that automated resolution never closes ambiguous threads.

Boundaries and compatibility

Ideal for

  • Review thread triage: produce a decision or artifact grounded in supplied evidence.
  • Comment intent classification: produce a decision or artifact grounded in supplied evidence.
  • Stale finding detection: produce a decision or artifact grounded in supplied evidence.

Out of scope

  • Inventing repository behavior or decisions
  • Replacing review by the accountable owner

Agent compatibility

  • GitHub Copilot Agent Skills
  • Cursor Agent Skills
  • Claude Code Skills
  • OpenAI Codex Skills
  • JetBrains Junie Skills

Tool policy: Advisory by default. No tools are assumed. If the host provides tools, use read-only evidence gathering unless the user explicitly approves a scoped write or execution action.