rubric-writer_skill

Overview

This skill writes a rubric-based peer review report (output/REVIEW.md) grounded in extracted claims and evidence gaps. It produces a concise referee report that evaluates novelty, soundness, clarity, and impact, and gives actionable, evidence-backed fixes rather than generic suggestions.

How this skill works

The skill reads output/CLAIMS.md and output/MISSING_EVIDENCE.md (and optionally output/NOVELTY_MATRIX.md or DECISIONS.md). It summarizes contributions, evaluates each rubric axis with direct references to extracted claims and missing evidence, and produces major and minor actionable items plus a final recommendation. All critiques are traceable to the provided artifacts.

When to use it

• Final stage of a peer-review pipeline when output/CLAIMS.md and output/MISSING_EVIDENCE.md exist
• When you need a referee-style REVIEW.md for submission or internal review
• When you require concrete, reproducibility-focused feedback tied to extracted claims
• When you want an evidence-first referee report that separates major vs minor issues

Best practices

• Ensure output/CLAIMS.md and output/MISSING_EVIDENCE.md are complete before running the skill
• Provide output/NOVELTY_MATRIX.md if you want a more detailed novelty evaluation
• Keep the extracted claims concise and numbered so critiques can cite them directly
• Use the DECISIONS.md file if a specific reviewer format or decision label is required
• Prefer precise missing-evidence entries (which baseline/dataset/metric) to enable actionable fixes

Example use cases

• Generate a referee report for an internal pre-submission check based on extracted claims
• Produce a reproducibility audit stating exactly which protocol/baseline details are missing
• Convert a claims+gaps artifact into a reviewer-style decision letter for program committees
• Create an author-facing checklist tied to major issues that must be addressed before resubmission