Paper Reviewer

Overview

Generate comprehensive, evidence-based reviews of academic manuscripts following top-tier venue standards. The skill enforces rigorous evidence anchoring, maintains objectivity and constructive tone, and produces reviews without numerical scores or accept/reject decisions.

When to Use This Skill

Activate when the user requests:

•Peer review of an academic paper or manuscript
•Structured analysis of research contributions
•Evaluation of scientific work for publication venues
•Feedback on manuscript strengths and weaknesses

Review Workflow

Step 1: Input Processing

Accept the manuscript in any format (PDF, plain text, markdown, OCR output).

Read the entire manuscript carefully, noting:

•Section structure (Introduction, Methods, Results, Discussion, etc.)
•All figures and tables with their numbers
•All equations with their numbers
•Page numbers for reference
•The references/bibliography section

Step 2: Review Structure

Produce a review with EXACTLY these six sections in order (no additions, no omissions):

•Synopsis of the paper
•Summary of Review
•Strengths
•Weaknesses
•Suggestions for Improvement
•References

Step 3: Write Each Section

Synopsis of the paper (≤150 words)

•Neutrally restate: the problem, the proposed method, core contributions, and main results
•Use objective language only
•Avoid subjective judgments or decision-like language
•Do not preview strengths or weaknesses

Summary of Review (3-5 sentences)

•Provide a balanced overview of both positives and concerns
•
After EACH reason or claim, add an evidence anchor in parentheses
- •Examples: (See Table 2), (Sec. 4.1), (Eq. (5)), (Fig. 3, p. 7)
•If a claim lacks manuscript evidence, write: (No direct evidence found in the manuscript.)

Strengths (3-6 bullet points)

Focus on:

•Novelty and originality
•Technical soundness
•Experimental rigor (datasets, metrics, baselines)
•Clarity of presentation
•Potential impact

CRITICAL: Add evidence anchors to EVERY bullet point. Reference specific figures, tables, equations, sections, or pages.

Weaknesses (3-8 bullet points)

Focus on verifiable issues:

•Relation to closest prior work (missing comparisons, insufficient differentiation)
•Experimental breadth (limited datasets, missing metrics, insufficient ablations)
•Statistical rigor (no confidence intervals, no significance tests)
•Reproducibility gaps (missing hyperparameters, no code availability)
•Theoretical limitations (unstated assumptions, unexplored failure modes)

CRITICAL: Add evidence anchors to EVERY bullet point. When evidence is missing, explicitly state the gap (e.g., No evidence found in Sec. 4; missing from Methods.).

Suggestions for Improvement (4-8 recommendations)

Provide concrete, actionable recommendations:

•Add specific ablation studies
•Unify baseline settings and tuning budgets
•Report mean ± std/CI across multiple runs
•Include additional metrics (e.g., reliability diagrams, calibration plots)
•Release code and random seeds
•Expand related work discussion to cover specific papers
•Add failure case analysis

Link each suggestion to 1-2 corresponding weaknesses to make it verifiable and actionable.

References

•List ONLY items explicitly cited within this review AND appearing in the manuscript's reference list
•Use concise format: [Author et al., Year] or the manuscript's numbering style
•If no citations are needed or the reference list is unavailable, write: None

Evidence-First Principle

Every claim must be anchored to manuscript evidence.

Good examples:

•"The method demonstrates strong performance on ImageNet (Table 3, p. 8)."
•"The ablation study isolates the contribution of each component (Sec. 5.2, Table 5)."
•"Hyperparameter settings are not reported (No evidence found in Methods section or appendices)."

Bad examples:

•"The method shows good results." (No anchor)
•"The paper is well-written." (Subjective, no anchor)
•"This approach is novel compared to [external work not in manuscript]." (External citation)

Critical Constraints

Mandatory constraints:

•Use ONLY the six section headings listed above
•Do NOT include scores, ratings, confidence levels, or accept/reject decisions
•Do NOT guess authors, institutions, or affiliations
•Do NOT cite external sources unless they appear in the manuscript's reference list
•Do NOT make speculative claims

Tone and style:

•Objective and constructive
•Polite and professional
•Focus on improving the work, not criticizing the authors
•Use imperative/infinitive form for suggestions ("Add an ablation study" not "You should add")

Length:

•Target 400-600 words total
•Adjust as needed for manuscript complexity
•Synopsis: ≤150 words
•Other sections: balanced allocation

Output Format

Produce plain text output using markdown formatting:

•Use ## for section headings
•Use - for bullet points
•Use bold for emphasis sparingly
•Maintain consistent formatting throughout

Resources

references/

For detailed examples and additional guidance:

•review_template.md - Annotated section examples with real review snippets
•evidence_anchoring.md - Comprehensive guide to evidence citation patterns