Imported from ouyangyipeng/AutoPaperBot (academic-paper-cheching-skill/SKILL.md). Install upstream with npx skills add ouyangyipeng/AutoPaperBot --skill academic-paper-cheching-skill. Copyright stays with the author.

Academic Paper Checking

Systematically evaluate a completed paper against top-tier conference standards. Produce a structured scoring report with per-dimension scores, issue identification, and actionable suggestions.

Evaluation Methodology

Four-Dimensional Framework

Evaluate the paper across four dimensions, each scored 0-10:

① Problem Criticality (问题关键性) — "Is the problem key?"

• Existence: Is the problem real and observable?
• Impact: Is it a critical bottleneck with significant consequences?
• Gap: Is it genuinely unsolved by existing work?

② Insight Validity (洞察有效性) — "Is the insight valid?"

• Essence: Does it reveal deep causes behind surface phenomena?
• Applicability: Is the viewpoint general, not specific to one case?
• Novelty: Does it challenge common sense or provide new perspective?

③ Design Inevitability (设计必然性) — "Is the design inevitable?"

• Continuity: Is it a natural derivation from the stated insight?
• Trade-off: Is there clear reasoning for why this design over alternatives?

④ Evidence Sufficiency (验证充分性) — "Is evidence sufficient?"

• Completeness: SOTA comparison? Ablation? Sensitivity analysis?
• Explanation: Are results explained (why), not just presented (what)?

Presentation Quality (表达质量)

Additional dimensions covering writing and visual standards:

⑤ Logical Closure (逻辑闭环) — "Does the paper form a closed loop?"

• Does Observation → Insight → Design → Evidence form a complete chain?
• Are there logical gaps or dangling arguments?

⑥ Cognitive Load (认知负荷) — "Is the paper easy to read?"

• Are key conclusions placed early and prominently?
• Is there unnecessary complexity in expression?
• Can figures be understood in 10 seconds?

⑦ Professional Rigor (专业严谨) — "Is the paper academically rigorous?"

• Are claims adequately supported by evidence?
• Is terminology consistent throughout?
• Are extreme/absolute words used carefully?

Checking Procedure

When invoked, follow this exact procedure:

Step 1: Read the Paper Thoroughly

Read every section of the paper. Take notes on:

• What problem is claimed
• What insight is stated
• What design is proposed
• What evidence is presented
• How the narrative flows

Step 2: Dimension-by-Dimension Evaluation

For each of the 7 dimensions, evaluate the paper and assign a score (0-10).

Scoring Guide

Step 3: Detailed Issue Identification

For each dimension, list specific issues found in the paper. Use this format:

[Dimension Name] — Score: X/10

Issues:
1. [Specific issue with location reference]
2. [Specific issue with location reference]
...

Strengths:
1. [Specific strength observed]
...

Step 4: Generate Improvement Suggestions

For each issue, provide actionable suggestions:

Issue: [description]
Suggestion: [specific, actionable fix]
Priority: [Critical/High/Medium/Low]

Step 5: Overall Assessment

Calculate the overall score and provide a final verdict.

Output Format

ALWAYS produce the report in this exact format:

# Paper Quality Assessment Report

**Paper Title:** [title from the paper]
**Evaluation Date:** [date]
**Evaluator:** AI Agent (academic-paper-checking skill)

---

## Overall Score: X/10

| Dimension | Score | Status |
|-----------|-------|--------|
| ① Problem Criticality | X/10 | [Excellent/Good/Adequate/Weak/Critical] |
| ② Insight Validity | X/10 | [Excellent/Good/Adequate/Weak/Critical] |
| ③ Design Inevitability | X/10 | [Excellent/Good/Adequate/Weak/Critical] |
| ④ Evidence Sufficiency | X/10 | [Excellent/Good/Adequate/Weak/Critical] |
| ⑤ Logical Closure | X/10 | [Excellent/Good/Adequate/Weak/Critical] |
| ⑥ Cognitive Load | X/10 | [Excellent/Good/Adequate/Weak/Critical] |
| ⑦ Professional Rigor | X/10 | [Excellent/Good/Adequate/Weak/Critical] |

**Overall Verdict:** [Ready for submission / Needs revision before submission / Major rewrite needed / Not submission-ready]

---

## Detailed Findings

### ① Problem Criticality — Score: X/10

**Existence Check:**
- [Findings on whether problem is real and observable]

**Impact Check:**
- [Findings on whether problem is a critical bottleneck]

**Gap Check:**
- [Findings on whether problem is genuinely unsolved]

**Issues:**
1. [Specific issue — Section X, Paragraph Y]
2. ...

**Strengths:**
1. [Specific strength]
2. ...

**Suggestions:**
1. [Actionable suggestion] — Priority: [level]
2. ...

---

### ② Insight Validity — Score: X/10

[Same sub-structure as above]

---

### ③ Design Inevitability — Score: X/10

[Same sub-structure as above]

---

### ④ Evidence Sufficiency — Score: X/10

[Same sub-structure as above]

---

### ⑤ Logical Closure — Score: X/10

**Loop Check:**
- Observation → Insight: [Connected? Gap?]
- Insight → Design: [Connected? Gap?]
- Design → Evidence: [Connected? Gap?]
- Evidence → Observation: [Does it close the loop?]

**Dangling Arguments:**
- [Any claims not supported? Any evidence not connected to claims?]

[Issues/Strengths/Suggestions as above]

---

### ⑥ Cognitive Load — Score: X/10

**10-Second Rule Check:**
- [Can each figure be understood in 10 seconds?]

**Conclusion Placement:**
- [Are key conclusions placed early?]

**Unnecessary Complexity:**
- [Are there overly complex expressions that could be simplified?]

[Issues/Strengths/Suggestions as above]

---

### ⑦ Professional Rigor — Score: X/10

**Claim Verification:**
- [Are all claims supported by specific evidence?]

**Terminology Consistency:**
- [Are terms used consistently throughout?]

**Absolute Language:**
- [Are extreme words used appropriately?]

[Issues/Strengths/Suggestions as above]

---

## Priority Action Items

List all suggestions sorted by priority:

### Critical (Must fix before submission)
1. [Item]
2. ...

### High (Strongly recommended)
1. [Item]
2. ...

### Medium (Recommended for improvement)
1. [Item]
2. ...

### Low (Optional polish)
1. [Item]
2. ...

---

## Submission Readiness Assessment

**Can a skeptical reviewer understand and believe this paper?**

| Checklist Item | ✅ Pass | ❌ Fail |
|---------------|---------|---------|
| Problem is real, important, clearly defined | | |
| Motivation has data and solid | | |
| Key insight explicitly stated and non-trivial | | |
| Design choices well-justified | | |
| Evaluation is complete | | |
| Results explained, not just presented | | |
| Figures self-explanatory with clear captions | | |
| Claims adequately supported by evidence | | |
| Writing clear, concise, easy to follow | | |
| Paper forms a logical loop | | |
| Each section delivers a clear takeaway | | |
| Terminology and concepts consistent | | |
| Reviewer perspective adopted | | |

**Note:** If any item is unchecked, a reviewer likely will too.

---

*Report generated by academic-paper-checking skill*
*Based on "From Engineering Implementation to Scientific Argument" methodology*

Specific Checks Per Dimension

Problem Criticality Checks

1. Is the problem explicitly stated? Find the sentence that defines the problem. Is it clear, specific, and verifiable?
2. Is there data supporting problem existence? Are there measurements, benchmarks, or studies showing the problem is real?
3. Is the problem impact quantified? "Performance drops by 10×" vs "performance is sometimes slow"
4. Is the gap clearly identified? Does the paper explain what existing work fails to address?
5. Scenario → Problem convergence: Does the paper narrow from a broad scenario to a specific, solvable problem?

Insight Validity Checks

1. Is insight explicitly stated? Can you find a clear sentence articulating the key insight? Or is it buried/implicit?
2. Does insight go beyond surface? Is it "X is slow" (surface) or "X is slow because of I/O-bound nature rather than compute-bound" (deep)?
3. Is insight generalizable? Does it apply beyond the specific system/dataset studied?
4. Does insight explain why existing methods fail? Not just "they don't work" but "they fail because they ignore cross-layer dependencies"
5. Premature solution check: Does motivation present solutions before establishing insight?

Design Inevitability Checks

1. Design-insight connection: Can you trace each design decision back to the stated insight? Is the path logical?
2. Trade-off reasoning: Does the paper explain WHY this design choice over alternatives? Or just describes WHAT was chosen?
3. Motivation-design separation: Is problem statement clearly separated from solution space? Or do they blur together?
4. Design covers all problems: Does every stated problem/challenge have a corresponding design component addressing it?
5. Module-by-module clarity: Can each design module be understood independently?

Evidence Sufficiency Checks

1. SOTA comparison: Is there comparison with state-of-the-art methods? If not, this is a hard flaw.
2. Ablation study: Is there evidence that each design component contributes to the overall result?
3. Sensitivity analysis: Are results shown across varying parameters/conditions?
4. Generality: Are results shown across diverse workloads/datasets/scenarios?
5. Results explained: For each result, is there analysis of WHY (not just WHAT)?
6. Anomaly handling: Are surprising/negative results acknowledged and explained?
7. Claim-evidence mapping: Can each claim be traced to specific experimental evidence?

Logical Closure Checks

1. Full loop: Observation → Insight → Design → Evidence → (back to) Observation
2. No dangling claims: Every claim has supporting evidence
3. No orphan evidence: Every experiment connects back to a design choice or problem
4. Consistent narrative: The story doesn't contradict itself across sections
5. Section takeaways: Each section has a clear conclusion that advances the argument

Cognitive Load Checks

1. 10-second figure rule: Can each figure be roughly understood in 10 seconds?
2. Front-loading: Are key conclusions and qualitative statements placed early in paragraphs/sections?
3. No clause trains: Are sentences reasonably short? No ... which ... which ... which ...
4. No isolated paragraphs: Are all paragraphs substantial and connected?
5. Self-explanatory figures: Can figures + captions alone convey 80% of the message?
6. Unnecessary detail removed: Is implementation detail that doesn't serve the argument removed?

Professional Rigor Checks

1. No overclaim: Are extreme words (best, optimal, most) used only with theoretical backing?
2. Consistent terminology: Same concept = same term throughout
3. Positive framing: Uses "is designed for" rather than "fails to"
4. Acknowledged limitations: Does the paper admit its own limitations?
5. Correct citations: Are referenced works real and correctly cited?
6. Reproducibility: Are experimental setups described sufficiently for reproduction?

Important Notes

• This skill evaluates, not fixes. Point out issues and suggest fixes, but don't rewrite the paper.
• Be specific: reference section numbers, paragraph locations, figure numbers.
• Be constructive: every issue should have an actionable suggestion.
• Be honest: don't inflate scores. A skeptical reviewer would be harsh; mirror that.
• Consider domain: systems papers have different expectations than ML theory papers. Adapt criteria accordingly.
• The four-dimensional framework (Problem/Insight/Design/Evidence) is the backbone; Presentation dimensions (Logic/Cognitive/Rigor) support it.