lingzhi227/agent-research-skills

self-review

Automatically review an academic paper using the NeurIPS review form with three reviewer personas, ensemble scoring, and reflection refinement.

ソースを見る

使い方

  1. インストールコマンドをコピーし、自分の環境へ導入する前にソースを確認します。
  2. 必要なツール、認証情報、対応環境を原文で確認します。
  3. 入力と成果物を指定し、出力を資料と照合します。
npx skills add https://github.com/lingzhi227/agent-research-skills --skill self-review

開始前の確認

  • 導入にはモデル契約、API枠、ファイルへのアクセス許可は含まれません。
  • スター数や導入数は活動の指標であり、独立した品質・安全性試験ではありません。
リポジトリの原文

見出し、例、コード、表、リンク、参照画像を含む原文を表示しています。

Self-Review

Review an academic paper using a structured review form with multiple reviewer personas.

Input

  • $ARGUMENTS — Path to PDF file or .tex file

Scripts

Extract text from PDF

bash
python ~/.claude/skills/self-review/scripts/extract_pdf_text.py paper.pdf --output paper_text.txt
python ~/.claude/skills/self-review/scripts/extract_pdf_text.py paper.pdf --format markdown

Tries pymupdf4llm (best) → pymupdf → pypdf. Install: pip install pymupdf4llm pymupdf pypdf

Parse PDF into structured sections

bash
python ~/.claude/skills/self-review/scripts/parse_pdf_sections.py \
  --pdf paper.pdf --output sections.json

Extracts title (via font size), section headings, and section text. Requires: pip install pymupdf Key flags: --format text, --verbose

Workflow

Step 1: Load Paper

  • If PDF: use extract_pdf_text.py to extract text
  • If .tex: read the LaTeX source directly

Step 2: Three-Persona Review

Run three independent reviews using different personas (from references/review-form.md):

  1. Harsh but fair reviewer: Expects good experiments that lead to insights
  2. Harsh and critical reviewer: Looking for impactful ideas in the field
  3. Open-minded reviewer: Looking for novel ideas not proposed before

For each persona, generate a review following the NeurIPS review JSON format in references/review-form.md.

Step 3: Reflection Refinement (up to 3 rounds per reviewer)

After each review, apply the reflection prompt: re-evaluate accuracy and soundness, refine if needed. Stop when "I am done".

Step 4: Aggregate

  • Combine all three reviews
  • Average numerical scores (round to nearest integer)
  • Synthesize a meta-review finding consensus
  • Weight scores using AgentLaboratory weights: Overall (1.0), Contribution (0.4), Presentation (0.2), others (0.1 each)

Step 5: Actionable Report

Output format:

## Review Summary
- **Overall Score**: X/10 (Weighted: Y/10)
- **Decision**: Accept / Reject
- **Confidence**: Z/5

## Strengths (consensus across reviewers)
1. ...
2. ...

## Weaknesses (consensus across reviewers)
1. ...
2. ...

## Questions for Authors
1. ...

## Specific Suggestions for Improvement
1. [Section X, Page Y]: ...
2. [Section Z, Page W]: ...

## Score Breakdown
| Dimension | R1 | R2 | R3 | Avg |
|-----------|----|----|-----|-----|
| Overall | ... | ... | ... | ... |
| Contribution | ... | ... | ... | ... |
| ... | ... | ... | ... | ... |

References

  • NeurIPS review form, scoring weights, personas, reflection prompts: ~/.claude/skills/self-review/references/review-form.md
  • PDF text extraction: ~/.claude/skills/self-review/scripts/extract_pdf_text.py

Missing Sections Check

You MUST verify that all required sections are present: Abstract, Introduction, Methods/Approach, Experiments/Results, Discussion/Conclusion. Reduce scores if any are missing.

Related Skills

タスクに関連する項目

同じリポジトリから

関連する Skills

すべての Skills
lingzhi227
コミュニティ

backward-traceability

Make every number in the final PDF traceable to the exact code line that produced it. Uses \hypertarget/\hyperlink LaTeX commands and \num{formula} evaluated at compile time. Use for reproducibility and data integrity verification.

1時間の導入数
2
GitHub Stars
451
更新日
2月27日
lingzhi227
コミュニティ

experiment-code

Write ML experiment code with iterative improvement. Generate training/evaluation pipelines, debug errors, and optimize results through code reflection. Use when implementing experiments for a research paper.

1時間の導入数
1
GitHub Stars
420
更新日
2月27日
lingzhi227
コミュニティ

paper-assembly

Orchestrate the full paper pipeline end-to-end. Manage state propagation between phases (literature → plan → code → experiments → figures → tables → writing → review), support checkpointing and resumption. Use for assembling a complete paper from components.

1時間の導入数
1
GitHub Stars
420
更新日
2月27日
lingzhi227
コミュニティ

literature-review

Conduct comprehensive literature reviews using multi-perspective dialogue simulation. Generate diverse expert personas, conduct grounded Q&A conversations, and synthesize findings into structured knowledge. Use when starting a new research project or writing a survey section.

1時間の導入数
1
GitHub Stars
414
更新日
2月27日