| name | paper-analysis |
| description | Deep analysis of a single paper: download PDF, extract figures, generate structured 20+ section note, fill LLM placeholders. |
| allowed-tools | Bash |
Paper Analysis Skill
Generate a comprehensive analysis note for a single paper.
Step 1: Download the paper
download_paper(paper_id="2510.24701", title="Paper Title", domain="LLM")
This caches the PDF locally under paper-notes/<title>/.
Step 2: Extract figures
extract_paper_images(paper_id="2510.24701")
Auto-detects the local PDF. Images go to paper-notes/<title>/images/.
Step 3: Generate the analysis note
analyze_paper(paper_json='{"title": "...", "authors": [...], "arxiv_id": "..."}', language="zh")
Returns:
note_path: path to the generated markdown note
quality_check: whether the note has unfilled placeholders
pdf_text: full text extracted from the PDF (if available)
instructions: if placeholders exist, instructions to fill them
Step 4: Fill placeholders
The generated note contains <!-- LLM: describe method --> style placeholders. When instructions is non-null, you MUST fill all placeholders using the pdf_text field:
Filling rules
- Read the note at
note_path.
- For each
<!-- LLM: ... --> placeholder, replace it with substantive content drawn from pdf_text.
- Be specific: cite numbers, dataset names, architecture details, formula descriptions.
- Do NOT leave any placeholder unfilled.
- Write the complete filled note back using the Write tool.
Section guide
| Section | What to write |
|---|
| Core information | One-sentence contribution, key result, paper type |
| Abstract translation | Translate or paraphrase the abstract with domain context |
| Research background | What problem existed, why it matters |
| Research questions | List the explicit or implicit questions the paper addresses |
| Method overview | Architecture, algorithm, pipeline — with specifics |
| Experimental results | Datasets, metrics, comparisons, ablation findings |
| Deep analysis | Strengths, weaknesses, assumptions, limitations |
| Comparison with related work | Position vs. prior art |
| Technical roadmap | Where this fits in the broader research trajectory |
| Future work | Open questions the paper raises |
| Comprehensive evaluation | Your own 1-5 rating with justification |
| My notes | Leave empty for user to fill |
Step 5: Convert to knowledge card (optional)
paper_to_card(paper_json='...', note_path='/path/to/note.md')
Step 6: Link keywords (optional)
link_paper_keywords()
Scans all notes and auto-creates [[wiki-links]] for technical terms.