---
name: paper-reading
description: Use when reading academic papers, PDFs, arXiv pages, figures, tables, methods, experiments, ablations, or literature-review notes. Provides a grounded paper-analysis workflow.
---

# Paper Reading

Use this skill when the user asks for paper reading, paper explanation, literature notes, PDF analysis, arXiv analysis, method comparison, experiment interpretation, or research synthesis.

## Reading contract

Never treat a paper as a blob of text. Treat it as a set of claims backed by evidence.

For each major claim, collect:

- where it appears
- what evidence supports it
- what assumption it needs
- what would falsify or weaken it

## Minimal report skeleton

1. Identity
   - title, authors, venue/year if available

2. Question
   - what problem the paper attacks
   - why previous work was insufficient

3. Contribution
   - separate method, empirical result, dataset, theory, and system contributions

4. Method
   - inputs and outputs
   - model or algorithm mechanism
   - training or inference procedure
   - key equations and their roles

5. Evidence
   - datasets
   - baselines
   - metrics
   - main tables
   - ablations
   - figures

6. Limits
   - stated limitations
   - hidden assumptions
   - failure cases
   - missing experiments

7. Takeaway
   - one precise sentence
   - do not exceed what the evidence supports

## Quality checks

- Did you attach evidence to every important claim?
- Did you separate the authors' claim from your interpretation?
- Did you read captions and tables, not only the abstract?
- Did you inspect ablations before praising the method?
- Did you mark uncertainty rather than smoothing it away?

## Output style

Write plainly. Prefer concrete nouns and direct verbs. Avoid generic praise.