Use a full PDF, full text or clearly bounded excerpts.
Research reading SkillUpdated 2026.09.29
Evidence-Led Paper Reading
Read research papers with AI while keeping important claims tied to source locations, limitations and the boundary between findings and inference.
For readers who need to assess evidence, not just collect summaries
Keep every consequential judgement connected to the paper
Start with the decision you need to make. The Skill maps the study, separates findings from interpretation and inference, and keeps scope limits visible.
Separate method, findings, limitations and inference.
Decide whether to cite, compare, reproduce, apply cautiously or hold.
Result
Find out what the paper supports before turning it into a summary
The Skill starts with the decision the reader needs to make. It maps the study, traces consequential claims back to the source and keeps conclusions inside the design and source coverage actually examined.
- Paper map
- Research question, claimed gap, design, data, primary outcome, main finding and boundary.
- Evidence table
- Claim, evidence identity, source locator, what it establishes and the caveat that changes interpretation.
- Next action
- Cite, compare, reproduce, apply cautiously or do not use yet.
Start here
Begin with one paper and one real decision
Provide a readable PDF or full text and explain whether you need to understand the argument, verify a claim, inspect the method, compare studies or decide whether the paper is useful enough to cite.
- 1
Set the reading decision
- 2
Declare source coverage
- 3
Map the paper
- 4
Locate consequential claims
- 5
Stress-test the conclusion
- 6
Choose the next action
Evidence contract
Keep four evidence identities separate
A paper-reported finding is not the same as the authors’ interpretation. A reader inference is recorded as an inference, and missing or unreadable information stays unknown. Important claims include a page, section, figure, table or appendix locator whenever the source permits it.
- Paper-reported finding
- A result directly supported by the paper's data or analysis.
- Author interpretation
- The paper's explanation of what that result means.
- Reader inference
- A reasoned conclusion added during the reading.
- Unknown
- Information that the available source does not establish.
Limits
Traceability is not a guarantee of correctness
The method can expose where a conclusion comes from and where it overreaches. It does not replace peer review, specialist judgement or independent replication, and one paper is not enough for a high-stakes medical, legal, safety or financial decision.
Access and version
Download the fixed v1.0.0 Skill
The public repository and versioned ZIP are available under the MIT License. The ZIP SHA-256 is ff98769953bab0a6a1c39247c513ae34ff384878cfe7979f61f94615a3df9b00. The core Skill requires no external service; PDF access depends on the reading tools available in the host agent environment.