---
title: "skill by agenticluke · skilld"
canonical_url: "https://skilld.dev/gh/agenticluke/literature-review-plus"
meta:
  description: "A clear workflow for reviews of academic, medical, technical, and science sources. It covers search plans, source checks, evidence tables, synthesis, and… From agenticluke/literature-review-plus."
  "og:description": "A clear workflow for reviews of academic, medical, technical, and science sources. It covers search plans, source checks, evidence tables, synthesis, and… From agenticluke/literature-review-plus."
  "og:title": "skill by agenticluke"
  "twitter:description": "A clear workflow for reviews of academic, medical, technical, and science sources. It covers search plans, source checks, evidence tables, synthesis, and… From agenticluke/literature-review-plus."
  "twitter:title": "skill by agenticluke"
---

`

[All skills](https://skilld.dev/skills)

[![agenticluke avatar](https://skilld.dev/_img/avatar?url=https%3A%2F%2Fgithub.com%2Fagenticluke.png%3Fsize%3D96)](https://skilld.dev/gh/agenticluke)

# **/skill**

[@2f51e4d](https://github.com/agenticluke/literature-review-plus/commit/2f51e4de41c97f860a8e9189c09ed58b6d2884a7 "Your agent reads SKILL.md at commit 2f51e4d")

by [agenticluke](https://skilld.dev/gh/agenticluke)· [agenticluke](https://skilld.dev/gh/agenticluke)/ [literature-review-plus](https://skilld.dev/gh/agenticluke/literature-review-plus)

A clear workflow for reviews of academic, medical, technical, and science sources. It covers search plans, source checks, evidence tables, synthesis, and citation checks.

- 1 file
- 9.5 KB
- Updated 2 weeks ago
- [GitHub](https://github.com/agenticluke/literature-review-plus/blob/2f51e4de41c97f860a8e9189c09ed58b6d2884a7/skill/SKILL.md "View SKILL.md on GitHub")

## SKILL.md

9.5 KB

**≈44** tokens always: the name and description. **≈2.4k** when used: this file.

## Literature Review

Use this skill to find, check, compare, and cite research.

Give loud credit to the original authors of all ideas, data, and text. Follow each source license.

### When to Use

Use this skill to:

- Write a narrative, scoping, or systematic review.
- Explain the current state of a research topic.
- Find gaps, conflicts, and open questions.
- Write a source-backed section for a paper or report.
- Compare papers, preprints, patents, reports, or standards.

Do not use it for a quick answer that needs only one known source.

### Pick the Review Type

- **Narrative review**: Gives a broad view of a topic.
- **Scoping review**: Maps ideas, methods, and research gaps.
- **Systematic review**: Uses a fixed plan that another person can repeat.
- **Meta-analysis**: A systematic review that also joins number-based results.

Ask the user which type they need.

If they do not choose:

- Use a scoping review for early study.
- Use a systematic review for public, medical, or safety claims.
- Do not run a meta-analysis unless the studies can be fairly joined.

State the chosen type before the search starts.

### Workflow

#### 1. Define the Question

Turn the request into one clear research question.

For medical work, use PICO:

- **Population**: Who is being studied?
- **Intervention or exposure**: What treatment, test, or risk is studied?
- **Comparator**: What is it compared with?
- **Outcome**: What result is measured?

For technical work, define:

- The system or field.
- The method being tested.
- The baseline used for comparison.
- The measures used to judge results.

Also set the date range, language, place, and source types.

If the question is too broad, split it into smaller questions.

#### 2. Write the Search Plan

Write the plan before collecting sources. Include:

- Databases to search.
- Full search terms for each database.
- Search date.
- Date range.
- Languages.
- Source types.
- Rules for including a source.
- Rules for leaving a source out.

Use at least:

- PubMed for medical and life science work.
- arXiv for computer science, math, physics, and preprints.
- Semantic Scholar or Crossref for broad research search.
- Field sources when needed, such as trial lists, patent sites, standards groups, or official product docs.

Use more than one database for broad claims.

Say when a database, paid paper, or full text cannot be reached. Do not act as if the search was complete.

#### 3. Search and Keep a Log

Record every search so another person can repeat it.

```
| Database | Search date | Exact query | Filters | Results | Saved data |
| --- | --- | --- | --- | ---: | --- |
| PubMed | 2026-05-11 | `("CRISPR"[tiab] OR "Cas9"[tiab]) AND "sickle cell"[tiab]` | 2020 to 2026, English | 86 | PMID list |
| arXiv | 2026-05-11 | `CRISPR sickle cell gene editing` | q-bio, 2020 to 2026 | 9 | BibTeX |
```

Keep raw IDs, links, DOI values, abstracts, and notes apart from the final review.

Do not count a search result page or short search text as evidence.

#### 4. Remove Copies

Remove copies in this order:

1. Same DOI.
2. Same PMID or arXiv ID.
3. Same title.
4. Near-match title with the same first author and year.

Link a preprint and its later paper as two forms of the same study. Use the final paper when it is available. Keep both only when the change matters.

Record how many copies were removed.

#### 5. Check Each Source

Check sources in three steps:

1. Title.
2. Abstract.
3. Full text.

For a systematic review, record one clear reason for each source left out:

- Wrong group.
- Wrong method or treatment.
- Wrong result.
- Not an original study.
- Copy of another record.
- Full text not available.
- Outside the date range.
- Wrong language.
- Paper was withdrawn.

Do not guess from the title when the abstract or full text is needed.

If two reviewers disagree, record the conflict and how it was settled.

#### 6. Check Source Health

Before using a source:

- Look for a withdrawal, correction, or new version.
- Check the journal or official source page.
- Mark preprints as preprints.
- Mark reviews as reviews.
- Mark company-funded work when this is known.
- Note if only an abstract is available.
- Note if the data or code cannot be checked.

Do not use a withdrawn paper as support. It may be named only to explain the record.

#### 7. Extract the Data

Use one row for each study.

```
| Study | Type | People or data | Method | Comparison | Result | Main finding | Limits |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Author, year | Trial, group study, review, or other | Sample or data set | What was done | Baseline or control | What was measured | Main result | Main risks or gaps |
```

For technical papers, also record:

- Data set.
- Test set.
- Measures.
- Baselines.
- Code and data access.
- Steps needed to repeat the work.
- Hardware or cost, when it changes the result.

Keep missing facts blank or mark them as “not reported.” Do not fill gaps with guesses.

#### 8. Judge Study Quality

Check whether each study:

- Fits the research question.
- Uses a fair comparison.
- Has enough people or data.
- Measures the right result.
- Reports missing data.
- Reports harm or failed results.
- Shares enough detail to repeat the work.
- Has funding or author ties that may shape the result.

Use a known quality check tool when the field has one. Name the tool.

Do not turn quality into one score unless the method calls for it.

#### 9. Combine the Evidence

Group findings by theme. Do not write one short report for each paper.

Cover:

- Strongest evidence.
- Results that agree.
- Results that conflict.
- Weak study methods.
- Limits in people or data sets.
- How new and repeatable the work is.
- Real-world meaning.
- Questions that are still open.

Explain likely reasons for conflicts, such as different groups, doses, data sets, measures, or study plans.

Rate each main claim:

- **High confidence**: Strong results repeat across several good sources.
- **Medium confidence**: The result seems sound but has clear limits.
- **Low confidence**: The result is early, weak, or based on one source.

If no good evidence is found, say so. Do not turn “no evidence found” into “proof of no effect.”

#### 10. Run a Meta-Analysis Only When It Fits

Join study numbers only when the studies ask close enough questions and report results that can be compared.

Before joining results:

- Pick the main result and effect measure.
- Check for major study differences.
- Avoid counting the same people twice.
- Explain how missing data was handled.
- Test how key choices change the result.
- Check for signs that failed studies may be missing.

If the studies are too different, use a written synthesis instead.

#### 11. Check Every Citation

Before the final answer:

- Open the source, not just the search page.
- Check the DOI, PMID, arXiv ID, or official link.
- Check the author names, title, and year.
- Make sure the source supports the exact claim.
- Cite the original study when making a result claim.
- Use reviews for context, not as a stand-in for original work.
- Mark preprints and withdrawn papers.
- Check that every citation appears in the source list.
- Check that every source-list item is cited in the text.

Do not cite a paper for a claim it does not make.

### Concrete Example

User request:

> Review whether standing desks reduce back pain in office workers.

Turn it into this question:

> In adult office workers, do standing desks, compared with seated desks, reduce reported back pain?

Example plan:

```
Review type: Scoping review
Databases: PubMed, Crossref, and trial lists
Date range: 2010 to 2026
Languages: English
Include: Adult office workers, standing or sit-stand desks, back-pain result
Exclude: Children, non-office jobs, no back-pain result, opinion pieces
Main search:
("standing desk" OR "sit-stand desk") AND ("back pain" OR "low back pain") AND office
```

Example final claim:

> The evidence suggests that sit-stand desks may reduce back pain for some office workers, but confidence is low because the studies are small and use different measures.

Add citations only after checking that each named study supports this claim.

### Output Template

```
# Literature Review: <Topic>

Prepared: <date>
Review type: <narrative | scoping | systematic | meta-analysis>
Search dates: <dates>
Databases: <list>

## Research Question

## Search Plan

## Include and Exclude Rules

## Search Results

Records found: <number>
Copies removed: <number>
Full texts checked: <number>
Studies included: <number>

## Evidence Summary

## Findings by Theme

## Conflicts in the Evidence

## Gaps and Limits

## Confidence in Main Claims

## Sources

## Search Log
```

### Common Mistakes

- Do not treat search text as evidence.
- Do not mix preprints, reviews, and original studies without labels.
- Do not hide failed, negative, or conflicting results.
- Do not call a review systematic without a saved plan and search log.
- Do not use one database for a broad claim unless the scope says so.
- Do not cite only papers that agree with the expected answer.
- Do not guess facts from a title or abstract.
- Do not hide missing full text or blocked sources.
- Do not count a preprint and its final paper as two separate studies.
- Do not claim that no effect exists just because no good study was found.
- Do not update an old review without showing the new search dates and terms.

Source: [SKILL.md on GitHub](https://github.com/agenticluke/literature-review-plus/blob/2f51e4de41c97f860a8e9189c09ed58b6d2884a7/skill/SKILL.md)

## Third-party checks

No third-party reports yet.

## Provenance

[Signed by skilld at 2f51e4d.](https://github.com/agenticluke/literature-review-plus/commit/2f51e4de41c97f860a8e9189c09ed58b6d2884a7 "2f51e4de41c97f860a8e9189c09ed58b6d2884a7") This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 weeks ago.

Activeupdated 2 weeks ago

## Capability

<dl>

<dt>origin</dt>
<dd>community</dd>

</dl>

## README badge

![README badge for agenticluke/literature-review-plus](https://skilld.dev/b/agenticluke/literature-review-plus?theme=light&label=0)

## Related skills

-
-
-
-
-
-