Source: https://github.com/aipoch/medical-research-skills

Academic Highlight Generator

Generate journal-ready

Highlights

that can be pasted directly into a submission system. This skill is for academic writing output, not for inventing missing results.

When to Use

• The user wants a

  Highlights

  section for a manuscript submission.
• The source is an English manuscript, abstract, results summary, or extracted full text.
• The paper falls into one of these types: Original Research, Meta-analysis, Review, Case Report, Bioinformatics, Bibliometrics, or Technical Note.
• The user needs a deterministic, concise output with strict bullet-count and length limits.

When Not to Use

• The user asks you to fabricate results, novelty claims, study counts, effect sizes, or conclusions that are not in the source.
• The source text is too short to identify study type or key findings reliably.
• The document is a perspective, commentary, editorial, or otherwise unsuitable for formal submission highlights.
• The user provides a binary

  .doc

  file. This package supports

  .txt

  ,

  .pdf

  , and

  .docx

  ; convert

  .doc

  before continuing.

Required Inputs

Provide one of the following:

• Plain manuscript text, abstract, or structured study summary.
• A supported source file path for

  scripts/extract_text.py

  :

  .txt

  ,

  .pdf

  , or

  .docx

  .

Recommended metadata if available:

• Manuscript type or target journal.
• Core method, main findings, and significance sentence.
• Any wording constraints such as British/American spelling.

Output Contract

Always return:

Highlights
- <bullet 1>
- <bullet 2>
- <bullet 3>
[- <bullet 4>]
[- <bullet 5>]

Hard requirements:

• Exactly

  3-5

  bullets.
• English bullets only unless the user explicitly requests Chinese.
• Maximum

  85

  characters per English bullet.
• Objective third-person tone.
• No first person (

  we

  ,

  our

  ).
• No undefined abbreviations, citation markers, or figure/table references.
• Every bullet must be grounded in source material.

Supported Execution Paths

Path A: Source text already provided

Use the provided text directly. This is the preferred path for speed and determinism.

Path B: Source file needs extraction

Use:

python scripts/extract_text.py <file_path>

Supported formats:

• .txt

• .pdf

• .docx

Unsupported format:

• .doc

  -> ask the user to convert to

  .docx

  or

  .pdf

  first.

Workflow

1. Validate source sufficiency

Before writing anything, confirm the source contains enough signal to identify:

• study type
• method or evidence base
• main finding or conclusion

If not, stop and use the refusal template in

## Fallback and Refusal Contract

.

2. Extract text if needed

If the user provided a file instead of text, run:

python scripts/extract_text.py <file_path>

If extraction fails:

• report the exact failure,
• preserve the original file path in the message,
• do not invent content from the missing file.

3. Detect article type

Use

references/prompts.md

to classify the manuscript into one of:

• Original Research
• Meta-analysis
• Review
• Case Report / Case Series
• Bioinformatics Study
• Perspective / Commentary
• Education / Policy Research
• Bibliometric Analysis
• Short Communication / Technical Note
• Other / Unclear

4. Generate draft highlights

Select the matching generation prompt from

references/prompts.md

.

Coverage priorities by type:

• Original Research: method, main result, mechanism/utility, significance
• Meta-analysis / Review: evidence base, synthesis method, conclusion, gap/future direction
• Case Report: case feature, diagnostic or treatment learning point, follow-up significance
• Bioinformatics: data source, analytic method, marker/pathway/model, biological relevance
• Bibliometrics: database, time span, tools, hotspots/trends, collaboration pattern
• Technical Note: method/device/process optimization, efficiency or usability gain

5. Self-critique and refine

Use the critique and refinement prompts in

references/prompts.md

.

The final output must satisfy all of these checks:

• 3-5

  bullets
• no bullet exceeds the limit
• the bullets are not copied verbatim from the abstract
• the set covers method + finding + value at least once
• no fabricated numbers or study claims

Fallback and Refusal Contract

If the source is unsuitable or insufficient, respond with this structure:

Cannot generate submission-ready Highlights yet.
Reason: <insufficient source / unsupported article type / unsupported file format>
Detected type: <type or Unknown>
Minimum additional input needed:
- <item 1>
- <item 2>

Use this refusal contract when:

• the article type is

  Other / Unclear

  ,
• the text is too short to ground claims,
• the user asks for invention rather than extraction,
• the file format is unsupported.

Deterministic Rules

• Keep the same output header every time:

  Highlights

  .
• Do not switch between sentence fragments and full sentences in one output.
• Prefer one factual claim per bullet.
• If a key value is unavailable, omit that value instead of guessing it.
• If the source supports only three safe bullets, output three rather than padding to five.

Quality Checklist

Before returning the final answer, verify:

• Study type and bullet focus are aligned.
• No unsupported causal overstatement appears.
• No clinical recommendation is implied unless the source itself states one cautiously.
• Each bullet is independently readable.
• The final output can be pasted into a journal submission form without reformatting.