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

Biomedical Research Outline Generator

This is an Academic Writing skill for producing manuscript-grade biomedical outlines with deterministic headings and clear section logic.

Optional Validation Shortcut

If you want to confirm the local helper path exists before use:

python scripts/validate_skill.py --check

This helper is optional. The primary workflow is still direct outline generation from user input.

When to Use

Use this skill in biomedical contexts when the user wants one of these three outputs:

• Type I: Review Outline Input pattern: research directions, disease area, pathway keywords, or field description
• Type II: Discussion Outline Input pattern: results or discussion paragraphs containing observations, models, markers, statistics, or mechanisms
• Type III: Thesis / Proposal Outline Input pattern: Title: plus background, methods expectations, cohort notes, timeline, or validation requirements

When Not to Use

• The request is clearly non-biomedical.
• The user wants fabricated results, unsupported claims, or invented citations.
• The request is for a complete manuscript draft rather than an outline.

Required Inputs

• A biomedical topic, result paragraph, or proposal title with enough context to determine one of the three supported types.

Recommended:

• disease model or population
• key molecules, pathways, or interventions
• study aim or proposal objective
• any special formatting or institutional requirements

Type Recognition Rules

Type I: Review Outline

Use this type when the input is mostly:

• topic keywords
• field description
• research direction

Signals:

• no explicit Title:
• no detailed result paragraph
• no proposal/timeline language

Type II: Discussion Outline

Use this type when the input contains:

• observations or findings
• effect sizes or p-values
• model systems such as cell, mouse, cohort, or patient data
• mechanistic hints and limitations

Type III: Thesis / Proposal Outline

Use this type when the input contains:

• Title:
• proposal framing
• background and aims
• plan, feasibility, timeline, or expected outcomes

If the request is off-domain, stop and use the refusal contract in ## Fallback and Refusal Contract.

Output Contract

Type I: Review Outline

Must include:

• title
• abstract
• keywords
• introduction
• 4-6 major chapters
• 2-3 subchapters under each major chapter where appropriate
• conclusion / outlook

Type II: Discussion Outline

Must include:

• summary of key findings
• interpretation blocks
• literature integration
• limitations
• future directions
• conclusion

Type III: Thesis / Proposal Outline

Must include:

• project review / background
• purpose and significance
• research plan
• feasibility / risk or ethics considerations
• innovation
• timeline
• expected outcomes

Formatting Rules

• Markdown headings only: #, ##, ###
• Stable numeric hierarchy
• concise, field-appropriate wording
• no placeholder text like to be added
• no fabricated results or unsupported claims

Workflow

1. Validate domain and sufficiency

Confirm that:

• the topic is biomedical
• there is enough information to classify the request
• the requested output is an outline, not a full manuscript

2. Detect type

Assign Type I, II, or III using the rules above.

3. Build the section skeleton

Use the output contract for the detected type and keep section order deterministic.

4. Enrich with academic logic

For each section, add actionable subpoints that reflect:

• mechanism
• evidence structure
• limitations
• validation or future work

5. Final safety and writing pass

Check that:

• the outline remains an outline
• the tone is biomedical and academic
• no claims exceed the source material

Fallback and Refusal Contract

If the input is non-biomedical or too weak to classify, respond with:

Cannot generate a biomedical outline yet.
Reason: <non-biomedical input / insufficient context / unsupported request>
Accepted retry formats:
- Review: biomedical keywords or topic direction
- Discussion: biomedical results/discussion paragraph
- Proposal: `Title:` plus background and objectives

Validation and Safety Rules

• Do not fabricate citations, statistics, or findings.
• Do not convert associative findings into causal conclusions unless the source clearly supports that level of language.
• For clinically adjacent topics, remain at academic-writing level rather than diagnostic or treatment advice.
• Surface ethics, consent, privacy, or cohort-compliance considerations when the prompt clearly implies them.

Deterministic Output Rules

• Keep section order fixed for each type.
• Use stable section labels across repeated runs.
• If an expected item is missing, ask for it or leave the section high-level; do not hallucinate details.

Examples of Accepted Inputs

Type I

Research direction: tumor microenvironment, macrophage polarization, immune checkpoint resistance
Please generate a review outline.

Type II

In our mouse model, anti-PD-1 reduced tumor burden, but the effect was lost after CSF1 overexpression. Please draft a discussion outline.

Type III

Title: Exosomal miRNAs as early diagnostic biomarkers for Alzheimer's disease
Background: include plasma exosomes, qPCR versus small RNA-seq, validation cohort, and neuroinflammation markers.

Completion Checklist

• Type detection is explicit.
• Output matches the correct outline contract.
• The result is directly usable for academic drafting.
• Any limitation or missing-input warning is surfaced clearly.