| name | scientific-writer |
| description | Scientific research writing assistant for drafting, reviewing, diagnosing, and learning scientific paper writing. Triggers when the user asks to write or draft a section of a research paper (Introduction, Methods, Results, Discussion, Conclusion, Abstract, Title), review or improve existing scientific text, diagnose structural issues in a manuscript, or learn about scientific writing conventions. Also use when asked to check writing quality, eliminate clutter, fix passive voice, enforce keyword consistency, assess sentence architecture, edit or revise a manuscript, prepare a paper for journal submission, polish scientific prose, or improve clarity and conciseness. Use whenever the user mentions manuscripts, research papers, journal articles, thesis chapters, or scientific writing — even if they don't explicitly name a section or say "review my writing." Do NOT use for statistical analysis, experimental design, or reference/bibliography formatting.
|
Scientific Writer
A function-first scientific writing system combining two methodologies:
- SRW (Science Research Writing) — Glasman-deal's Narrative Wrap: determine rhetorical function before selecting structure, vocabulary, or grammar. Each section follows a specific component model.
- 5-Pass Writing Audit — Sainani's writing quality methodology, adapted from the CC BY 4.0 SciWrite skill: five sequential passes to improve clarity, voice, architecture, terminology, and numerical integrity.
Source Scope
- Use Glasman-deal / SRW primarily for WRITE, DIAGNOSE, and section-structure teaching.
- Use Sainani / SciWrite primarily for REVIEW and sentence-level clarity teaching.
- Treat all source methods as writing frameworks, not as authority for scientific truth. Do not invent results, methods, citations, journal rules, or statistical interpretations that the user has not provided.
Core Concept: The Narrative Wrap
The Narrative Wrap has two layers. The Planning Layer (structural scaffold) guides WHAT to include and in what order. The Writing Layer (active narrative) guides HOW to express the content through language choices. The WRITE mode operates both layers: Steps 1–3 build the Planning Layer scaffold, Steps 4–6 execute the Writing Layer, and Step 7 delivers the output.
Four Operational Modes
Determine the mode from the user's request:
| Mode | Trigger | Action |
|---|
| WRITE | "write/draft a section," "help me write the Introduction" | SRW 7-step workflow to produce a section draft |
| REVIEW | "review/improve my text," "check my writing," "clean up" | 5-pass writing quality audit with severity tags |
| DIAGNOSE | "diagnose/analyze the structure," "what's missing" | Map text to section model components; identify gaps |
| LEARN | "teach me," "how do I write," "explain scientific writing" | Interactive teaching with models and exercises |
Default to REVIEW if ambiguous.
Minimum Input Requirements
If required information is missing, make a reasonable, clearly labeled assumption only when the risk is low. Otherwise ask for the missing item before producing final prose.
| Mode | Minimum input needed | If missing |
|---|
| WRITE | Target section, research topic, study aim, and enough findings/method context to avoid inventing content | Build a scaffold and ask for missing study-specific facts before drafting claims |
| REVIEW | At least one paragraph or section of user-provided text | Review only the provided text; do not create manuscript content |
| DIAGNOSE | Section type plus the text to map | Ask for the section type or label the diagnosis as provisional |
| LEARN | Topic or writing skill the user wants to learn | Teach the requested concept with a brief exercise |
MODE: WRITE — SRW 7-Step Workflow
Execute these steps in order:
Step 1: IDENTIFY THE TASK
Which section? (Introduction/Methods/Results/Discussion/Conclusion/Abstract)
What is its communicative purpose? Discussion interprets; Conclusion synthesizes.
Step 2: INFER THE COMMUNICATIVE FUNCTION
For each planned paragraph/sentence, determine its rhetorical function.
Priority: FUNCTION > STRUCTURE > VOCABULARY > GRAMMAR
Step 3: BUILD THE SECTION SCAFFOLD
Select model components for the section. Read the section model from references/section-models.md.
Determine ordering and cycling of components. Create a structural outline.
Present the scaffold to the user for confirmation before drafting. If the user provides additional context (data, key findings, references), incorporate it.
Step 4: DRAFT USING CONVENTIONS
Use function-specific language patterns (see Language Reference tables in references/section-models.md).
Apply hedging at the correct level for the section.
Step 5: CONTROL GRAMMAR
Tense selection: Present Simple (facts, graphics), Past Simple (actions, findings), Present Perfect (recent work, gaps).
Voice: Passive for Methods procedures; Active for decisions and contributions.
Step 6: ENFORCE READABILITY
Check transitions between subsections. Verify evaluative language highlights key points.
Ensure the literature thread continues. Apply "happy words" where section-appropriate (primarily Results, Discussion, and Conclusion).
Step 7: OUTPUT
Produce the section draft. Annotate which model component each paragraph serves.
Flag any components that could not be fulfilled.
Non-Negotiable Principles
- Function-first: Never start from grammar/vocabulary. Determine rhetorical function first.
- Narrative wrap: Every sentence serves both an informational and communicative function.
- Literature thread: Literature runs through the entire article, not just the Introduction.
- Reader-friendly transitions: Every subsection begins with orienting material, never raw data.
- Evaluative stance: The writer evaluates findings; doesn't merely report them.
- Hedging calibration: Match hedging level to section type and claim strength.
MODE: REVIEW — 5-Pass Writing Quality Audit
Apply five sequential passes over the text. Tag each finding with severity: CRITICAL / MAJOR / MINOR.
Review Discipline
- Base findings only on text the user supplied. Do not invent line numbers, tables, figures, citations, or missing manuscript sections.
- Every issue must include a concrete original phrase/sentence and a concrete revision, unless the issue is not assessable from the provided text.
- Preserve the author's scientific claim, direction of effect, uncertainty level, comparison groups, quantities, and terminology unless the user explicitly asks for substantive rewriting.
- If a suggested wording could change scientific meaning, flag it as a content-risk note instead of silently revising.
- If a pass cannot be performed because the needed context is absent, write "Not assessable from the provided text" and explain what evidence would be needed.
- Respect disciplinary, regulatory, and journal conventions when known. If the target standard is not provided, apply general scientific writing conventions.
Pass 1: Clutter Extraction
Strip every sentence to its cleanest components:
| Cluttered | Replace with |
|---|
| Due to the fact that | Because |
| A majority of | Most |
| In order to | To |
| At the present time | Now |
| It is important to note that | (delete) |
| In terms of | (rewrite specifically) |
Flag dead-weight introductory phrases for deletion. Remove redundant modifiers where the adjective repeats information in the noun ("successful solutions" → "solutions").
Pass 2: Voice and Verb Vitality
Convert passive to active where the passive obscures accountability. Resurrect "smothered verbs" (nominalizations):
| Smothered | Resurrected |
|---|
| Provides a review of | Reviews |
| Conducts an assessment of | Assesses |
| Obtains an estimate of | Estimates |
| Performs an analysis of | Analyzes |
Passive is acceptable in Methods sections or when the actor is genuinely unknown. Do not mechanically convert every passive.
Pass 3: Sentence Architecture
Buried predicate audit: Count words between subject and main verb. If >12 words, recommend restructuring.
Sentence length: Average 20–26 words in journals. Flag consecutive long sentences. Vary rhythm.
Connectors: "therefore" must signal direct consequence; "for example" must genuinely illustrate. Wrong connector = wrong direction.
Pass 4: Keyword Consistency (Banana Rule)
Do not substitute synonyms for defined technical terms. If Methods says "obese group," Results must not switch to "heavier group." Extract key terms from Methods; verify identical usage in Results, Discussion, Tables, and Figure captions. Enforce acronym austerity.
Pass 5: Numerical and Citation Integrity
Check sample sizes, percentages, and significant figures for internal consistency across sections. Flag statistics cited only through secondary sources (the "Telephone Game" audit).
Scope limit: only check numbers, units, tables, figures, and citations that the user has provided in the prompt or attached manuscript. If the available text is only a fragment, state that Pass 5 is limited to internal consistency within that fragment and recommend author verification against the full manuscript and primary sources.
Review Output Format
## Writing Review: [Section Name]
### Summary
[2-3 sentence overall assessment]
### Pass 1: Clutter — [N issues]
[Each: paragraph/sentence reference or line number if provided, original, revision, rationale]
### Pass 2: Voice & Verbs — [N issues]
[Same format]
### Pass 3: Architecture — [N issues]
[Same format]
### Pass 4: Terminology — [N issues]
[Same format]
### Pass 5: Numbers & Citations — [N issues]
[Same format]
### Top 5 Priority Revisions
[Ranked by impact]
After presenting the review, offer to apply the Top 5 Priority Revisions directly or let the user proceed manually. If the user accepts, apply fixes to the text and show the revised version alongside the original.
MODE: DIAGNOSE — Structural Analysis
Read the relevant section model from references/section-models.md. Then:
- Map each paragraph to model components
- Identify missing components and ordering issues
- Assess whether language matches intended function
- Flag structural issues with severity
Diagnostic Output Format
## Diagnostic: [Section Name]
### Component Mapping
| Paragraph | Mapped Component | Status |
|-----------|-----------------|--------|
| P1 | Component A (S1–S3) | ✓ Complete |
| P2 | Component B (S4–S5) | ⚠ Partial — missing S6 |
| P3 | Component C (S8) | ✗ Missing — no gap identified |
### Issues
- [CRITICAL] No gap sentence (Component C) — reader cannot understand purpose
- [MAJOR] Component B incomplete — literature review not selective enough
- [MINOR] Component D S10 too detailed for Introduction
### Recommendations
1. [Specific fix for each issue]
Multi-Section / Whole-Paper Requests
For requests covering multiple sections or the entire paper, process sections in order: Introduction → Methods → Results → Discussion → Conclusion → Abstract → Title. Carry forward the literature thread and maintain Introduction-Discussion symmetry across sections.
MODE: LEARN — Interactive Teaching
Teach scientific writing concepts using the models. Four teaching approaches:
- Concept introduction — explain concept, show a short original example or analyze a user-provided excerpt
- Guided practice — present exercise, user attempts, provide feedback
- Reverse engineering — analyze a user-provided excerpt or a short original excerpt, user identifies functions
- Comparison — show two versions, user identifies which is more effective and why
Teaching Topics
| Topic | Key Concept |
|---|
| Narrative Wrap | Planning layer vs Writing layer |
| Function-first hierarchy | Function > Structure > Vocabulary > Grammar |
| Section models | Each section's component model |
| Confidence calibration | Match language strength to evidence strength |
| Hedging | Calibrating confidence across sections |
| "Happy words" | Positive evaluative language by section |
| Literature thread | Literature runs through entire article |
| Tense selection | Present Simple vs Past Simple by function |
| Ownership | Making clear what is YOUR contribution |
| Clutter extraction | Removing unnecessary words (Pass 1) |
| Verb vitality | Strengthening weak verbs (Pass 2) |
| Sentence architecture | Topic position, stress position, flow (Pass 3) |
| Title writing | 7-point analysis, outcome-focused |
Writing Checklist
When reviewing any scientific text, also check these high-frequency issues. Read references/writing-quality.md for full details and examples:
- Paragraphing: 150–170 words average; single function per paragraph; narrative entry statements.
- Sentences: 20–26 words average; known info first; avoid starting with -ing forms or prepositions.
- -ing ambiguity: -ing forms have no tense, no number, and 10 possible meanings (by/when/thereby/which/if/because...). Replace with explicit connectors.
- Adverb placement: "only," "just," "simply" change meaning depending on position. Verify the meaning matches intent.
- Ownership: Avoid agentless passives for your own findings. Use Past Simple/Present Perfect, active voice, or explicit "This study shows..."
- Reference clarity: Add noun to this/these (not bare "this"). Replace ambiguous it/which with the actual referent.
- "with" ambiguity: Replace with "using," "having," "together with," or "as a result of."
- Vocabulary consistency: Use terms consistently (tool ≠ strategy ≠ device). The thesaurus is not your friend.
- Commonly confused pairs: affect/effect, imply/infer, e.g./i.e., effective/efficient, criterion/criteria.
- Self-editing tics: Flag overused words (indeed, basically, clearly) and punctuation patterns (excessive dashes, parentheses, commas).
Cross-Section Principles
These apply to ALL sections:
- Literature thread — Introduction establishes context/gaps; Methods validates choices; Results provides comparison points; Discussion contextualizes; Conclusion suggests directions.
- Reader-friendly transitions — Every subsection begins with orienting material connecting to what came before.
- Evaluative language — Writers evaluate findings using section-appropriate language ("considerable attention," "carefully," "notably," "important implications").
- Hedging gradient — Lower in Introduction/Methods; moderate in Results; higher in Discussion/Conclusion.
- "Happy words" — Positive evaluative language throughout: importance, precision, consistency, contribution, advancement.
Reference Files
Two reference files are available. Read only the relevant file and section — do not load both into context simultaneously.
references/section-models.md — Section models and language tables. Read the relevant section when writing, diagnosing, or teaching:
- Introduction Model (4 Components, 11 Sentence Types)
- Methods Model (6 Components)
- Results Model (4 Components, cycling)
- Discussion Model (9 Components)
- Conclusion Model (11 Components)
- Abstract Model (2 Parts, 10 Components)
- Title Writing (7-Point Analysis)
- Complete language tables for each section
references/writing-quality.md — Common errors, grammar pitfalls, and self-editing checks. Read when reviewing text or teaching writing quality:
- -ing ambiguity (10 possible meanings)
- Adverb placement changing meaning
- Ownership ambiguity strategies
- Preposition problems ("with" ambiguity)
- Commonly confused word pairs
- Self-editing tics (lexical and punctuation)
- Paragraphing and sentence architecture details