Science Research Writing

Related Skills

• When the task expands into staged document collaboration across outlining, section-by-section ideation, drafting, refinement, and reader checks, use doc-coauthoring from the anthropics/skills repository.
• When the user needs a concept figure, framework diagram, schematic, or other polished single-page visual artifact, use canvas-design from the anthropics/skills repository.
• When the task requires citation lookup, metadata verification, DOI-to-BibTeX conversion, or bibliography cleanup, use citation-management from the K-Dense-AI/claude-scientific-writer repository.
• When the scientific structure is already sound but the prose still feels synthetic or machine-regular, use humanizer from the blader/humanizer repository for the final tone pass.
• Do not add absolute paths here: if these skills are installed, the model can discover them by skill name; if one is unavailable, say that explicitly to the user and continue with the best in-skill fallback.

Workflow

1. Identify the job: draft, revise, or diagnose.
2. Identify the target section and read the matching playbook in references/manuscript-map.md.
3. Read references/style-controls.md when the task involves tense, voice, hedging, cohesion, causality, articles, terminology, or sentence-level diagnosis.
4. Read references/workflow-extensions.md when the task involves manuscript assembly order, tables and figures, review articles, response letters, or lay summaries.
5. Build the paragraph plan in sentence functions before writing. Decide what each sentence must do for the reader, not just what information it contains.
6. Draft or revise from general to specific. Orient the reader early, present evidence clearly, and add interpretation or limitation when the reader would otherwise have to guess.
7. Calibrate certainty. Match tense, modality, and causal wording to the strength of the evidence.
8. Run references/revision-checklist.md before handing text back.

Section Routing

• Use references/manuscript-map.md for task triage, section-level move patterns, paragraph roles, and section-specific failure modes.
• Use references/style-controls.md for sentence-level control over clarity, ownership, certainty, causality, and rewrite tactics.
• Use references/workflow-extensions.md for manuscript assembly order, table/figure storytelling, review articles, lay summaries, and response-to-reviewers discipline.
• Use references/revision-checklist.md for final review across structure, sentence logic, evidence discipline, and delivery readiness.

Maintenance Notes

• Keep this entrypoint concise. Put new tactical depth in references/ unless the routing itself has changed.
• Before adding a new adjacent task here, decide whether it should be routed to another installed skill instead of expanding this one.
• When updating related-skill names or fallback behavior, keep the names aligned with the actual installed skills in the current environment.
• For future maintenance and skill-improvement work only, reviewer-oriented checks may reference Skill Reviewer from the openclaw/skills repository, and iterative fix loops may reference Skill Improver from the sickn33/antigravity-awesome-skills repository. Do not add absolute paths here: if those skills are installed, the model can discover them by name; if they are unavailable, say that explicitly to the user. Do not route normal research-writing tasks to these maintenance-only helpers.

Output Requirements

• Preserve the scientific meaning unless the user explicitly asks to change the claim.
• Preserve numbers, comparisons, and citations unless they are clearly inconsistent with the source text.
• Prefer explicit logic over stylistic variation.
• Repeat the same term for the same concept unless a distinction is intended.
• Avoid unsupported novelty claims, inflated causal claims, and vague pronoun references.

Common Transformations

• Convert notes into prose by mapping sentence functions first, then drafting the paragraph.
• Convert data-heavy text into readable Results prose by adding commentary around the numbers.
• Convert overclaimed language into evidence-matched language by weakening certainty or causality where needed.
• Convert flat prose into reader-guided prose by making contribution, limitation, or implication explicit.
• Convert display-first study materials into a manuscript section by building the story from tables and figures.
• Convert reviewer criticism into a calm, specific response plus the nearby textual repair it implies.

Example Requests

• Use $science-research-writing to rewrite this abstract so the gap, method, and main finding are explicit.
• Use $science-research-writing to revise this Discussion paragraph without strengthening the claim.
• Use $science-research-writing to turn these bullet points into an Introduction paragraph for a biomedical journal.
• Use $science-research-writing to diagnose why this Results paragraph is hard to follow and then rewrite it.
• Use $science-research-writing to draft a point-by-point reply to these reviewer comments without sounding defensive.
• Use $science-research-writing to turn this paper abstract into a lay summary for non-specialist readers.