---

name: academic-writing description: "Supports evidence-disciplined medical manuscript, review, abstract, figure, table, submission, peer-review, and grant writing without fabricating data, citations, or compliance." disable-model-invocation: true metadata: disco-role: operating license: MIT

Academic Writing

Use this route for medical or biomedical manuscripts and reviews: planning or revising IMRaD sections, structured abstracts, figures and tables, citation integrity, reporting checklists, journal adaptation, peer-review responses, submission preflight, and grant components. The researcher remains the author and must approve every scientific claim.

Intake gate

Before drafting, establish:

• artifact: full manuscript, review, section, abstract, figure/table text, response letter, submission package, or grant;
• study/review type, research question, evidence available, and target audience;
• target journal or funder, format, word limit, declarations, and deadline;
• whether data, source papers, references, figures, and reviewer comments are actually supplied; record missing items rather than inferring them;
• privacy status and whether the package is double-blind. Do not request or reproduce identifiable patient information.

If the evidence is insufficient, ask focused questions or produce an outline, input checklist, or marked partial draft—not plausible-looking prose.

Route the work

1. Read workflows.md for the artifact-specific workflow and handoffs.
2. Use the claim ledger and source-fit tests in citations-and-integrity.md.
3. Apply troubleshooting.md when inputs, privacy, blinding, references, figures, reporting checklists, or submission constraints fail validation.
4. For a long review, use staged outline → evidence search/library → chapter-by-chapter drafting → verification → abstract/conclusion → format; preserve intermediate state and obtain author checkpoints.
5. For Methods, Results, and Discussion, keep boundaries explicit: Methods reports what was done, Results reports what was observed, and Discussion interprets only those results in light of supplied literature.

Non-negotiables

• Never invent participants, outcomes, effect sizes, p-values, software, ethics IDs, approvals, citations, reviewer facts, grant preliminary data, or journal policy. Use [AUTHOR TO SPECIFY: ...], [CITE: ...], or an uncertainty note when necessary.
• Match claim strength to design and evidence: association is not causation; exploratory, animal, in-vitro, single-center, or unvalidated findings must not be generalized into clinical certainty.
• Identify the applicable reporting framework from the actual design. Use a hybrid review when necessary rather than falsely certifying one checklist.
• Do not rewrite or silently alter numerical results, denominators, figure labels, reference identities, or author contributions. Check consistency across title, abstract, Methods, Results, Discussion, tables, figures, and supplements.
• For journal recommendations, rank scope/design/evidence fit and label current metrics as approximate; verify live instructions and policies on the official journal or funder site. Never predict acceptance.
• For submission, remove PHI and local identifiers. For double-blind review, separate title-page metadata and scrub names, institutions, acknowledgments, self-identifying phrasing, file metadata, and revealing repository links as the journal requires; do not conceal mandatory ethics or conflict disclosures from the confidential submission fields.

Handoffs

• Route literature retrieval, evidence maps, source appraisal, and replacement citations to evidence-insight; return with traceable source records.
• Route statistical computation, data cleaning, model fitting, uncertainty, and figure generation to data-analysis; this skill reports supplied outputs but does not manufacture analyses.
• Route study design, endpoints, randomization, power, eligibility, protocol, and ethics planning to protocol-design before writing Methods or grants.
• After a stable Results section, hand the evidence boundary to Discussion; after a compliance or integrity review, revise the owning section and rerun the relevant check.

Output contract

Deliver the requested artifact plus a compact audit note: inputs used, study or reporting framework selected, claims or values left unresolved, citation and privacy checks, journal/funder assumptions, and next handoff. Stop and report a gap when verification cannot be completed.