Sci manuscript architect

SCI Manuscript Architect

This skill builds biomedical SCI manuscripts from study materials and first decides where the project sits in its field. It combines ToolUniverse-backed field positioning, Supervisor-Skills-inspired logic gates, and biomedical evidence discipline: scan the field, lock the motivation and innovation, estimate contribution level, map claims to evidence, design the manuscript and figures, align language with a field corpus, then draft and audit.

Default language policy:

• Use Chinese for analysis, decisions, risks, matrices, and explanations.
• Use English for manuscript titles, abstracts, section prose, captions, replacement sentences, and journal-facing text.

Use This Skill For

• Evaluating whether a biomedical project is worth writing and what submission level it can realistically target.
• Building a biomedical SCI manuscript from experiment notes, figures, result summaries, protocols, PDFs, references, datasets, or partial drafts.
• Creating Field Scan Reports, Motivation and Gap Locks, Innovation and Contribution Matrices, Submission-Level Estimates, PaperSpine Maps, Evidence Ledgers, Citation Support Banks, IMRAD Blueprints, Figure Architecture Plans, reviewer audits, and journal checklists.
• Checking whether claims are supported by user evidence, citations, figures, statistical results, reporting-guideline requirements, and external field context.
• Preparing English manuscript prose while keeping Chinese reasoning visible to the user.
• Building or using a field corpus to align terminology, collocations, hedging, journal style, and context-aware narrative logic with high-quality papers in the same biomedical field.

Operating Rules

1. Do not draft the full manuscript until the field position, central motivation, contribution level, and evidence boundary are clear. If the user asks for prose too early, first produce the missing strategy artifacts.
2. Never fabricate experiments, sample sizes, p-values, datasets, figures, citations, ethics approvals, software versions, guidelines, clinical relevance, or therapeutic claims.
3. Treat user materials as authoritative for study results. Literature, ToolUniverse outputs, and exemplar papers can shape positioning, motivation, background, and style, but cannot create new findings.
4. If no live database or literature scan has been performed, mark novelty, field position, and submission-level judgments as provisional with FIELD_SCAN_REQUIRED.
5. Mark missing evidence explicitly as MISSING, uncertain citations as VERIFY, overbroad claims as OVERCLAIM, incomplete evidence as EVIDENCE_INCOMPLETE, unvalidated mechanisms as VALIDATION_REQUIRED, and submission-level overreach as CLAIM_OVERREACH_RISK.
6. Use reporting guidelines by study type. If the study type is unclear, ask the user or state the assumption before applying a checklist.
7. Corpus materials are style and terminology evidence, not claim evidence. Do not copy protected full text or imitate distinctive sentences.
8. Supervisor-Skills is a structure source only. Translate its logic gates into biomedical SCI terms; do not force CS top-conference paper patterns onto clinical, wet-lab, or biomedical discovery papers.
9. Submission-level estimates are strategic guidance, not acceptance predictions.

Standard Workflow

Follow this order unless the user requests a specific artifact:

1. Project Intake and Field Scan: collect study type, disease or biological process, materials, core finding, target journal, evidence available, and target output. Use templates/project_intake.md and templates/source_inventory.md. When field positioning is requested, read references/field_positioning.md and use ToolUniverse-backed sources when available.
2. Motivation, Gap, and Innovation Lock: decide whether the motivation is a real clinical unmet need, biological mechanism gap, methods bottleneck, or translational opportunity. Read references/motivation_and_gap_engine.md, references/innovation_taxonomy.md, and references/contribution_level_matrix.md; use the matching templates.
3. Evidence and Claim Architecture: map every major claim to user data, figure, statistic, citation, guideline, or missing evidence. Read references/evidence_ledger.md, references/citation_support_bank.md, and references/paper_spine_map.md.
4. IMRAD and Figure Narrative Design: create the biomedical Introduction flowchart, IMRAD blueprint, writing rationale, and figure architecture plan. Read references/introduction_flowchart.md, references/writing_rationale_matrix.md, and references/figure_architecture.md.
5. Language and Corpus Alignment: when field-style alignment is requested or enough topic information is available, create or update a corpus profile using public metadata, abstracts, user-provided references, and open-license full text only. Read references/corpus_intelligence.md and references/terminology_and_style.md.
6. Drafting and Submission-Level Audit: write English manuscript sections from approved strategy artifacts, then run reviewer, journal-readiness, and severity audits. Read references/reviewer_audit.md, references/journal_checklist.md, references/submission_level_estimator.md, references/submission_severity_audit.md, and references/ai_research_integrity.md.

Required Core Artifacts

For a full from-materials workflow, produce these artifacts in order:

• Project Intake
• Source Inventory
• Field Scan Report or FIELD_SCAN_REQUIRED provisional note
• Motivation and Gap Lock
• Innovation and Contribution Matrix
• Submission-Level Estimate
• Motivation Lock
• PaperSpine Map
• Evidence Ledger
• Citation Support Bank
• Biomedical Introduction Flowchart
• IMRAD Blueprint
• Figure Architecture Plan
• Writing Rationale Matrix
• Term Bank, Phrase Pattern Bank, Journal Style Profile, and Language Guard when corpus materials are available
• English manuscript draft or requested section draft
• Reviewer Audit
• Journal-Ready Audit
• Submission Severity Audit

If the user requests only one artifact, produce that artifact and note which upstream inputs are assumed, missing, or provisional.

ToolUniverse Field-Positioning Sources

Use ToolUniverse when the user asks for current field progress, novelty, submission level, clinical relevance, translational value, or target maturity. Use only the tools needed for the task:

• PubMed_search_articles and PMC_search_papers for field progress, recent work, reviews, and open full text.
• iCite_search_publications and iCite_get_publications for citation count, RCR, APT, NIH percentile, and human/animal/molecular flags.
• MeSH_search_descriptors and BioPortal_annotate_text for concept normalization.
• PubMed_Guidelines_Search and TRIP_Database_Guidelines_Search for clinical guidelines and unmet clinical need.
• Pharos_search_targets and TargetMine_search for target development level, target-disease context, druggability, and translational relevance.

If these tools are not used, clearly mark field and journal-level judgments as provisional.

Supervisor-Skills Adaptation

Use these Supervisor-Skills ideas only after translating them into biomedical SCI context:

• intro-drafter: biomedical six-part Introduction flowchart.
• tech-paper-template: background, gap, objective, study design, evidence, contribution self-consistency chain.
• figure-designer: study design figure, workflow figure, mechanistic model, main finding figure, validation figure, and graphical abstract logic.
• pre-submission-reviewer: CRITICAL, MAJOR, MINOR submission severity taxonomy.
• benchmark-paper-template: only for dataset, resource, tool, benchmark, or model-comparison papers.
• vibe-research-workflow: AI integrity rules; AI may accelerate organization, code, figures, and language polish, but cannot own scientific judgment.

Submission-Level Estimate

Every submission-level estimate must include:

• Recommended tier: Top/high-impact, solid specialty, method/resource, or descriptive/lower-level.
• Evidence basis: field heat, novelty, evidence strength, translational value, target or mechanism maturity, guideline relevance, and journal fit.
• Risk labels: FIELD_SCAN_REQUIRED, EVIDENCE_INCOMPLETE, VALIDATION_REQUIRED, and/or CLAIM_OVERREACH_RISK.
• Upgrade path: concrete evidence, validation, analysis, figure, or framing changes needed to target a higher tier.

Biomedical SCI Defaults

• Default workflow: strategy first, then manuscript building from materials.
• Default audience: biomedical journal reviewers and editors.
• Default manuscript structure: IMRAD unless the target journal requires a different structure.
• Default reporting-guideline candidates:
  • CONSORT for randomized trials.
  • STROBE for observational studies.
  • PRISMA for systematic reviews and meta-analyses.
  • ARRIVE for animal studies.
  • TRIPOD for prediction model studies.
• Default output: Chinese analysis plus English manuscript prose.

Reference Routing

• For field progress and ToolUniverse scan planning, read references/field_positioning.md.
• For motivation and gap quality, read references/motivation_and_gap_engine.md.
• For innovation type, read references/innovation_taxonomy.md.
• For contribution level, read references/contribution_level_matrix.md.
• For submission-level strategy, read references/submission_level_estimator.md.
• For central motivation and claim scope, read references/motivation_lock.md.
• For the argument spine, read references/paper_spine_map.md.
• For claim support and missing-evidence checks, read references/evidence_ledger.md.
• For literature-to-claim matching, read references/citation_support_bank.md.
• For Introduction structure, read references/introduction_flowchart.md.
• For writing-unit planning, read references/writing_rationale_matrix.md.
• For figure logic, read references/figure_architecture.md.
• For corpus collection, field tracking, and allowed corpus sources, read references/corpus_intelligence.md.
• For terminology, phrase patterns, and style constraints, read references/terminology_and_style.md.
• For peer-review simulation, read references/reviewer_audit.md.
• For severity-ranked final audit, read references/submission_severity_audit.md.
• For reporting and submission checks, read references/journal_checklist.md.
• For AI-assisted research boundaries, read references/ai_research_integrity.md.

Keep the main response concise. Load only the reference files needed for the current user request.

Context-Aware Narrative Learning Defaults

When building a long-term corpus for manuscript writing, prefer a two-layer corpus-to-narrative structure:

corpus_workspace/
  corpus_metadata.jsonl
  high_impact_corpus_metadata.jsonl
  field_scan_report.md
  term_bank.md
  phrase_pattern_bank.md
  journal_style_profile.md
  language_guard.md
  narrative_learning/
    01_online_context_engine/
    02_manuscript_specific_tools/
    03_general_move_library/
    legacy_first_pass/

The 01_online_context_engine layer should classify source-use environment, evidence type, journal role, section function, and claim-strength context before any language pattern is adapted. This is especially important for non-native English scientific writing: the goal is to learn paragraph function and evidence logic, not to imitate sentences.

Corpus Scripts

Optional helper scripts live in scripts/:

• collect_corpus.py: collect PubMed/OpenAlex metadata and abstracts into a local corpus workspace.
• extract_terms.py: extract terminology and phrase-pattern candidates from the corpus.
• build_style_profile.py: build a journal style profile and language guard from the extracted corpus artifacts.

Use scripts when the user wants repeatable corpus updates. For a one-off strategy or writing request, it is acceptable to create the same artifacts manually from supplied papers, references, or ToolUniverse outputs.