---

name: journal-style-learn description: >- Learn a journal's structural and rhetorical patterns from supplied papers or an existing style analysis, and adapt the local nature skill family to that journal. Use for 把这本期刊适配到nature系列、新增期刊、期刊风格迁移、从论文生成期刊规则, bilingual chapter profiles and journal routing. Can also learn style only. Adapts reusable skills; polishing a manuscript belongs to nature-polishing.

Journal Style Learn — Nature family adapter

This skill owns learning and registering journal adaptations, not manuscript rewriting. Reuse the installed nature-writing, nature-polishing and nature-shared architecture. Keep the current skill name so existing invocations still work. Chinese reading copy: SKILL-zh.md.

Determine the requested outcome

• “适配到 nature 系列 / add this journal / 生成期刊适配技能” means learn or reuse grounded rules, generate bilingual resources, register the journal in the selected local nature family, validate and package the adaptation.
• “学习期刊风格 / 分析这些论文 / 只生成规则” means learn-only: produce journal_style_rules.md, leaving installed skills untouched. Explicit preview or package-only requests likewise do not authorize installation.
• “把我的论文改成这个期刊风格” means manuscript revision: use the existing nature-polishing route. Build a new adaptation only if the user requests it; do not silently edit skill infrastructure during a manuscript edit.

Infer journal identity, corpus location, discipline/article type and installation root from the request and available files before asking. Match explicit journal context; do not confuse a cited venue with the target. A folder of mixed venues requires separating primary journal examples from supplemental material. Five to eight relevant articles is a useful starting sample, not a mandatory quota or proof of a journal-wide convention. Ask only for an essential unresolved identity or inaccessible corpus, continuing independent work where possible.

Learn from evidence

1. Inventory supplied PDFs, Markdown, Word or existing journal_style_rules.md. Read existing analysis first when provided; mark derivative synthesis clearly instead of claiming a new full-text analysis. Do not invent sample counts.
2. Read usable sources with available format tools. Prefer existing readable text; use PDF extraction/OCR when needed. MinerU is optional, not a prerequisite. Check section order, paragraph boundaries and damaged formulas/tables. Exclude unreadable passages from inference and report coverage limits.
3. Extract structural observations by paper and section with source locations: entry point, paragraph function chain, literature comparison, gap-to-design connection, method explanation, results-to-claim relation, limits and language. Describe rhetorical patterns; do not copy source prose, data or findings into rules or future manuscripts. A paper title/DOI can identify provenance.
4. Aggregate only comparable papers. Separate observed patterns, operational synthesis, general writing advice and verified official requirements. A missing feature in a few papers is not a forbidden practice. Keep exceptions and evidence strength visible; paragraph/word/contribution counts are not quotas.
5. Where helpful, load one relevant base norm from base_rules/ (economics, ml_cv_nlp, cs_engineering or general_academic). It is a general fallback, not journal evidence. Do not impose canned forbidden phrases irrespective of use.

The reusable analysis records journal/key/aliases, corpus and article-type scope, source coverage, per-section patterns and exceptions, source locations, confidence, and synthesis rationale. Keep it self-contained and readable. For learn-only, return this analysis and stop; do not force chapter generation.

Adapt the nature family

Read references/nature-adaptation.md for an adaptation request. Follow its discovery, file contract, registration, preservation, validation and packaging steps. Inspect the actual target installation rather than assuming this machine's paths or a fixed number of family members.

Preserve the target installation's original guides, configured journal routes, evidence-completion behavior, natural-prose rules, language preferences and verification annotations. The adapter supplies journal patterns and references; it must not reset core workflows or import another journal's formal limits. Generate English runtime files and Chinese reading copies beside originals.

Do not represent an official requirement as verified without its current primary source. Consult the official author guidance when requirements are being generated; if inaccessible or browsing is forbidden, mark those fields unverified and omit unsupported numeric constraints. A source-corpus pattern is never official policy.

Validate and report

Run scripts/validate_nature_adapter.py --skills-root PATH --journal KEY after integration; the script reads only and checks the current file contract. Inspect existing-journal preservation and effective instructions as well as paths. Report static checks separately from real behavioral tests. A genuine writing trial needs manuscript evidence; do not fabricate research or claim a trial from routing checks.

Return the installed/package paths, affected skills, provenance/scope, unresolved requirements and validation results. Provide short user prompts for writing and polishing with the new journal. Do not require the author to fill a setup template.