Nature Polishing

Nature-Style Academic Polishing — Router

This skill is split into two layers:

• A static layer under static/ that holds versioned, reusable content fragments (core principles, paper-type playbooks, per-section guidance, language-specific rules, per-journal style).
• A dynamic layer (this file plus manifest.yaml) that detects the request's axes and loads only the fragments needed for the current job.

Do not try to apply the polishing logic from memory or from this router. Always load fragments from disk as described below.

Routing protocol

Follow these five steps every time the skill is invoked.

1. Load the manifest and the core layer

Read manifest.yaml. It declares the axes (paper_type, section, language, journal), the allowed values, and the file paths each value maps to.

Also read every file listed under always_load. These hold the default stance, failure-mode diagnosis, ethics, and output format that apply to every polish job.

2. Detect the axis values for this request

For each axis in the manifest, decide the value using the manifest's detect: hint and the user's input:

• paper_type — research / methods / hypothesis / algorithmic / review. Default: research.
• section — abstract / intro / results / discussion / conclusion / title / methods. May be multiple. Ask the user if it is ambiguous and matters for the polish.
• language — en or zh-to-en. Detect from the draft itself.
• journal — nature / nat-comms / nat-mach-intell / generic. Default: generic. Use nature only for flagship Nature, nat-comms for Nature Communications and nat-mach-intell for Nature Machine Intelligence (NMI). Do not route another Nature Portfolio title through flagship Nature rules.

State the detected axis values in one short line to the user before proceeding, so they can correct you cheaply. This is a progress update, not an approval gate; continue unless a necessary decision remains unresolved.

3. Load the matching fragments

For each axis value, Read the file mapped in the manifest. Skip the section axis only if the user has supplied free-floating prose with no section context.

Do not read every fragment in static/. Load only what step 2 selected.

4. Polish using the loaded material

Apply the loaded fragments in this priority order, matching the paper type -> section job -> paragraph logic -> claim/evidence/boundary -> sentence polish rule from core/failure-modes.md:

1. Paper-type playbook (architecture, writing order).
2. Section-specific job and failure modes.
3. Journal-specific framing and constraints.
4. Language-specific sentence and paragraph rules (apply last).
5. Core stance and ethics throughout.

If a paragraph's structural problem cannot be fixed without inventing content, flag it instead of papering over it.

For Results, full-main-text compression, main-versus-SI allocation, or prose added during revision, load ../nature-shared/core/main-text-discipline.md before sentence polishing. Classify each result, retain the shortest sufficient evidence chain, and require every addition to trigger a deletion or replacement check across the affected paragraph.

For flagship Nature, Nature Communications, Nature Machine Intelligence, or another Nature Portfolio title, load the matching shared Nature-style corpus guidance:

• Results or Discussion → ../nature-shared/core/nature-results-discussion.md
• Introduction or whole-manuscript narrative → ../nature-shared/core/nature-introduction.md
• Abstract → ../nature-shared/core/nature-abstract.md

Preserve claim escalation, the fast question funnel, Introduction–Results alignment, discovery-centred abstract compression, evidence-bound local interpretation, and cross-Results synthesis. These defaults were initially distilled from published NMI papers; treat them as corpus-derived guidance, not official policy, and obey the target journal's current rules when they differ.

For any Discussion polish or restructuring job, also load ../nature-shared/core/discussion-argument-language.md. Use its function labels to remove Results replay, repair the movement from specific findings to bounded implications, calibrate modal and reporting verbs to evidence strength, and make limitations and future work resolve named claim boundaries. Treat it as general writing guidance, not journal policy.

5. Reach for references only when needed

The files under references/ are deep references, not defaults. Open them on demand per the references.on_demand table in the manifest, for example when the user explicitly asks for phrasebank-style alternatives or a stricter style audit.

When the target is Nature Machine Intelligence and exact limits, availability sections, conference-extension disclosure or production checks affect the revision, load ../nature-shared/journal-formats/nature-machine-intelligence.md.

When the job is a whole manuscript rather than a passage, or the text has already been through more than one round of editing, also load ../nature-shared/core/consistency-sweep.md. Polishing passage by passage cannot see accumulated drift: one experimental factor under several names, the same quantity in two units, a metric at two precisions, or a superlative the paper's own table contradicts. Sweep for those before working on sentences, and repeat the sweep until a pass finds nothing new.

Layout/typesetting (排版) requests are different. If the user asks to fix placement rather than wording — loose/sparse pages, stranded headings, figures that don't fill the page or split across pages, "Float too large", multi-panel arrangement, sparse Supplementary Information — skip the prose axes (paper_type, section, language, journal) and load references/latex-layout.md directly. That file is self-contained: it carries the diagnosis workflow (render → contact-sheet → read the log), the float-glue and [H]/\clearpage/placeins patterns, and the "regenerate wide figures taller at the source" rule. Always compile and visually inspect rendered pages before and after — never judge layout from the .tex alone.

Why this split

• The static layer is versioned and reviewable. Adding a new journal style or paper type is one new file plus one manifest line.
• The dynamic layer keeps each invocation cheap: only the fragments relevant to this draft enter context, instead of the full 1000-line monolith.
• The router itself is short on purpose. Update fragments, not this file, when adding scope.