SCIGS — Scientific Journal Submission Formatter

This is the SciXZ child Skill for journal-specific manuscript formatting and submission-package preparation.

This skill combines two complementary patterns: package-first fact recovery and pre-submit verification, plus journal-family baselines and explicit journal profiles. It is DOCX-first and content-locked: formatting is allowed, silent scientific rewriting is not.

When It Applies

Use it for requests such as:

• “Format this existing Word manuscript for Journal X.”
• “Convert this paper to another journal’s submission style.”
• “Prepare blinded and non-blinded DOCX files.”
• “Check figures, supplements, references, and metadata before submission.”

Do not use it to rewrite scientific claims, repair unsupported results, invent declarations, or click a final external submission button.

Required Workflow

1. Lock the source. Copy the manuscript and supplied package. Record filenames and SHA-256 values. Inventory DOCX parts, fields, metadata, figures, tables, and supplements.
2. Verify the journal. Use the current official journal or publisher Instructions for Authors and article-type page. A family profile is only a baseline. Save the source URL, access date, evidence snapshot, and unresolved conflicts in journal_requirements.yaml.
3. Review extracted rules. Run scripts/fetch_official_requirements.py only to create candidate rules. Do not mark a rule verified until the official page has been read and the rule is confirmed. Map confirmed rules to explicit actions.
4. Format conservatively. Run scripts/format_docx.py. It applies only explicit actions such as page size, margins, font, spacing, line numbering, and page numbering. It refuses to overwrite the source and compares body/footnote text before and after.
5. Convert references safely. Keep Zotero/EndNote/Mendeley fields intact. If dynamic fields are present, scripts/reference_style.py blocks plain-text replacement by default. For plain numeric references, use DOI metadata from scripts/fetch_references.py, then convert only mapped entries.
6. Prepare anonymity. If the journal requires blinded review, run scripts/anonymize_docx.py on a copy, then scripts/anonymity_scan.py with all known author, institution, repository, grant, and project terms. A low scan result is not proof of anonymity; inspect comments, revisions, custom properties, filenames, and supplements manually.
7. Audit assets. Run scripts/asset_inventory.py and scripts/reference_audit.py. Reconcile in-text citations, captions, embedded/separate figures, table numbering, supplements, file names, and technical limits from the official instructions.
8. Render and inspect. Run scripts/render_docx.py; visually inspect every page at readable zoom. A missing renderer, failed conversion, clipping, overlap, broken table, missing glyph, or wrong page break is a blocker.
9. Package. Run scripts/prepare_submission.py, which writes FORMAT_CHANGE_LOG.md, submission-result.json, and an explicit submission-package/. Only required files are included.

Release States

• READY: all mandatory rules are verified, content integrity passes, references and assets pass, and every rendered page was inspected.
• READY_WITH_MANUAL_ACTIONS: formatting is complete but named portal or human actions remain.
• BLOCKED: any mandatory rule is unverified, a content digest changes unexpectedly, dynamic fields would be flattened, rendering is unavailable, identity leakage remains, or reference/asset integrity is unresolved.

Guardrails

• Preserve the untouched source and never silently edit scientific wording, numbers, units, tables, figures, claims, conclusions, author order, or declaration facts.
• Do not infer a mandatory journal rule from memory, a cached profile, or a third-party summary.
• Do not add optional graphical abstracts, highlights, separate figure sets, or supplementary files unless required or requested.
• Do not submit externally. Stop before Submit, Approve submission, Run checks and submit, or equivalent final action unless the user explicitly confirms that exact action.

Execution

Install dependencies with python -m pip install -r requirements.txt. Run python scripts/prepare_submission.py --help for the full command. The total pipeline intentionally requires an official requirements YAML and a real rendering tool; missing evidence or tooling must remain visible as BLOCKED.

Read the relevant references only when needed: references/requirements-workflow.md, references/docx-formatting.md, references/references-citations.md, references/anonymization.md, references/figures-supplements.md, and references/quality-gates.md.