Food Paper

Food-Paper — Whole-Process Manuscript System for Food & Nutrition Science

Take a food/nutrition study from data and idea to a submission-ready, journal- formatted manuscript, using a team of subagents for each stage of the research and writing process. Original work; architecture informed by open community paper-writing and Nature-style skills (see the repo README Acknowledgements).

First move — resolve the target journal (once)

Before drafting, load journal-selector/SKILL.md (a shared procedure, not an installed skill) and follow it — it asks the author which journal they are targeting (they may answer 'generic' for APA 7.0 defaults). Do this once: record the resolved journal and its constraints and reuse them for every subagent and stage — do not ask again. Re-run journal-selector only if the author asks to switch journals, or reuse the choice already resolved by food-pipeline/an earlier turn. The constraints govern structure, word/abstract limits, reference style, and the figure spec passed to food-figure.

Modes

• full (default) — the whole pipeline: field → questions → data/stats → figures → argument → draft → polish → self-review.
• plan — Socratic planning of the paper chapter by chapter (no full draft).
• outline — detailed outline + evidence map only.
• section — draft or rewrite one section (intro/methods/results/discussion/abstract).
• stats — statistical analysis plan/execution guidance only.
• revise — revise against an existing review (a food-review report and/or margin comments on a Word file). Edit the original .docx with Tracked Changes (do not start a fresh copy), resolving each comment. Inside food-pipeline: update the existing Review & Response Report (.docx) in place with each item's response — no separate response letter. Standalone (real journal reviewers): produce a point-by-point response letter as a new Word document. All deliverables are .docx, never Markdown. See references/revision-response.md.
• format-convert — convert a draft to the target journal's structure + reference style; output Markdown, LaTeX (.tex), or DOCX, and build a PDF via Pandoc or latexmk (see references/latex-guide.md). Yes — this skill can prepare and edit LaTeX drafts. When the source is a Word file, preserve EndNote/Zotero/Mendeley citation fields (references/word-field-codes.md) — don't flatten field codes into text.
• polish — language editing to publication-quality English and removal of AI writing tells (polisher runs references/human-writing.md): inflated significance, "-ing" tack-ons, vague attribution ("studies have shown" → a real citation), stock AI vocabulary, "serves as" → "is", filler, hedge stacking, generic upbeat endings — then the two-pass check "what still reads as machine-written?". Keeps calibrated hedging, passive Methods, and journal-mandated form; never changes a number, claim scope, or citation. Good for non-native-English drafts and for anything AI helped write.

Subagent team (dispatch via the Agent tool; independent stages run in parallel)

Workflow

flowchart TD
    A[intake<br/>type, journal, data, goals] --> JS[journal-selector<br/>load journal constraints]
    A --> L[literature_lead<br/>-> food-research evidence base]
    L --> Q[question_framer<br/>RQs / hypotheses / contribution]
    A --> D[data_curator<br/>curate + check dataset]
    D --> S[statistician<br/>analysis plan + results]
    S --> V[viz_designer<br/>-> food-figure figures & tables]
    Q --> ST[structure_architect<br/>journal-mapped outline + evidence map]
    S --> ST
    V --> ST
    ST --> AR[argument_builder<br/>CER chains, results->discussion]
    AR --> W[draft_writer<br/>section drafts]
    W --> P[polisher<br/>publication English]
    P --> C[citation_manager<br/>journal reference style]
    C --> IR[internal_reviewer<br/>-> food-review panel]
    IR -- revise --> W
    IR -- ready --> OUT[Submission-ready manuscript]

Food & nutrition reporting defaults (enforced by draft_writer / data_curator / statistician)

Composition as g/100 g (basis + AOAC method); sensory panel type/size/scale + ethics; microbial counts log CFU/g with LOD; TPA/rheology parameters + settings; HPLC/GC/LC-MS conditions, LOD/LOQ, recovery, identification by standards/MS-MS; mean ± SD/SEM with n; the statistical model, test, post-hoc, and threshold, with significance shown consistently. Reproducible Methods (cultivar/breed/batch, prep, storage). Ethics/food-safety statements where relevant.

References (load as needed)

• references/paper-structure.md — structure_architect: IMRaD/review patterns + abstract types.
• references/writing-style.md — draft_writer/polisher: scientific style, title/intro rhetoric.
• references/human-writing.md — write like a scientist, not a chatbot: the AI writing tells to remove (inflated significance, vague attribution, stock vocabulary, hedge stacking), the academic exceptions to keep (calibrated hedging, passive Methods, journal-mandated form), and the two-pass check. Canonical for the suite; used by food-research, food-deep-research, and food-review too.
• references/writing-quality-check.md — self-check before internal_reviewer.
• references/statistics-reporting.md — statistician: what to report and which test.
• references/declarations-guide.md — CRediT, funding, COI, data availability, ethics.
• references/apa7-quickref.md — default citation style for citation_manager (canonical APA 7.0 for the suite).
• references/faithfulness-and-citation.md — grounding rules + four-gate citation check; the suite's no-fabrication contract. Run scripts/verify_citations.py on the reference set.
• references/latex-guide.md — prepare/edit LaTeX drafts and build the PDF (Pandoc / latexmk).
• references/revision-response.md — revise mode: tracked changes (original Word opt-in under food-pipeline) + either updating the one Review & Response Report in place (pipeline) or a point-by-point response letter (standalone). All .docx.
• references/word-field-codes.md — preserve EndNote/Zotero/Mendeley citation fields when editing a .docx (don't flatten field codes into visible text); verify with scripts/check_docx_fields.py.
• references/privacy-and-confidentiality.md — privacy check before delivery (no local paths/secrets); run scripts/privacy_scan.py.

Grounding (non-negotiable)

Write only from the user's data and verified literature. Never invent references, DOIs, numbers, or results; unsupported content is marked [UNVERIFIED]/[EVIDENCE GAP], never filled from memory. citation_manager and draft_writer enforce the four-gate citation check in references/faithfulness-and-citation.md.

Handoffs

food-research (evidence in) → food-paper → food-review (external panel) → food-paper revise. Orchestrated by food-pipeline.