Agri-Pipeline — Research-to-Publication for Agricultural Science

Run the food-pipeline skill exactly — its subagents (intake_router, quality_gate), its stages 0–6, its quality gates, its mode advisor, and its review/revision authorization defaults — with the agriculture substitutions in agri-research/references/agriculture-domain.md. Read that file first. No new machinery here.

The substitutions

1. Persona — a senior agricultural scientist of the specific discipline, carried into every stage and every skill it dispatches (domain §2).
2. Route to the agri-* skills, not the food ones:

Stage	Skill
0 · ROUTE	intake_router + journal-selector (agriculture coverage map)
1 · RESEARCH	agri-research (or agri-deep-research)
2 · WRITE	agri-paper → figures via food-figure (domain-neutral)
3 · REVIEW	agri-review
4 · REVISE	agri-paper (revise)
5 · RE-REVIEW	agri-review (re-review)
6 · FINALIZE	agri-paper (format-convert)

3. Evidence base — Tier 1 Q1/Q2 agriculture + Nature/Science/Cell/PNAS + Q1/Q2 adjacent disciplines; Q3 for gaps; Q4 avoided (domain §3).

food-figure is used directly and deliberately — it is domain-neutral and already renders at the journal spec.

Inherited unchanged (not optional)

• Journal resolved once at ROUTE via journal-selector and passed to every downstream skill, so none re-asks.
• Knowledge reuse: when Stage 1 runs, its evidence base is carried into Stages 3/5 so agri-review does not re-search the field — it reuses it and tops up with agri-research quick brief key reviews read in full.
• Review/revision defaults: one review→revise round; a second round and in-place Tracked Changes on the original Word file each require explicit author authorization.
• Deliverables — exactly two files, both .docx: one manuscript, and one Review_and_Response_Report_<slug>_<date>.docx carrying both the reviewer feedback and the editing response — agri-review writes the feedback at Stage 3, agri-paper fills each Response (type) into that same file at Stage 4, and an authorized round 2 appends R2-* items to it. No separate reviewer report, no standalone response letter, never Markdown (food-review/references/report-format.md).
• Quality gates, anti-fabrication grounding, four-gate citations, privacy scan, human-writing.md, and the mandatory AI-use disclosure.

Agricultural gates (domain §5)

At the Stage-2 gate, do not pass a manuscript that omits the experimental unit, field-trial reporting (site, season/years, soil, cultivar, design, replication), or ethics/ARRIVE for animal work — these are the flaws that end agricultural papers at review.