Agri-Deep-Research — Source-Validated Reviews for Agricultural Science
Run the food-deep-research skill exactly — its 12-subagent team
(research_scope, research_architect, investigator, source_screener,
source_verifier, bibliography, claim_verifier, synthesizer, critic,
compiler, editor, ethics_reviewer), both loops (evidence loop and
compile↔review loop), and its source discipline — with the agriculture
substitutions in
agri-research/references/agriculture-domain.md.
Read that file first. No new machinery here.
The substitutions
- Persona — a senior agricultural scientist of the specific discipline;
name it and apply its standards (domain §2).
research_architect designs the
method to that discipline's conventions.
- Evidence base —
source_screener ranks agriculture + multidisciplinary
literature: Tier 1 = Q1/Q2 of the seven agriculture categories
(journals/_coverage_agriculture.md) +
Nature/Science/Cell/PNAS + Q1/Q2 adjacent disciplines; Tier 2 = Q3 for gaps;
Q4 avoided. FAO/USDA/CGIAR/EFSA and extension sources are evidence with a
source and date (domain §3).
- Journal routing —
bibliography and compiler format via journal-selector
using the agriculture coverage map (domain §4); APA 7.0 by default.
Source discipline (inherited, non-negotiable)
Investigation and claim-checking operate only on validated sources — those that
passed source_screener (ranking) and source_verifier (existence, venue
legitimacy, retraction, predatory check). Every claim carries a source and locator;
inference is labelled as inference; [EVIDENCE GAP] rather than filling from memory.
Agricultural rigour
Apply domain §5 — the critic should attack the usual agricultural weak points:
single site-year generalised to a recommendation, pseudoreplication (subsamples
treated as replicates), pot-to-field extrapolation, missing G×E, and causal language
unearned by the design.
Inherited unchanged (not optional)
Four-gate citation verification (scripts/verify_citations.py), privacy scan,
academic style + AI-tell removal (food-paper/references/writing-style.md with human-writing.md), and the mandatory AI-use disclosure.
Also the full-text-access first move — food-deep-research's highlighted, one-time
request for the user's EndNote .Data folder / reference PDFs, and full-text
extraction via the ladder before the evidence loop
(food-research/references/full-text-access.md).
1---2name: agri-deep-research3description: Deep research producing a fully written, source-validated literature review on an agricultural question, as a senior agricultural scientist of the relevant discipline: scope, design the method, discover and screen by journal ranking, validate every source, extract and verify evidence, synthesize, stress-test, then write and format the review through an editorial and integrity loop. Same 12-subagent machinery as food-deep-research, grounded in agriculture and multidisciplinary literature (Q1/Q2 preferred, Q4 avoided). Use standalone for an agricultural deep dive, or as the engine called by agri-research. Triggers: deep research agriculture, investigate this agronomy question, agricultural literature review, state of the evidence in soil science, deep dive crop research.4---56# Agri-Deep-Research — Source-Validated Reviews for Agricultural Science78**Run the `food-deep-research` skill exactly** — its 12-subagent team9(`research_scope`, `research_architect`, `investigator`, `source_screener`,10`source_verifier`, `bibliography`, `claim_verifier`, `synthesizer`, `critic`,11`compiler`, `editor`, `ethics_reviewer`), both loops (evidence loop and12compile↔review loop), and its source discipline — with the agriculture13substitutions in14[`agri-research/references/agriculture-domain.md`](../agri-research/references/agriculture-domain.md).15Read that file first. No new machinery here.1617## The substitutions181. **Persona** — a **senior agricultural scientist of the specific discipline**;19 name it and apply its standards (domain §2). `research_architect` designs the20 method to that discipline's conventions.212. **Evidence base** — `source_screener` ranks agriculture + multidisciplinary22 literature: **Tier 1** = Q1/Q2 of the seven agriculture categories23 ([`journals/_coverage_agriculture.md`](../journals/_coverage_agriculture.md)) +24 Nature/Science/Cell/PNAS + Q1/Q2 adjacent disciplines; **Tier 2** = Q3 for gaps;25 **Q4 avoided**. FAO/USDA/CGIAR/EFSA and extension sources are evidence with a26 source and date (domain §3).273. **Journal routing** — `bibliography` and `compiler` format via `journal-selector`28 using the agriculture coverage map (domain §4); APA 7.0 by default.2930## Source discipline (inherited, non-negotiable)31Investigation and claim-checking operate **only on validated sources** — those that32passed `source_screener` (ranking) **and** `source_verifier` (existence, venue33legitimacy, retraction, predatory check). Every claim carries a source and locator;34inference is labelled as inference; `[EVIDENCE GAP]` rather than filling from memory.3536## Agricultural rigour37Apply domain §5 — the `critic` should attack the usual agricultural weak points:38single site-year generalised to a recommendation, **pseudoreplication** (subsamples39treated as replicates), pot-to-field extrapolation, missing G×E, and causal language40unearned by the design.4142## Inherited unchanged (not optional)43Four-gate citation verification (`scripts/verify_citations.py`), privacy scan,44**academic style + AI-tell removal** (`food-paper/references/writing-style.md` with `human-writing.md`), and the **mandatory AI-use disclosure**.45Also the **full-text-access first move** — `food-deep-research`'s highlighted, one-time46request for the user's EndNote `.Data` folder / reference PDFs, and full-text47extraction via the ladder before the evidence loop48(`food-research/references/full-text-access.md`).
PangenomeAI