跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
搜索与研究

food-pipeline

Master orchestrator for the whole food & nutrition research-to-publication workflow. Coordinates the specialist skills — each with its own subagent set — into one governed path: journal selection, research (food-research / food-deep-research), writing & analysis (food-paper), figures (food-figure), peer review (food-review), revision, and finalization. Use when the user wants the entire process managed end to end, or a project routed to the right skills with quality gates. Triggers: run the full paper workflow, take this from research to submission, manage the whole project, research to publication, end-to-end paper, orchestrate my paper.

PangenomeAI · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add PangenomeAI/academic-skills-food-nutrition --skill food-pipeline

技能说明


name: food-pipeline description: "Master orchestrator for the whole food & nutrition research-to-publication workflow. Coordinates the specialist skills — each with its own subagent set — into one governed path: journal selection, research (food-research / food-deep-research), writing & analysis (food-paper), figures (food-figure), peer review (food-review), revision, and finalization. Use when the user wants the entire process managed end to end, or a project routed to the right skills with quality gates. Triggers: run the full paper workflow, take this from research to submission, manage the whole project, research to publication, end-to-end paper, orchestrate my paper." metadata: version: "2.2.0" verified: "2026-07" related_skills: [journal-selector, food-research, food-deep-research, food-paper, food-figure, food-review] subagents: [intake_router, quality_gate] references: - references/pipeline-state-machine.md - references/quality-gates.md - references/mode-advisor.md

Food-Pipeline — Master Research-to-Publication Orchestrator

The top-level conductor. It does not do research, writing, or review itself — it routes the project to the specialist skills (each a multi-subagent system), enforces quality gates between stages, and manages the review→revise loop. Original work.

Skills it orchestrates (each brings its own subagent team)

  • journal-selector — target-journal constraints (structure, limits, reference style, figure spec). A shared procedure, not an installed skill: load journal-selector/SKILL.md and follow it.
  • food-research — literature/evidence synthesis (quick brief / full review / systematic PRISMA + OHAT). Use food-deep-research instead for an open-ended, source-validated deep dive or a standalone literature review.
  • food-paper — whole-process manuscript system (field → questions → data/stats → figures → argument → draft → polish → self-review).
  • food-figure — submission-grade figures at the journal spec (invoked within food-paper).
  • food-review — multi-reviewer peer-review panel + formatting compliance.

Own subagents

  • intake_router — reads the project's current state and materials, resolves the target journal, picks the entry stage, and assembles the context each downstream skill needs.
  • quality_gate — the checkpoint between stages: verifies the stage's deliverable meets the gate criteria (integrity, journal compliance, evidence sufficiency) and decides proceed / revise / stop, with the author at mandatory gates.

Stages

StageSkill / agentDeliverableGate
0 · ROUTEintake_router + journal-selectorEntry point + journal constraints—
1 · RESEARCHfood-research (or food-deep-research)Evidence brief / gap list / (systematic report)evidence sufficiency
2 · WRITEfood-paperDraft: analysis, figures (food-figure), argument, referencesintegrity + journal compliance
3 · REVIEWfood-reviewReview & Response Report (.docx) — feedback + editorial decision — plus margin comments on the manuscript (when Word tooling available)mandatory author decision
4 · REVISEfood-paper (revise)Revision + response entries — tracked changes on the original Word only if the author authorizesissues resolved
5 · RE-REVIEWfood-review (re-review)Only if the author authorizes a second round — verifies the revision; may add new commentsaccept / stop (no auto third round)
6 · FINALIZEfood-paper (format-convert) + writerSubmission-ready manuscript (.docx) + the one Review & Response Report (.docx); optionally an editable .pptx deck via food-ppt if the author wants slidesfinal compliance

Knowledge reuse — don't research the same field twice

When the pipeline ran Stage 1 (it entered at Stage 0 or 1), the field has already been searched and synthesized. Stages 3 · REVIEW and 5 · RE-REVIEW must therefore carry the Stage-1 evidence base into food-review rather than let its knowledge_builder repeat a full literature search:

  • Pass forward the Stage-1 output (food-research / food-deep-research: validated sources, evidence matrix / synthesis, grading, gap list) as the review's field-knowledge foundation; don't re-fetch what Stage 1 already validated.
  • Top it up with the food-research quick brief stream to find the field's key review publications and read those reviews in full (state of the art, consensus vs contested, standard methods, benchmark ranges).
  • Knowledge base = Stage-1 knowledge + key-review knowledge. The manuscript's own cited sources are still read and audited (Pathway A), reusing Stage-1 records where the source was already retrieved.

If Stage 1 did not run (entry at Stage 2/3 with a finished draft), there is nothing to inherit — food-review builds its knowledge base the full way (Pathway A + B). Using food-review standalone is unaffected by this rule. See food-review/agents/knowledge_builder.md.

Review & revision defaults (Stage 3 onward) — explicit authorization

Default: one review→revise round, then FINALIZE. Do not auto-run a second round or silently rewrite the author's original Word file.

Ask once (consolidate) before Stage 4 when a .docx (or equivalent) is in play:

  1. Second review round? Default no. Run Stage 5 (RE-REVIEW) only if the author explicitly authorizes it. Hard cap remains 2 rounds total.
  2. Edit the original Word with Tracked Changes? Default no. Only modify the original manuscript in place when the author explicitly authorizes it. Without that authorization: deliver a revised copy (or a change log / marked draft) plus the Review & Response Report — leave the original file untouched.

Deliverables — exactly two files, both Word (.docx)

The pipeline produc

阅读完整来源 ↗