技能创建器
skill-creator
创建、改进和评估自定义技能,把专业流程转化为可重复使用的 AI 能力。
food-proposal【数据与研究】Master orchestrator for preparing research proposals and funding-application proposals in food, nutrition & agricultural science. Like food-pipeline, but for proposals: it resolves the target funding scheme's format first (headings, order, page/word limit, font, assessment criteria), then drafts from scratch (food-deep-research for the knowledge base…
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
The proposal counterpart of food-pipeline. It does not research, write, or review
itself — it routes the proposal to the specialist skills and enforces the target
funding scheme's format, length limit, and assessment criteria at every gate.
Original work; the orchestration mirrors food-pipeline.
A proposal is written to a funding scheme's format, the way a manuscript is written
to a journal's. Before anything else, resolve the target scheme
(references/scheme-selection.md): ask the user which scheme, then load its spec from
schemes/<id>.md + schemes/schemes.json (required headings in order, page/word limit,
font, assessment criteria). Built-in: arc-discovery-eoi, arc-discovery-full,
uom-major-minor-project (see schemes/INDEX.md). If the scheme isn't built in,
don't guess — ask the user to upload the scheme's guideline / sample form and run
scheme_extractor to capture its format into a new schemes/<id>.md + schemes.json
entry. Every scheme carries a length limit — enforce it with
scripts/proposal_wordcount.py.
proposal_router — reads what the user has (topic only, or an existing draft),
resolves the scheme, picks the entry stage, and assembles each downstream skill's
input. Mirrors food-pipeline's intake router.scheme_extractor — when the scheme isn't built in, extracts its format
(headings, order, limits, font, criteria) from the user's uploaded guideline/sample
into a new scheme spec.compliance_gate — the checkpoint between stages: verifies required headings are
present and in order, the length limit is met (proposal_wordcount.py), the font/
format matches, and the content addresses every assessment criterion.| Stage | Skill / agent | Deliverable | Gate |
|---|---|---|---|
| 0 · SCHEME | proposal_router + scheme resolution | Target scheme spec (headings, limit, criteria) | — |
| 1 · RESEARCH | food-deep-research (or agri-deep-research; food-research) | Knowledge base / evidence for the case | evidence sufficiency |
| 2 · DRAFT | food-paper (or agri-paper) writing to the scheme's headings | Proposal draft in the scheme format | compliance_gate: headings + length + criteria |
| 3 · REVIEW | food-review (or agri-review) as a mock funding panel vs the criteria | Review report + editorial decision | mandatory author decision |
| 4 · REVISE | food-paper (revise) | Revised proposal (tracked changes; original Word opt-in) | issues resolved + still within limit |
| 5 · FINALIZE | food-paper (format-convert) | Submission-ready proposal (.docx/PDF) to scheme format | final compliance |
Slides? For a proposal pitch, food-ppt can turn the finished proposal into a deck.
food-pipeline builds a paper, but
to the scheme's headings and limit.food-research
quick brief / reuse of any provided library, as food-pipeline does before editing),
then edit to fix gaps and fit the scheme — never rewrite blindly.Anti-fabrication grounding + four-gate citations, privacy scan, academic style +
human-writing.md, the mandatory AI-use disclosure (some schemes, e.g. the UoM
project, require stating the AI model and prompts used — put it in Acknowledgements),
and the full-text-access first move for the research stage. Figures via food-figure.