integrable-lattice-harvest
Project-specific literature harvesting for integrable-lattice. Use when Codex needs to collect candidate papers, metadata, abstracts, identifiers, and source availability for unresolved finite-symbol
SKILL.mdintegrable-lattice/skills/integrable-lattice-harvest/SKILL.md
GitHubIntegrable Lattice Harvest
Goal
Collect a broad but traceable paper set for the integrable-lattice project. The output is not a bibliography; it is raw material for statement extraction.
Required Context
Before deciding queries or scope, read:
CLAUDE.md/AGENTS.md(mandatory before starting) — the repository's top-priority conventions: proof form (structured text understructured-latex/; never author new Typst), naming rules, verification commands, the definition of "done", and "文書・定理を番号や記号で管理しない" (never manage documents, theorems, or classifications by number or symbol — call them by name).integrable-lattice/README.mdandintegrable-lattice/docs/tasks/auto-loop-runbook.md— project goal and the auto-loop procedure.docs/discussion/対数順序群上の統計力学/anddocs/discussion/可算性の効用/— primary sources for the countable core (ℕ⊂ℚ⊂Λ⊂ℚ̄) and for treating ℝ/ℂ as a convenience for applications; the countable/ℝ boundary is the decidability boundary.docs/research/R-Lambda-duality/— the ℝ-side / Λ-side duality and what counts as a closed form on the countable side.docs/research/場の量子論の数学的定式化/— which layers of existing physics close algebraically and which escape to ℝ.integrable-lattice/README.mdintegrable-lattice/inputs/seeds/models.mdintegrable-lattice/inputs/seeds/operations.mdintegrable-lattice/inputs/seeds/axes.mdintegrable-lattice/inputs/queries/seed-queries.md
Workflow
- Identify the target slice: model family, operation type, boundary condition, parameter regime, or seed paper.
- Expand search phrases from the seed files. Preserve exact phrases for reproducibility.
- Collect metadata from stable sources first: arXiv, DOI pages, OpenAlex, Semantic Scholar, publisher pages.
- Prefer arXiv records when available because later extraction can use LaTeX source.
- Record enough metadata to deduplicate papers: title, authors, year, arXiv id, DOI, URL, abstract, source availability.
- Save curated query strings under
integrable-lattice/inputs/queries/. - Save harvested metadata under
integrable-lattice/inputs/corpus/when it is small and curated. For large corpora, save only manifest/query instructions.
Output Shape
Use this minimal record shape in markdown, YAML, JSON, or JSONL:
id:
title:
authors:
year:
source:
arxiv_id:
doi:
url:
abstract:
model_hints:
operation_hints:
source_available:
notes:
Guardrails
- Do not call a paper relevant only because it contains generic terms like "integrable" or "lattice"; connect it to at least one seed model or operation.
- Do not judge novelty here. Mark only retrieval facts and coarse relevance.
- Keep failed or noisy queries when they explain search coverage.