Lx050
User书生国智科探挑战赛 · 赛道六 AI for Social Science · 古代碑帖与拓片 — 可复现证据链科研 Skill 系统:12 个离线、确定性、纯标准库 Skill(哑舍小分队 · 西北大学)
Categories
Indexed Skills (13)
rubbing-to-knowledge
碑帖与拓片研究的可复现证据链工作流。12 个可独立运行的 Skill 组成一条链：材料登记与哈希、确定性图像 baseline、OCR 候选与逐字坐标回链、人工审校交接、文本结构化、文献证据定位、假说反思、诚信审计、知识网络与时间轴/关系路径查询，由主编排 Skill 管理研究状态与阶段门。纯 Python 标准库、完全离线、不需要任何凭证或云服务：一条命令即可复算全部测试（包内实测 12 Skills / 835 项测试 / status pass）。核心差异不在修图或识字，而在把观察 OBS、派生 DER、来源 SRC、推断 INF、假说 HYP 结构性分开：机器与 AI 输出默认只是候选，任何研究主张都必须回链到原图字位或文献定位，失败、冲突、反证与阻断都是正式产物而非被删掉的噪声。当前状态如实说明：知识网络与文本结构化只有合成通道可跑，真人审校未完成，本作品不产出任何历史结论、释文或准确率数字。
epigraphy-knowledge-network
Turn located, human-adjudicated epigraphic text and located literature evidence into an auditable knowledge graph of candidate mentions, human-adjudicated assertions and evidence-backed edges, with deterministic timeline, relationship-path and entity queries. Use when Codex must build, verify, query or export a rubbing-derived knowledge graph without letting OCR, AI proposals, name matches, modern place names, calendar conversions or graph paths become historical facts.
epigraphy-text-structure
Validate qualified epigraphic transcriptions; draft append-only, evidence-linked proposals for punctuation, annotation, character normalization, metadata, people, places, offices, era dates, relations, and ambiguities. Use when structuring a human-reviewed rubbing transcription, a located source-edition transcription, or an explicitly synthetic test fixture without treating OCR, AI suggestions, normalized characters, modern places, calendar conversions, or entity identities as fact.
research-hypothesis-reflection
Review a ResearchCase v1 after evidence or experiment feedback arrives. Use when a research workflow must detect supporting-versus-challenging evidence conflicts, negative/null/invalid results, dependent evidence, or stalled inquiry; preserve prior hypotheses in a versioned ReflectionMemo; and choose only continue, reroute, rollback, human_review, or stop_unresolved.
research-integrity-audit
Audit the ResearchCase v1 publication gate for missing evidence, HYP or rejected support, unresolved conflicts, unadjudicated machine output, unaccountable adjudication records, broken hash bindings, sticky blockers, evidence-layer confusion, and vacuous completion attempts. Use before packaging or publishing claims, after claim-evidence links change, or when a workflow needs a machine-readable pass/warn/block decision.
research-literature-evidence
Build an append-only, auditable path from preregistered literature searches to precisely located evidence candidates while keeping snippets, OCR, AI interpretations, self-attested human review, synthetic fixtures, and unclear rights from becoming academic or public evidence. Use for books, papers, catalogues, webpages, inscriptions, and other research sources when queries, discoveries, failures, editions, carriers, locators, excerpts, conflicts, and review gates must remain reproducible.
research-provenance-ledger
Register local research materials in an append-only provenance ledger, capture SHA-256, size, MIME, source, license, and rights status, and reverify originals or copies without executing them. Use when a research workflow needs auditable material intake, duplicate-ID protection, integrity checks, or explicit hash-mismatch reports before OCR, analysis, publication, or packaging.
research-quarantine-promotion
Safely assess one already verified quarantine JSON as bounded inert data, using a preregistered exact binding, an exclusive source-verification receipt, same-buffer identity and strict parsing, content-free structure reports, and offline producer verification. Use when research-source-acquisition has preserved an application/json asset in quarantine and a later internal workflow needs an auditable B0 gate before any IIIF, OCR, schema, literature, or knowledge-graph interpretation; do not use for arbitrary JSON, URL following, semantic validation, canonical materialization, academic admission, or public release.
research-run-ledger
Execute approved research commands without a shell and preserve tamper-evident command events, exit status, separate stdout/stderr artifacts, exact input/output snapshots, and verified append-only output checkpoint chains. Use when a research workflow, experiment, data conversion, growing JSONL ledger, model invocation, validation, or multi-Skill handoff must be reproducible and auditable at command level.
research-source-acquisition
Safely preflight, acquire, and reverify one already rights-approved online research asset into an immutable quarantine run with exact URL approval, HTTPS allow-listing, SSRF-resistant DNS and redirect checks, streaming limits, MIME/magic validation, SHA-256, and auditable manifests. Use when a reviewed碑帖、拓片、古籍影像、权利页面或馆藏元数据 URL must be fetched for internal research or a later approved pipeline; do not use for open-web discovery, arbitrary URL fetching, scraping, or bulk download.
rubbing-image-preparation
Inspect and deterministically preprocess digitized rubbings, steles, copybooks, and inscription images before OCR or scholarly comparison. Use when Codex must measure image quality, reject unsafe or unsupported image inputs, preserve an immutable parent hash, create grayscale/contrast/polarity baselines, compare preprocessing variants, or register non-generative derived assets without presenting enhancement as recovered historical evidence.
rubbing-ocr-review
Run auditable OCR baselines on registered rubbing, stele, copybook, and inscription images, preserving empty or failed results, and create append-only, actor-declared review decisions and pending human-review handoffs. Use when comparing OCR across an original and deterministic derivatives, binding readings to image hashes and pixel regions, retaining uncertain or rejected readings, separating AI prelabels from declared-human review, or handing OCR evidence to text-structure workflows.
rubbing-research-orchestrator
Orchestrate provenance-first research on rubbings, steles, inscriptions, and related historical sources. Use when Codex must initialize a ResearchCase, split an epigraphy question into auditable stages, route work across independent Skills or Agents, react to positive/negative/null/invalid results, enforce research gates, or decide whether to continue, reroute, request human review, stop unresolved, or package a study.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.