Skill Claude CodeCodex
Add a research paper to the Pathology-Wiki knowledge base. Use when the user provides a DOI, arXiv id, PubMed id, journal URL, PDF path, or pasted abstract and asks to ingest, cite, add, extract, or save the paper. Produces all five mandatory artifacts (object YAML, graph entries, Markdown page, SKILL.md, update…
Skill Claude CodeCodex
Add a benchmark node to the Pathology-Wiki knowledge base. Use when the user provides a benchmark name (e.g. SpaPath-Bench, WSI-PFM Benchmark) or a benchmark paper and asks to add or document a benchmark. Defines goal, datasets, tasks, metrics, baselines, and the executable workflows for adding models / datasets /…
Skill Claude CodeCodex
Add a public dataset to the Pathology-Wiki knowledge base. Use when the user provides a dataset homepage / Kaggle / Zenodo / Hugging Face URL or release paper and asks to add a dataset. Captures access info, modalities, organs/diseases, labels, tasks, preprocessing, and known pitfalls. Produces all five mandatory…
Skill Claude CodeCodex
Add a method node to the Pathology-Wiki knowledge base, positioning it in the canonical 8-branch method map. Use when the user provides a method name plus a representative paper (or set of papers) and asks to add, document, or position a method. Produces all five mandatory artifacts per CLAUDE.md §2.
Skill Claude CodeCodex
Add a software library or repository (a "tool") to the Pathology-Wiki knowledge base. Use when the user provides a GitHub URL, package name, or docs URL and asks to add, ingest, or document a tool. Captures install method, capabilities, agent-callability, license, and links to methods / datasets / benchmarks. Produces…
Instructions file
Claude Code instructions for Bokai-Zhao/Pathology-Wiki, covering claude.md, 0. repository state — read first, 1. project identity, 2. the five-artifact rule (the most violated rule) and 3. repository structure.
Skill Claude CodeCodex
The origin paper for attention-based MIL aggregation. Cite as the foundational reference whenever attention-MIL appears in a pathology paper.
Skill Claude CodeCodex
Top-level dispatcher for adding any article to the knowledge base. Classifies the article, then hands off to the type-specific extractor.
Skill Claude CodeCodex
Broad orientation review of deep learning in histopathology (206 refs). Cite as a single background reference for any thesis intro, grant background, or related-work section.
Skill Claude CodeCodex
The clinical-grade scale precedent for weakly-supervised MIL in pathology. Cite when arguing that slide-level labels suffice for clinical-grade accuracy.
Skill Claude CodeCodex
The standard reference for modern ViT-based nuclei segmentation + classification. Cite whenever SAM/DINOv2-pretrained ViT encoders meet HoVer-Net-style decoders.
Skill Claude CodeCodex
Clinical-grade-validated pathology foundation model. 19 cohorts, 32 cancer types, 60k WSIs, 24 hospitals. Cite for cross-cohort generalisation and clinical breadth.
Skill Claude CodeCodex
The standard reference for modern weakly-supervised MIL in pathology. Cite whenever a pipeline uses the mahmoodlab/CLAM toolkit or CLAM aggregator.
Skill Claude CodeCodex
Extract clinical-focus fields from an article that primarily defines clinical pathology / diagnostic criteria / biomarkers / cohort design.
Skill Claude CodeCodex
Architectural foundation of every modern pathology vision-language model. Cite when introducing image-text contrastive learning in any pathology context.
Skill Claude CodeCodex
Modern default pathology vision-language foundation model. ViT-B/16 + text Transformer, CoCa-style dual objective on textbook/paper image-caption pairs. Successor to PLIP.
Skill Claude CodeCodex
The pre-UNI open-source pathology SSL reference. Cite whenever a pipeline uses CTransPath features or when documenting the evolution of pathology PFMs.
Skill Claude CodeCodex
Dual-stream MIL aggregator with SSL pretraining. Standard sibling baseline alongside ABMIL / CLAM / TransMIL on Camelyon16-type WSI benchmarks.
Skill Claude CodeCodex
Sparse deformable-attention MIL aggregator. DETR-style sparse keys instead of full self-attention. Sibling to TransMIL in the Transformer-MIL family.
Skill Claude CodeCodex
Pseudo-bag construction + double-tier feature distillation MIL. Particularly strong in low-slide-count regimes. Standard sibling reference to ABMIL / CLAM / TransMIL / DSMIL.
Skill Claude CodeCodex
First published slide-level pathology foundation model. Cite when discussing slide-level pretraining or two-stage patch-then-slide PFM recipes.
Skill Claude CodeCodex
The canonical reference for joint nuclei instance segmentation + classification on H&E. Cite whenever HV-decoder or three-branch nuclei architectures appear.
Skill Claude CodeCodex
Benchmarks 4 SSL recipes on pathology data and releases the Lunit-DINO ViT-S/16 checkpoint (TCGA + TULIP). Established DINO as the canonical pretraining recipe for pathology PFMs.
Skill Claude CodeCodex
First widely-cited multistain slide-level pathology PFM. Cross-stain contrastive pretraining (H&E + IHC + special stains) yields stain-invariant slide embeddings.