Pathology Skills Collection

Comprehensive clinical pathology toolkit for Claude - 15 specialized skills for surgical pathology QA, template generation, staging, coding, clinical workflow optimization, research integrity, and digital-pathology tool guidance.

🌐 [pathologyskills.patoloji.dev](https://pathologyskills.patoloji.dev/)  ·  Related: [Pathology Report Checker](https://reportskill.patoloji.dev/) (live demo · Hugging Face)

What This Provides

A modular collection of Claude skills for clinical pathologists, QA teams, and researchers:

• ✅ Compliance checking against CAP and ICCR guidelines
• 📋 Synoptic template generation with optional pre-fill
• 🔢 TNM staging calculation (AJCC 8th edition)
• 🏥 SNOMED CT / ICD-O-3 coding suggestions
• 📊 Tumor board summaries for MDT meetings
• 🔄 Free-text to synoptic conversion for legacy reports
• 🔍 Scientific similarity checking for academic integrity / plagiarism screening
• 📚 Reference verification (4-level citation auditor) for manuscripts
• 📐 Statistical methods review for research articles (9-aspect rubric)
• 🔬 QuPath guidance for digital-pathology workflows and Groovy scripting

Supported Tumor Types

• 🎀 Breast invasive carcinoma (CAP Breast.Invasive)
• 🔬 Colorectal resection (CAP ColoRectal)
• 🫁 Pancreas exocrine carcinoma (CAP Panc.Exo)
• 🫀 Gastric carcinoma (CAP Stomach)

Prerequisites

Required: Claude Code CLI

# Verify installation
claude --version

Optional (for batch processing): Python 3.8+ with dependencies:

pip install -r requirements.txt

Installation

Pick the option that matches how your colleagues use Claude.

Option 1: Plugin Marketplace — Recommended for Claude Code users

Works in Claude Code (terminal CLI, VS Code extension, or JetBrains plugin). No cloning, no scripts, two slash commands inside a Claude Code session:

/plugin marketplace add patolojiAI/pathology-skills-collection
/plugin install pathology-skills@pathology-skills-collection

That installs all 15 skills at once.

Updating later:

/plugin update          # pull the latest version from the marketplace
/reload-plugins         # apply the update to the current session

/plugin update only downloads — it does not activate the new version in the current session. Run /reload-plugins afterwards to pick up changes to skills, agents, hooks, and plugin MCP servers without restarting Claude Code. A fresh Claude Code session loads the updated version automatically, so /reload-plugins is only needed if you want the changes immediately in your current session.

Option 2: Clone + install.sh — for Claude Code users who prefer scripts

git clone https://github.com/patolojiAI/pathology-skills-collection.git
cd pathology-skills-collection
./install.sh

# Test
claude "What stage is pT2 N0 M0 for breast using tnm-stage-calculator"

Symlinks all 15 skills into ~/.claude/skills/. Uninstall with ./uninstall.sh.

Option 3: Claude.ai App (browser / desktop) — upload .skill files

Each release publishes a standalone .skill file for every skill (self-contained zip, no shared dependencies). Upload one through the Claude.ai web app or the Claude desktop app:

1. Right-click the skill you want in the table below → Save Link As… to

download the .skill file. (Direct-clicking starts the download for some browsers but may try to "open" it for others.)

1. In claude.ai, open Settings → Capabilities → Skills → Upload skill and

select the downloaded file. Repeat for each skill.

All links point at the latest release automatically — they don't need updating when a new version is published.

Download all 15 from the [Releases page](https://github.com/patolojiAI/pathology-skills-collection/releases/latest), or pick individual files:

No terminal, no git, no admin rights needed. Updates: download the new release's .skill file and re-upload.

Older alternative (still works): the Claude.ai Project setup walkthrough — paste SKILL.md into a Project's custom instructions. Use this if your account doesn't yet have the Settings → Capabilities → Skills upload UI.

Option 4: Community Marketplace (auto-discovery after 5⭐)

Once this repo reaches 5+ GitHub stars it's auto-listed on claudemarketplaces.com within 24 hours, which gives a browse-and-install UI. The two-line /plugin marketplace add ... in Option 1 works right now without waiting.

How to Use

Single Report Analysis

# Compliance checking
claude "Check this breast report using breast-pathology-specialist" < report.txt

# TNM staging
claude "What stage is pT3 N1b M0 for colorectal using tnm-stage-calculator"

# Template generation
claude "Generate blank Whipple template using pathology-template-generator"

# Tumor board summary
claude "Create tumor board summary using pathology-tumor-board-summary" < report.txt

# SNOMED coding
claude "SNOMED code for invasive ductal carcinoma grade 2 using pathology-coder"

Batch Processing Multiple Reports

✅ RECOMMENDED: Use Claude CLI directly (no API key needed)

# Process Excel file with 100 reports - Claude analyzes each with LLM intelligence
claude "Read reports.xlsx and use breast-pathology-specialist to analyze each report. Provide compliance scores and export results to Excel."

# Process all reports in a directory
claude "Analyze all .txt files in reports_dir/ using pathology-compliance-checker and create summary Excel."

# Use the helper script
./batch_process_cli.sh reports.xlsx colorectal-pathology-specialist results/

⚠️ ALTERNATIVE: Python API script (requires ANTHROPIC_API_KEY)

Only use this if you DON'T have Claude CLI or need programmatic integration:

# Requires: export ANTHROPIC_API_KEY="your-key-here"
pip install -r requirements.txt
python scripts/process_skill.py --skill breast-pathology-specialist reports.xlsx

🚫 DON'T: Ask Claude to "generate a Python script" - it will create regex-only code without LLM intelligence!

Skills

15 skills in four groups: clinical workflow tools for daily report QA (including the all-in-one pathology-report-checker), all-in-one specialist skills that handle a tumor type end-to-end, research-integrity skills for manuscripts and statistical reviews, and digital-pathology tooling for software platforms used in practice.

All 15 skills auto-detect English and Turkish input. The four tumor specialists share the four core clinical workflows (compliance, staging, templates, summaries) with the single-purpose skills — pick the specialist when you want one skill to own a whole case, or the single-purpose skills when you want them usable across tumor types.

Clinical workflow — single-purpose

#### 1. pathology-compliance-checker

Validates surgical pathology cancer reports against CAP (College of American Pathologists) and ICCR (International Collaboration on Cancer Reporting) protocols. Performs element-completeness checking, severity-weighted scoring (Critical −15, Major −5, Minor −2 per missing element), and automatic cross-validation of pT vs tumor size, pN vs node count, margins vs R classification, and stage-group consistency.

• Triggers: "check compliance", "validate against CAP", "score this report", "audit this synoptic", "find missing elements", "verify pT/pN consistency"
• Output: compliance score 0–100, missing-element list, cross-validation flags, prioritized recommendations
• Tumor types: breast, colorectal, pancreas, gastric · Languages: EN / TR

claude "Check this report using pathology-compliance-checker" < report.txt

#### 2. pathology-template-generator

Generates blank or pre-filled CAP-compliant synoptic report templates with every required, conditional, and recommended data element listed. Useful for standardizing department reporting, training residents on synoptic structure, or pre-filling templates before dictation.

• Triggers: "generate a synoptic template", "create a CAP report skeleton", "blank breast lumpectomy template", "Whipple specimen template", "pre-fill template"
• Output: markdown synoptic template with [___] placeholders or pre-filled values
• Specimen types: mastectomy, lumpectomy, colectomy, LAR, APR, Whipple, distal pancreatectomy, gastrectomy, EMR/ESD, polypectomy, sentinel / axillary LND · Languages: EN / TR

claude "Generate blank Whipple template using pathology-template-generator"
claude "Pre-fill breast template with 2.3cm Grade 2 IDC using pathology-template-generator"

#### 3. tnm-stage-calculator

Fast TNM stage-group calculation from pT, pN, pM categories using AJCC 8th edition criteria. Flags inconsistencies (tumor-size-vs-pT, node-count-vs-pN), validates the pT/pN/pM combination actually exists in AJCC 8th, and gives brief prognostic context.

• Triggers: "what stage is pT2 N1 M0", "calculate TNM stage", "stage group for pT3 N1b breast", "is pT2 N0 M0 stage I or II"
• Output: stage group (0, I, IIA, IIB, IIIA, IIIB, IIIC, IV) with one-line prognostic note
• Tumor types: breast, colorectal, pancreas, gastric (uses pathologic stage groups, AJCC 8th 2017)

claude "What stage is pT2 N1a M0 for breast using tnm-stage-calculator"

#### 4. pathology-coder

Suggests SNOMED CT and ICD-O-3 morphology / topography codes for pathology diagnoses, procedures, and biomarkers. Includes hierarchical parent / child relationships and synonyms so you can pick the right specificity for LIS integration, tumor-registry reporting, or billing support.

• Triggers: "SNOMED code for [diagnosis]", "ICD-O-3 code for [tumor type]", "code this diagnosis", "morphology code for"
• Output: SNOMED CT code + preferred term + hierarchy; ICD-O-3 morphology/behavior + topography
• Coverage: breast, colorectal, pancreas, gastric, prostate, lung, lymphoma, GIST + common procedures and biomarkers

claude "SNOMED code for invasive ductal carcinoma grade 2 using pathology-coder"

#### 5. pathology-tumor-board-summary

Distills full pathology reports into concise 3–5 line summaries for multidisciplinary tumor board (MDT) meetings. Extracts diagnosis, TNM stage, margins, lymph nodes, and key biomarkers — only the data points that drive MDT decisions.

• Triggers: "create a tumor board summary", "MDT summary", "oncology consult summary", "3-line summary for the MDT"
• Output: 3–5 line paragraph or bulleted summary
• Use cases: MDT preparation, oncology consults, referral letters, case presentations

claude "Create tumor board summary using pathology-tumor-board-summary" < report.txt

#### 6. report-converter

Two-in-one skill: converts free-text narrative pathology reports into structured CAP synoptic format and generates professional amendments, addenda, or corrections for previously-issued reports. Helps clean up legacy reports and supports proper documentation when biomarker results arrive late or errors need correcting.

• Triggers: "convert to synoptic", "convert to CAP format", "structure this report", "generate an amendment", "add an addendum", "issue a corrected report"
• Output: CAP synoptic version of the original report OR a professional addendum / amendment block
• Coverage: breast, colorectal, pancreas, gastric, general carcinomas

claude "Convert to synoptic format using report-converter" < narrative_report.txt
claude "Create addendum for HER2 results using report-converter" < original_report.txt

Clinical workflow — all-in-one specialists

The specialist skills combine compliance, staging, templates, biomarkers, summaries, and coding for a specific tumor type into a single skill. Pick a specialist when you want one tool to handle every part of a case end-to-end.

#### 7. breast-pathology-specialist

Comprehensive breast cancer pathology workflow combining compliance checking (CAP Breast.Invasive, ICCR), TNM staging (AJCC 8th), synoptic templates, biomarker reporting (ER / PR / HER2 / Ki-67 per ASCO/CAP guidelines), tumor board summaries, and SNOMED coding.

• Triggers: "check breast report", "validate breast pathology", "stage this breast cancer", "interpret ER/PR/HER2", "generate breast template"
• Output: depends on request — score, stage, biomarker interpretation, template, or summary
• Specimens: mastectomy, lumpectomy, wide local excision, re-excision, sentinel LN biopsy, axillary LND · Languages: EN / TR

claude "Check this breast report using breast-pathology-specialist" < report.txt

#### 8. colorectal-pathology-specialist

Comprehensive colorectal cancer pathology workflow with compliance checking (CAP ColoRectal, ICCR), AJCC 8th staging, MSI/MMR testing interpretation (MLH1, PMS2, MSH2, MSH6 patterns and Lynch-syndrome implications), MERCURY mesorectal-excision quality grading (complete / nearly complete / incomplete), tumor board summaries, and SNOMED coding.

• Triggers: "check colorectal report", "validate CAP ColoRectal", "stage this colorectal cancer", "interpret MSI/MMR", "assess TME quality"
• Output: score, stage, MMR interpretation, TME grade, template, or summary
• Specimens: colectomy, low anterior resection, abdominoperineal resection, polypectomy, transanal excision · Languages: EN / TR

claude "Check this colorectal report using colorectal-pathology-specialist" < report.txt

#### 9. pancreas-pathology-specialist

Comprehensive pancreatic cancer pathology workflow for exocrine pancreas carcinoma. Covers Whipple / distal-pancreatectomy specimen dissection guidance, axial slicing protocol, the six margin assessments (SMA, SMV, posterior, anterior, pancreatic neck, bile duct), AJCC 8th staging (CAP Panc.Exo, ICCR), compliance scoring, and SNOMED coding.

• Triggers: "check pancreas report", "validate CAP Panc.Exo", "stage this pancreatic cancer", "assess Whipple margins"
• Output: score, stage, margin interpretation, R classification, template, or summary
• Specimens: Whipple (pancreaticoduodenectomy), distal pancreatectomy, total pancreatectomy, biopsy · Languages: EN / TR

claude "Check this Whipple report for all margins using pancreas-pathology-specialist" < report.txt

#### 10. gastric-pathology-specialist

Comprehensive gastric cancer pathology workflow covering Lauren classification (intestinal / diffuse / mixed), WHO histologic typing, HER2 IHC / FISH interpretation per ASCO/CAP, MSI/MMR status, AJCC 8th staging (CAP Stomach, ICCR), compliance scoring, tumor board summaries, and SNOMED coding.

• Triggers: "check gastric report", "validate CAP Stomach", "stage this gastric cancer", "interpret Lauren", "interpret HER2 IHC/FISH"
• Output: score, stage, HER2 interpretation, Lauren type, template, or summary
• Specimens: total / subtotal / proximal gastrectomy, esophagogastrectomy, EMR / ESD · Languages: EN / TR

claude "Check this gastric report, ensure Lauren type specified using gastric-pathology-specialist" < report.txt

Research integrity

These three skills work on research manuscripts, not clinical reports. They save their outputs as markdown files in the current working directory (HTML on request) so you can review them outside the chat.

#### 11. scientific-similarity-checker

Academic-integrity / literature-discovery tool. Takes a scientific article (PDF, image, or plain text) and searches PubMed, OpenAlex, Semantic Scholar, IEEE, arXiv, Crossref, and the web for similar work. Detects similarity by abstract content, topic overlap, author network, and journal / venue, then issues tiered warnings (🔴 likely misconduct · 🟡 needs investigation · 🟢 normal practice · ℹ️ informational) for self-plagiarism, duplicate publication, salami slicing, or related-work blind spots.

• Triggers: "check this paper for similar work", "duplicate publication check", "misconduct screening", "novelty check", "plagiarism screen", "find the authors' other work"
• Output: <author>-<year>-<title>.md (similarity report) saved in cwd, optional HTML, plus the verdict badge in chat
• Use cases: pre-submission self-check, journal pre-screening, integrity review, literature surveillance

claude "Use scientific-similarity-checker on this paper" < manuscript.pdf

#### 12. reference-verifier

Four-level citation auditor for manuscripts.

Catches fake or hallucinated references and citations that misrepresent the source's findings.

• Triggers: "audit references", "verify citations", "check the bibliography", "generate a .bib file", "detect fake references", "verify citation context"
• Output: <stem>_audit.md markdown audit + <stem>_refs.bib cleaned BibTeX (HTML optional), saved in cwd
• Best for: biomedical / PubMed-indexed manuscripts, but works for any discipline

claude "Audit references using reference-verifier" < manuscript.pdf

#### 13. statistical-methods-reviewer

Critically reviews the statistical methods of a research article against a 9-aspect rubric scored 0–2 each (total /18):

1. Design–method alignment
2. Assumptions & diagnostics
3. Sample size & power
4. Multiplicity control
5. Model specification & confounding
6. Missing-data handling
7. Effect sizes & confidence intervals
8. Validation & calibration
9. Reproducibility & transparency

Also checks the paper against a curated red-flag catalogue (chi-square with expected counts <5, stepwise regression without validation, Cox PH never tested, AUC alone with no calibration, immortal-time bias, dichotomized continuous predictors, etc.) and recommends concrete alternative analyses where the rubric flags issues. Cross-references CONSORT / STROBE / STARD / TRIPOD / PRISMA / REMARK / ARRIVE compliance when applicable.

• Triggers: "review the stats", "check the statistical methods", "audit the analysis", "are these tests correct", "score this paper's statistics"
• Output: <slug>-stats-review.md saved in cwd (HTML optional). Includes Overall Rating (🟢 Robust 15–18 / 🟡 Moderate 8–14 / 🔴 Weak 0–7), scored rubric table, red-flag list, and recommendations table

claude "Review the stats in this paper using statistical-methods-reviewer" < paper.pdf

Digital pathology tooling

A guidance skill (not a report-processor) that helps you actually drive the software platforms used in digital pathology.

#### 14. qupath-guide

Expert guidance for QuPath, the open-source digital pathology platform (qupath.github.io). Covers four pillars: troubleshooting (errors, "Run for Project does nothing", performance), UI / navigation (tools, panels, keyboard shortcuts, preferences), analysis workflows (cell detection, tissue / pixel classification, TMA dearrayer, IHC scoring for Ki-67 / HER2 / PD-L1, multiplex / mIF), and Groovy scripting (batch processing, custom measurements, data export, automation). Adapts depth to the user's level — from "new to QuPath" to extension developer. When version-specific details matter, fetches official QuPath docs (qupath.readthedocs.io, javadoc, forum.image.sc). Targets QuPath 0.5.x and 0.6.x.

• Triggers: "QuPath", .qpproj / .qpdata files, "Run for Project", "stain vectors", "TMA dearrayer", "detection objects", "Brightfield (H-DAB)", "QuPath cell detection", "IHC scoring", "Groovy script for slides", "QuPath error"
• Output: menu paths, step-by-step workflows, ready-to-run Groovy snippets, and links to authoritative docs when the local references don't have enough detail
• Use it for: anything QuPath. Don't use it for ImageJ/Fiji-only questions or standalone Python deep-learning (StarDist / Cellpose outside QuPath)

claude "Use qupath-guide: my Run for Project script does nothing, no errors"
claude "Use qupath-guide: walk me through Ki-67 scoring on an IHC slide"

#### 15. pathology-report-checker

The all-in-one clinical skill: a single skill that does the whole report-QA loop end to end — CAP/ICCR compliance scoring with severity weighting, pTNM staging validation (AJCC 8th), blank/pre-filled synoptic templates, a 3–5 line tumor-board summary, SNOMED CT / ICD-O-3 coding, free-text → synoptic conversion, and amendment/addendum drafting. Self-contained (bundles its own references), and the same skill is published standalone at patolojiAI/pathology-report-checker-skill. It complements rather than replaces pathology-compliance-checker — reach for the report-checker when you want one skill to own a whole case, or the single-purpose skills when you want each step on its own.

• Triggers: "check this report for CAP compliance", "verify staging", "generate synoptic template", "tumor board summary", "convert to synoptic", "calculate TNM stage", "code with SNOMED"
• Tumor types: breast invasive carcinoma, colorectal resection, exocrine pancreas, gastric carcinoma (EN/TR)
• Use it for: an end-to-end pass on a single report. Don't use it for non-cancer, cytology, or unsupported tumor types

claude "Check this pancreas Whipple report for CAP/ICCR compliance" < report.txt

Documentation

• [Getting Started Guide](docs/getting-started.md) - Installation and setup
• [Claude App Setup (browser/desktop)](docs/CLAUDE_APP_SETUP.md) - 5-min recipe for colleagues who don't use the CLI
• [Quick Reference](docs/QUICK_REFERENCE.md) - 1-page trigger phrases and reference card
• [Features](docs/FEATURES.md) - Full feature documentation
• [Workflow Guide](docs/WORKFLOW.md) - Step-by-step compliance workflow
• [Batch Processing Guide](docs/BATCH_PROCESSING.md) - How to process multiple reports correctly
• [Local LLM Setup](docs/LOCAL_LLM_SETUP.md) - Ollama/LM Studio setup for offline use
• [Examples](docs/examples.md) - Common use cases
• [Multi-Format Guide](docs/MULTI_FORMAT_GUIDE.md) - PDF, Excel, images support
• [Skills Reference](docs/pathology-skills.md) - Detailed skill documentation
• [Contributing Guide](docs/CONTRIBUTING.md) - How to contribute

Project Structure

pathology-skills-collection/
├── pathology-skills/          # 15 individual skills
├── shared-references/         # Common guidelines (TNM, codes, templates)
├── scripts/                   # Python tools (batch, single, watch, file readers)
├── samples/                   # Sample reports with expected outputs
├── examples/                  # Example inputs/outputs
├── docs/                      # Documentation
├── install.sh                 # Installation script
├── uninstall.sh              # Uninstallation script
├── requirements.txt           # Python dependencies
├── README.md                  # This file
├── CLAUDE.md                  # Development guide
└── LICENSE                    # MIT license

Contributing

Contributions welcome! See docs/CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

Support

• Project website: pathologyskills.patoloji.dev
• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Related: Pathology Report Checker — live demo · Hugging Face

Built for clinical pathologists by pathologists 🔬