histai-skillsets

HistAI Skillsets

A skills repository providing clean and secure integration between AI coding agents (Claude Code, OpenAI Codex, Gemini CLI) and the HistAI computational pathology platform — covering the Whole Slide Image Data Hub, the CellDX AI Autopilot model trainer, training job monitoring, and slide-level AI analysis.

Overview

This repository bundles four complementary skills:

• 🔬 Cohort Builder (Data Hub) — search pathology cases by diagnosis, organ, age, stains, and clinical criteria; build research cohorts; export whole slide images (WSI) with comprehensive metadata; manage costs with volume-based pricing and selective slide downloads.
• 🤖 CellDX AI Autopilot (Model Trainer) — submit slide classification training jobs on Azure GPU compute, with a full pipeline (parameter tuning → multi-strategy comparison → best model → deployment) and a quick mode for fast single-model iteration.
• 📈 Training Monitor — monitor running training jobs with real-time epoch-by-epoch metrics, stop/adjust/resume training, compare strategy results, and deploy trained models as widgets.
• 🧪 Slide Analyzer — upload your own WSIs to "My Cases", install AI widgets (public store or custom-trained), run them on slides, and retrieve results including segmentation masks.

Security & Access Requirements

To use these skills, you must:

1. Have an active subscription on our CellDX platform
2. Have 2FA authentication enabled on your CellDX account
3. Generate an API Key from Profile Settings → API Keys

Integration

This repository is compatible with Claude Code, OpenAI Codex, and Gemini CLI. Choose your preferred agent below:

Claude Code

Quick Install (Recommended)

1. Register this repository as a plugin marketplace:

   /plugin marketplace add histai/skillsets
   

2. Install all skills at once:

   /plugin install histai-skillsets
   

   This is the preferred way — the skills are designed to work together (e.g. build a cohort → train a model → monitor training). Installing individually is possible but not recommended.

3. Configure your API key:

   • Generate an API key from CellDX Platform → Profile Settings → API Keys
   • Set it as an environment variable:

     export CELLDX_API_KEY="your-api-key"
     

4. Start using:

   • Example: "Search for breast cancer cases with HER2 IHC staining"
   • Example: "Train a classifier on my breast cancer cohort"
   • Example: "How is my training job going?"
   • Example: "Upload this slide and run the melanocytic widget on it"
   • Example: "List my cases and show which widgets are installed"

Manual Installation (Alternative)

If you prefer local installation:

1. Clone this repository:

   git clone https://github.com/histai/skillsets.git
   cd skillsets
   

2. Register as a local plugin:

   /plugin add-local
   

   Point to the repository root directory.

OpenAI Codex

Codex automatically detects skills via the agents/AGENTS.md file.

1. Clone the repository:

   git clone https://github.com/histai/skillsets.git
   cd skillsets
   

2. Verify skills are loaded:

   codex --ask-for-approval never "Summarize the current instructions."
   

3. Configure API key:

   export CELLDX_API_KEY="your-api-key"
   

For more details, see the Codex AGENTS guide.

Gemini CLI

This repository includes gemini-extension.json for Gemini CLI integration.

1. Install from GitHub:

   gemini extensions install https://github.com/histai/skillsets.git --consent
   

   Or install locally:

   git clone https://github.com/histai/skillsets.git
   cd skillsets
   gemini extensions install . --consent
   

2. Configure API key:

   export CELLDX_API_KEY="your-api-key"
   

3. Start using:

   • The Gemini CLI will automatically load all four skills
   • Interact with the HistAI Datahub through natural language

See Gemini CLI extensions docs for more help.

Available Skills

🔬 Cohort Builder

Search pathology cases by diagnosis, organ, age, and stains. Filter datasets (benign/malignant, cancer types), build research cohorts, and export whole slide images with clinical and technical metadata.

Full documentation: skills/cohort_builder/SKILL.md

🤖 CellDX AI Autopilot (Model Trainer)

Submit slide classification training jobs on Azure GPU compute. Supports a full pipeline (parameter tuning → multi-strategy comparison → best model → deployment) and quick mode for fast single-model iteration.

Full documentation: skills/ai_model_trainer/SKILL.md

📈 Training Monitor