Add comprehensive Agent Capabilities Guide to document all available tasks

[Copilot]

Problem

Users asking "what kind of tasks can I do with these agents?" had no single comprehensive resource to discover the full capabilities of the Sherlock multi-agent system. While documentation existed for individual agents (e.g., Refinery Agent Guide, ML Workflow Guide), there was no centralized guide showing all available agents and their actions.

Solution

Created a comprehensive Agent Capabilities Guide (AGENT_CAPABILITIES.md) that serves as a one-stop reference for all agent capabilities in the system.

What's Included

📄 AGENT_CAPABILITIES.md (1,382 lines)

Quick Reference Table - Overview of all 4 agents and their purposes:

• EDA Agent (port 8001) - Exploratory data analysis
• Refinery Agent (port 8005) - Data quality & feature engineering
• Model Agent (port 8002) - Machine learning workflows
• Custom Agent - Custom scripts and integrations

Complete Action Documentation - All 48 agent actions with:

• Clear descriptions and purposes
• Real-world use cases
• JSON examples for immediate use
• 100% coverage verified against config.yaml

Organized by Agent:

• EDA Agent: 5 actions (analyze, clean, transform, explore, preprocess)
• Refinery Agent: 22 actions split into:
  • 7 data quality checks (read-only validation)
  • 15 feature engineering operations (transformative)
• Model Agent: 17 actions covering:
  • Class imbalance & sampling (3 actions)
  • Model training & evaluation (5 actions)
  • Baseline & sanity checks (4 actions)
  • Experiment tracking (5 actions)
• Custom Agent: 4 actions for extensibility

3 Complete Workflow Examples:

1. End-to-end data science pipeline
2. Data quality assessment workflow
3. Model experimentation and comparison

Additional Sections:

• API integration examples with Master Orchestrator
• Best practices (5 key guidelines)
• Configuration guide
• Monitoring & observability setup
• Troubleshooting common issues
• Links to additional resources

📝 README.md Updates

Added prominent Quick Start section that:

• Immediately directs new users to the capabilities guide
• Provides a visual overview with icons for each agent type
• Links directly to AGENT_CAPABILITIES.md

Updated Documentation & Examples section to list the new guide first for easy discovery.

Validation

Created and ran validation script that confirms:

• ✅ All 48 actions from config.yaml are documented
• ✅ 100% coverage across all 4 agents
• ✅ Consistent formatting and structure
• ✅ All examples follow JSON schema conventions

User Impact

Users can now:

1. Discover capabilities - Quickly see what each agent can do
2. Get started fast - Copy-paste working JSON examples
3. Learn workflows - Follow complete end-to-end pipeline examples
4. Follow best practices - Understand recommended usage patterns
5. Troubleshoot issues - Find solutions to common problems

Example Usage

{
  "run_name": "customer_churn_analysis",
  "tasks": [
    {
      "agent": "refinery",
      "action": "comprehensive_quality_report",
      "args": {"data_path": "customer_data.csv", "target_col": "churn"}
    },
    {
      "agent": "refinery",
      "action": "execute_feature_pipeline",
      "args": {"data_path": "customer_data.csv", "run_id": "churn_001"}
    },
    {
      "agent": "model",
      "action": "train_model",
      "args": {
        "experiment_id": "exp_churn_001",
        "model_type": "random_forest"
      }
    }
  ]
}

The guide makes it clear which agent handles which tasks and provides production-ready examples for immediate use.

---

Files Changed:

• AGENT_CAPABILITIES.md (new file, 1,382 lines)
• README.md (updated with Quick Start section and documentation links)

Original prompt

what kind of tasks can i do with these agents?

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.