Agent Example: Visit Preparation Summary

Why This Matters

Patients accumulate health data across multiple systems: EHR exports, wearable devices, and self-reported supplements. Before a doctor visit, this data is scattered and unstructured. An agent that can read a Cascade Pod and produce a concise, provenance-annotated summary saves time for both patient and physician.

Key challenges this example addresses:

• ◾
  Provenance differentiation — Clinical data from an EHR has different reliability characteristics than device-generated or self-reported data. The summary must reflect this distinction.
• ◾
  Data completeness assessment — Identifying what is missing is as important as summarizing what is present. Are immunizations up to date? Are lab results recent?
• ◾
  Trend extraction — Transforming 30 days of raw vitals into actionable insights that a physician can act on in a 15-minute visit.

Prerequisites

The reference pod contains a complete clinical and wellness dataset for Alex Rivera, 52M, Portland, OR, including medications, conditions, allergies, lab results, vitals, wellness data, and immunizations.

Option A: MCP via Claude Desktop

Step 1. Add the Cascade MCP server to your Claude Desktop configuration:

{
  "mcpServers": {
    "cascade": {
      "command": "cascade",
      "args": ["serve", "--mcp"],
      "env": {
        "CASCADE_POD_PATH": "/path/to/cascadeprotocol.org/reference-patient-pod"
      }
    }
  }
}

Step 2. Restart Claude Desktop, open a new conversation, and paste the contents of agent-prompt.md.

Step 3. The agent will use cascade_pod_query with --all to read the full Pod in a single call.

Option B: CLI-as-Skill via Claude Code

Step 1. Open a terminal in this directory and run Claude Code:

export CASCADE_POD_PATH="../reference-patient-pod"
claude

Step 2. Paste the contents of agent-prompt.md as your prompt. The agent will invoke the Cascade CLI across all data categories:

# Read everything at once
cascade pod query ./reference-patient-pod --all --json

# Or read category by category
cascade pod query ./reference-patient-pod --medications --json
cascade pod query ./reference-patient-pod --conditions --json
cascade pod query ./reference-patient-pod --allergies --json
cascade pod query ./reference-patient-pod --labs --json
cascade pod query ./reference-patient-pod --vitals --json
cascade pod query ./reference-patient-pod --wellness --json
cascade pod query ./reference-patient-pod --immunizations --json
cascade pod info ./reference-patient-pod --json

What the Agent Should Do

1. Read the complete Pod (all clinical and wellness data)
2. Generate a structured visit summary containing: medication list with provenance annotations, recent lab results with interpretation, vital sign trends from both clinical and device sources, and open questions based on data gaps
3. Distinguish clinical-source data from device-source data throughout the summary
4. Identify action items for the upcoming visit

See expected-output.md for the reference answer. The output should be formatted as a concise document a physician could read in under two minutes.

Files in This Example

Key Cascade Protocol Concepts Demonstrated