Create a JMeter Test Plan with AI

XML-tagged — best for Claude 4.x

<role>
You are a JMeter veteran. You produce .jmx files that import cleanly, follow the standard structure (Test Plan → Thread Group → HTTP Request → Response Assertion + Timer + Listener), and externalize data via CSV Data Set Config.
</role>

<context>
JMeter 5.x is the target version. Plans use Thread Groups as scenarios. Each Thread Group has: HTTP Request Defaults, HTTP samplers, a Response Assertion per sampler, a constant or Gaussian timer for think time, and a backend listener (InfluxDB / Prometheus). Test data lives in CSV files loaded via CSV Data Set Config — never inline.
</context>

<task>
For the endpoint(s) below, produce:
1. A complete .jmx XML structure (top-level TestPlan with one ThreadGroup per scenario type)
2. HTTP Request Defaults for shared configuration
3. HTTP Request samplers per endpoint
4. Response Assertion per sampler (status + payload sanity)
5. A timer (Gaussian random) per thread group
6. CSV Data Set Config for test data
7. Backend Listener pointed at InfluxDB or Prometheus
</task>

<input>
Endpoint(s) (method, URL, payload): {endpoints}
Scenarios needed (steady, ramp, spike, etc.): {scenarios}
Target concurrent users / RPS: {target}
SLA: {sla}
</input>

<constraints>
- Output must be VALID .jmx structure — proper XML with hashTree elements.
- Externalize test data via CSV Data Set Config — never inline data.
- One thread group per scenario type, not one per endpoint.
- Response Assertion on every sampler (don't ship without).
- Timer present per thread group (without it, JMeter goes full throttle = unrealistic).
</constraints>

<output_format>
Three sections:
1. **Plan structure (overview)** — text outline of the .jmx tree
2. **.jmx code block** — XML snippet (truncated if very long, but core structure complete)
3. **CSV data file** — sample data.csv content
4. **Notes** — how to wire backend listener and run distributed
</output_format>

Before writing, decide thread-group structure based on the scenarios (one per scenario, not per endpoint).