agent-evaluation-pro
Agent and LLM evaluation (professional)
Skill text is English; match the user's response language from Cursor User Rules / project rules when applicable.
Use evaluation frameworks (Promptfoo, Braintrust, LangSmith, Arize Phoenix, Langfuse) and academic literature on LLM evaluation as authority. This skill encodes metric design, dataset curation, regression testing, red-teaming, observability, and CI integration for agentic systems. Confirm evaluation goal (correctness, latency, cost, safety), baseline, and dataset size before proposing a pipeline.
Boundary
agent-evaluation-pro owns metric design, dataset curation, regression testing, red-teaming, agent observability (tracing, logging, scoring), and CI evaluation pipelines. It does not own building the agent, prompt engineering depth, or general software testing — combine with ai-integration-pro, prompt-engineering-pro, and testing-pro as needed.
| Skill | When to combine with agent-evaluation-pro |
|---|---|
ai-integration-pro |
When the evaluation is part of building or iterating on the agent itself |
prompt-engineering-pro |
When comparing prompt variants or designing prompt templates for eval |
testing-pro |
General software testing strategy, test pyramid, CI structure |
security-pro |
Red-teaming for safety, jailbreak detection, adversarial testing |
ci-cd-pro |
Pipeline design, deployment gates, artifact management |