This skill from the Strands Agents project helps a coding agent write evaluations with the strands-agents-evals Python package. It covers setting up cases and experiments, choosing evaluators such as output quality, trajectory, helpfulness, faithfulness, tool selection or goal success, scoring from traces in CloudWatch, OpenSearch, OpenInference or LangChain, simulating users and tools, and finding the root cause of failures. You need the strands-agents-evals package installed. You get working evaluation code and test cases for your agent.
Build AI agents in Python with Pydantic AIThis official skill from the Pydantic team guides a coding agent through building agents with the Pydantic AI framework. It covers creating an agent, adding tools and capabilities such as thinking and web search, structured output, streaming, workspaces where the agent runs commands or edits files, defining agents from YAML, multi-agent handoffs, hooks and observability with Logfire. It also shows how to test agent behavior. The framework changes quickly, so check the skill matches the version you have installed.AI agent framework
Evaluate an ADK agent with agents-cliThis skill runs and reads evaluations for Google ADK (Agent Development Kit) agents, through the agents-cli eval commands. It walks a loop of preparing a dataset, running the eval, studying failures, and fixing the agent, then repeating. It also helps pick metrics, including LLM-as-judge scoring, and lists usual reasons a score drops.AI agent frameworkAgentsEvaluation
Pick the cheapest AI model that passes your testsThis skill helps you choose a model for one AI workflow with a small test set instead of by impression. You collect 20 to 50 real examples with correct answers, decide pass or fail rules first, run candidate models from smallest to largest, and record pass rate, cost per 100 runs and latency. It then picks the cheapest model within an agreed margin of the best and writes a routing rule, such as escalating to a larger model when confidence is low. You get a results table per workflow and a routing rule to re-test when models or prices change.AI agent framework
Security best practices (OpenAI)Performs language/framework-specific security best-practice reviews when explicitly requested.AI agent framework
Build Azure AI agent apps from Microsoft CAIRA referencesMicrosoft's caira skill is the entry skill for coding agents drawing on the CAIRA repository as reference material for generative AI apps. The agent asks a few questions about what you want and what you already run, then picks the smallest matching reference parts and adapts them into your project. It targets Azure AI Foundry and Azure OpenAI-compatible endpoints, along with agent frameworks, APIs and front ends. It needs access to the microsoft/caira repository.AI agent frameworkAzure
Use AgentSquad for on-device multi-agent Swift apps2FastLabs' agent-squad-swift skill guides a coding agent working on an iOS or macOS app that uses AgentSquad, a framework for running several AI agents on the device. It explains when to use the orchestrator, the agent types, classifier routing, OpenAI-compatible model clients, native and MCP tools, local storage, tracing and live voice. It is a map rather than an API reference, so the agent is told to read exact signatures in the framework source.AI agent frameworkSwiftiOS
