What problem does it solve? Building AI agents that reliably perform multi-step tasks with tools is difficult: agents often skip planning, misuse tools, lose state across calls, and lack safety guardrails. This Skill provides proven architecture patterns, workflow templates, and framework-specific code for building production-grade agents. ## Core Features & Use Cases - Framework Selection Guidance: Defaults to Pydantic AI for Python (type-safe, validated outputs) and Claude Agent SDK for Node.js/Next.js, with model selection tables for task complexity. - Explore-Plan-Execute-Verify Workflow: A complete agent loop pattern with explicit planning, per-step verification, and self-correction on failure. - Tool Design & Guardrails: Tool definition schemas with risk levels, human-in-the-loop approval for high-risk actions, and multi-layer input/output validation. - Multi-Agent Patterns: Agent-as-tool, handoff, and orchestrator patterns, plus memory management and testing strategies (unit, behavior, evaluation). - Use Case: You need to build a research agent in Python that searches the web, reads pages, and returns structured results. Use the Pydantic AI patterns to define typed tools, structured output models, and streaming responses. ## Quick Start Ask the AI to scaffold a Pydantic AI research agent with web search tools and structured output following the agentic-development patterns.
