AI Agent Design

Operational playbook for designing production LLM agents in 2026. Covers agent loop patterns (ReAct, Reflexion, Plan-and-Execute, function calling, Anthropic tool use), tool schema engineering (JSON Schema, strict mode, description craft, anti-patterns), the Agent Skills open standard (Anthropic Dec 2025), Model Context Protocol (MCP) integration, multi-agent orchestration (LangGraph, CrewAI, AutoGen — with cost trade-offs), error handling and infinite-loop prevention, observability via OpenTelemetry GenAI semantic conventions, the OWASP LLM Top 10 v2025 and OWASP Top 10 for Agentic Applications 2026, and the context-engineering discipline that replaced 'prompt engineering' in 2026. Use when designing a new agent, picking an orchestration framework, debugging a misbehaving agent, or hardening tool exposure. Output: agent specs with named patterns, cost projections, observability hooks, and OWASP-mapped risk surfaces.

Muvon Updated 6 repo stars

File contents

Muvon/octomind-tap commit 8daae957b7

Frequently asked questions

npx get-skill add muvon/ai-agent-design