AI Agent Development

Comprehensive patterns and implementation guidance for building AI agents that reason, plan, use tools, and collaborate. Covers the full spectrum from single-agent architectures to multi-agent orchestration, with production-grade error handling, observability, and evaluation.

Effort-level note: when agents run inside loops or parallel fan-out, default the effortLevel to high, not max. Iterative and concurrent runs compound effort-level cost without matching quality gains. See the Effort-Level Strategy section of prompt-engineering/SKILL.md for the decision table. A full Opus 4.7 Anti-Patterns table for agent workloads is maintained alongside this skill.

When to Use This Skill

Use this skill for:

• Designing agent architectures (ReAct, plan-and-execute, reflection)
• Implementing tool-use patterns with function calling or MCP
• Building memory systems (short-term, long-term, episodic, semantic)
• Creating planning and reasoning loops
• Orchestrating multi-agent systems
• Adding guardrails and safety layers to agents
• Evaluating and testing agent behavior
• Instrumenting agents for observability