What problem does it solve? Autonomous AI agents often look impressive in demos but fail in production because compounding error rates destroy reliability — a 95% per-step success rate drops to roughly 60% by step ten. This Skill provides the architectural patterns and hard-won lessons needed to build agents that work consistently rather than agents that merely look capable. ## Core Features & Use Cases - Agent Loop Patterns: Implement ReAct (alternating reasoning and action) and Plan-Execute (separating planning from execution) architectures with clear guidance on when each applies. - Reliability Engineering: Apply guardrails, hard cost limits, step-count reduction, and structured logging to contain compounding errors before they reach production. - Anti-Pattern Detection: Identify and avoid unbounded autonomy, blind trust in agent outputs, and general-purpose autonomy traps. - Use Case: When designing an agent that researches topics and writes reports autonomously, use this Skill to decompose the goal into validated steps, add reflection checkpoints, and set cost and step limits before deployment. ## Quick Start Ask the agent to help you design a reliable autonomous agent architecture for your specific task, including which loop pattern to use and what guardrails to apply.
