Audit an agent harness

Design and audit an AI agent harness

This skill is a provider-neutral guide for designing, auditing, and troubleshooting the system around an AI agent, including its loop, tools, memory, permissions, and prompts. It applies to any domain, from coding to support, finance, and operations, and covers topics such as context compaction, MCP connectors, evals, observability, and safety. You ask it for an MVP blueprint, a review of an existing harness, or an explanation of a pattern. It targets OpenAI, Anthropic, and compatible APIs.