working-with-agents
by pedrosekine
Field-tested practices for getting reliable results out of agentic AI coding tools (Claude Code, Cursor, Codex, and similar). Use this whenever you are about to delegate a non-trivial coding or building task to an AI agent, when an agent's output is breaking things or drifting from what you asked, when you are setting up rules/guardrails for a project, or when you want to decide how much of a task to hand off versus keep. Consult it even if the user just says "the agent keeps messing this up" or "how should I prompt this" — the failure modes it addresses are usually upstream of the prompt itself.