AGENTS.md patterns for context engineering in coding agents: safer command output, token efficiency, validation, and prompt-injection resistance.
