AXI
Design principles for agent ergonomics
...The project proposes agent-native CLIs as an alternative to human-oriented command-line tools and structured protocols such as MCP. Its principles cover compact output, minimal schemas, truncation, precomputed aggregates, explicit empty states, structured errors, and noninteractive behavior. It also emphasizes ambient context, useful default output, contextual next steps, and consistent help. Reference implementations cover GitHub operations, browser automation, human review, and local usage monitoring.