Agent-First CLI Design

Design CLIs where AI agents are first-class consumers alongside humans. Human DX optimizes for discoverability and forgiveness. Agent DX optimizes for predictability and defense-in-depth. Both can coexist in the same binary, but agent ergonomics must be intentionally designed — they do not emerge from human-first defaults.

When to Use

• Building a new CLI that agents will invoke
• Adding agent support to an existing human-first CLI
• Reviewing CLI designs for agent compatibility
• Wrapping REST/gRPC APIs as CLI tools
• Writing CONTEXT.md, AGENTS.md, or skill files for a CLI

When NOT to Use

• Pure library/SDK design (no CLI surface)
• GUI or TUI-only applications
• Internal scripts not exposed to agents

Core Principle