What problem does it solve? Traditional software hardcodes features and workflows, leaving agents as thin wrappers that can only execute pre-built functions. This Skill guides you to build agent-native applications where agents operate in loops with atomic tools, achieving outcomes you describe in prompts rather than code you write. ## Core Features & Use Cases - Architecture Design: Apply five core principles (Parity, Granularity, Composability, Emergent Capability, Improvement Over Time) with a pre-implementation checklist covering tool design, workspace patterns, and context injection. - Tool & Prompt Design: Build atomic MCP primitives with full CRUD, dynamic capability discovery, explicit completion signals, and system prompts that inject runtime app state. - Testing & Refactoring: Run outcome-based capability tests, automated action-parity audits, and migrate existing code toward prompt-native behavior. - Use Case: When building a notes app, instead of writing a classify_and_organize_files function, you expose read_file, write_file, and move_file primitives plus a prompt describing the organization outcome, letting the agent handle edge cases with judgment. ## Quick Start Ask the agent to help design an agent-native architecture for your application and choose one of the thirteen intake topics such as tool design, action parity, or testing.
