Can I apply these patterns to existing projects?

Yes, the skill provides specific strategies for refactoring traditional applications into prompt-native systems, enabling more flexible and autonomous functionality.

Why are atomic tools better than complex features?

Atomic tools empower an agent's reasoning capabilities, allowing it to handle edge cases and solve unexpected problems that fixed, complex code sequences might not anticipate.

How does this skill improve MCP tool design?

It teaches patterns for creating granular, primitive tools that allow agents to compose complex behaviors dynamically, rather than being restricted by rigid, high-level functions.

What is the principle of Action Parity?

Action Parity ensures that every action a human user can perform through a graphical user interface is also available to the agent via its toolset, preventing 'dead ends' in autonomous workflows.

What is an agent-native application?

An agent-native application is software designed where agents are first-class citizens, achieving outcomes through autonomous loops and atomic tools rather than predefined, hardcoded logic paths.

Agent-Native Architecture

Name: Agent-Native Architecture
Author: gvkhosla

bygvkhosla

•

開発者ツール

Architects autonomous software systems where features are emergent outcomes of agents operating through atomic tools and action parity.

This skill provides a comprehensive framework for designing agent-native applications, shifting the paradigm from hardcoded features to agent-driven outcomes. It guides developers in implementing core principles like Action Parity—ensuring agents can perform any task available in the UI—and Tool Granularity, which favors atomic primitives over monolithic functions. By adopting this architecture, you can build self-improving, highly composable systems where new capabilities are unlocked through prompt engineering and emergent agent reasoning rather than traditional code refactoring.

主な機能

01Atomic tool design patterns for maximum agent reasoning flexibility

02Architectural blueprints for self-modifying and emergent system behaviors

0335 GitHub stars

04Action Parity framework for aligning UI capabilities with agent tools

05Composable prompt strategies to define features as agent-driven outcomes

06Guidance for MCP tool development and shared workspace integration

ユースケース

01Developing self-correcting applications that refine their own logic through feedback and prompts

02Designing autonomous file management systems that organize data through reasoning loops

03Building collaborative workspaces where agents have full functional parity with human users

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add gvkhosla/compound-engineering-pi agent-native-architecture

For use in Claude.ai and ChatGPT

Download Skill