What is agent-native architecture?

It is a design philosophy where software features are outcomes achieved by agents using atomic tools in a loop, rather than traditional hardcoded functions.

What is the principle of parity in this context?

Parity ensures that any action a human user can perform through a traditional UI, an agent can also achieve using its available set of tools.

Can I use this for MCP tool development?

Yes, the skill includes specific guidance on building primitive and domain tools that follow Model Context Protocol (MCP) best practices for granularity.

How does this skill improve Claude's behavior?

It provides Claude with specific frameworks and checklists to ensure any system it builds is optimized for agentic autonomy, tool parity, and composability.

Does this skill help with self-modifying code?

Yes, it provides patterns for creating systems where agents can safely refine their own prompts or code logic based on feedback and accumulated context.

Agent-Native Architecture

Name: Agent-Native Architecture
Author: gvkhosla

bygvkhosla

•

生产力与工作流

Builds autonomous applications where features are outcomes achieved by agents operating in a continuous loop.

This skill provides a comprehensive framework for designing and implementing agent-native systems, treating software agents as first-class citizens rather than secondary scripts. It leverages five core principles—parity, granularity, composability, emergent capability, and iterative improvement—to help developers build applications where complex features are defined through prompts rather than hardcoded logic. By emphasizing a shared workspace and atomic tool-based interaction, it enables the creation of self-modifying systems and autonomous workflows that adapt to user needs and discover latent demand through usage patterns.

主要功能

01Promotes atomic tool design for high composability

02Enables emergent capabilities through outcome-oriented prompts

03Ensures parity between UI actions and agent capabilities

04Supports the development of self-modifying system architectures

05Provides structured checklists for agent-native architecture reviews

0635 GitHub stars

使用场景

01Developing complex MCP tools that enable agents to perform multi-step tasks

02Refactoring existing codebases to support agentic workflows and tool parity

03Architecting new autonomous agent systems from the ground up

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