What does 'Parity' mean in agent-native design?

Parity is the principle that an agent should be able to achieve any outcome that a user can perform through the UI. If a user can click a button to tag a file, the agent must have a tool to perform that same action.

What is an agent-native application?

An agent-native application is software designed where agents are central participants. Instead of hard-coding every feature, developers provide agents with atomic tools and define features as outcomes achieved through autonomous loops.

How do agent-native apps improve over time without code changes?

They improve through accumulated context, where the agent learns user preferences, and through prompt refinement, where behavior is tuned by editing natural language instructions rather than refactoring backend code.

Why is tool granularity important in this architecture?

Granularity ensures tools are simple primitives (like read/write). This empowers the agent to use judgment to compose these primitives into complex solutions, allowing for emergent capabilities that weren't explicitly programmed.

Agent-Native Architecture

Name: Agent-Native Architecture
Author: jayteealao

byjayteealao

0•

개발자 도구

Design and build applications where autonomous agents are first-class citizens using atomic tools and prompt-driven features.

This skill provides a comprehensive framework for building 'agent-native' software, shifting the focus from hard-coded features to outcomes achieved by autonomous agents operating in loops. By following core principles like UI-agent parity and tool granularity, developers can create flexible systems where new capabilities are defined through prompts rather than code changes. This approach enables emergent behaviors, where agents solve problems the developer didn't explicitly anticipate, and facilitates self-improving architectures that evolve through context accumulation and prompt refinement.

주요 기능

01Architectural patterns for autonomous agent-in-the-loop systems

02Principles for designing atomic MCP (Model Context Protocol) tools

03Guidelines for building self-modifying and self-improving software systems

04Strategies for achieving parity between UI actions and agent capabilities

05Frameworks for prompt-based feature composition and emergent behavior

060 GitHub stars

사용 사례

01Building an autonomous file management system that organizes workspaces based on high-level user objectives.

02Designing a project management tool where the agent can perform every action available to the human user via tools.

03Refactoring traditional SaaS applications to allow agents to automate complex multi-step workflows via atomic primitives.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jayteealao/htbase agent-native-architecture

For use in Claude.ai and ChatGPT

Download Skill