What is the core principle of agent-native architecture?

The core principle is that features are defined in prompts rather than code. You describe the desired outcome and provide primitive tools, letting the agent determine the best way to achieve that outcome.

When is it inappropriate to use a prompt-native approach?

Avoid this approach for high-frequency operations requiring thousands of calls per second, scenarios requiring 100% deterministic output, or extremely cost-sensitive environments.

What is action parity in agent-native design?

Action parity is the discipline of ensuring that every action a human user can perform via the UI has a corresponding tool that the AI agent can use to perform the same action.

How should tools be designed in this architecture?

Tools should be primitives (like read_file or write_file) that provide general capabilities rather than encoding specific business logic or complex workflows into the code.

Agent-Native Architecture

Name: Agent-Native Architecture
Author: RBozydar

byRBozydar

API 开发

Implements a prompt-native engineering philosophy where AI agents achieve outcomes using primitive tools and natural language definitions rather than hard-coded workflows.

关于

The Agent-Native Architecture skill enables developers to build sophisticated AI systems by inverting traditional software patterns. Instead of writing rigid code for an agent to execute, this skill provides the framework to define outcomes in prompts and provide agents with 'primitive' tools (like file system access or generic API calls) to figure out the implementation independently. It includes comprehensive guidance on Model Context Protocol (MCP) design, self-modifying code patterns, and maintaining action parity between users and agents to ensure high autonomy and flexibility.

主要功能

Dynamic runtime context injection for long-running sessions
Prompt-native feature definition and outcome-based logic
Model Context Protocol (MCP) primitive tool design
Action parity mapping between UI capabilities and agent tools
Self-modification and autonomous code evolution patterns
0 GitHub stars

使用场景

Building autonomous systems that can refactor and evolve their own codebase
Creating flexible MCP servers that provide low-level primitives to LLMs
Transitioning legacy hard-coded workflows into dynamic, prompt-driven AI features

关于

主要功能

Dynamic runtime context injection for long-running sessions
Prompt-native feature definition and outcome-based logic
Model Context Protocol (MCP) primitive tool design
Action parity mapping between UI capabilities and agent tools
Self-modification and autonomous code evolution patterns
0 GitHub stars

使用场景

Building autonomous systems that can refactor and evolve their own codebase
Creating flexible MCP servers that provide low-level primitives to LLMs
Transitioning legacy hard-coded workflows into dynamic, prompt-driven AI features