Can I use different LLMs for different parts of the agent process?

Yes, the skill provides a Multi-Model Architecture pattern that allows you to route planning to fast models and complex reasoning or coding to more powerful models.

Does this skill include patterns for browser automation?

Yes, it covers tool design patterns for browser interaction, including opening URLs, clicking elements, and capturing screenshots for agent observation.

How does it handle AI safety and permissions?

It includes patterns for multi-level permissions ranging from 'Auto' for low-risk tasks to 'Ask Each' for high-risk commands like terminal execution or file deletion.

What is the core agent loop provided in this skill?

The skill utilizes a Think-Decide-Act-Observe cycle where the LLM reasons about a task, decides on a tool call, executes the action, and observes the result to inform the next iteration.

Autonomous Agent Patterns

Name: Autonomous Agent Patterns
Author: claudiodearaujo

byclaudiodearaujo

0•

개발자 도구

Provides architectural blueprints and implementation patterns for building safe, effective autonomous coding agents and tool-calling systems.

This skill offers a comprehensive framework for designing autonomous AI agents, drawing inspiration from industry leaders like Cline and OpenAI Codex. It covers essential architectural components including robust agent loops (Think-Decide-Act-Observe), multi-model selection strategies, and standardized tool schemas for file operations and terminal execution. Crucially, it prioritizes safety with detailed patterns for granular permission management, sandboxed execution, and human-in-the-loop approval workflows, making it an indispensable resource for developers building the next generation of AI-driven development tools.

주요 기능

010 GitHub stars

02Multi-model orchestration patterns for specialized task handling

03Granular permission systems and safety sandboxing patterns

04Pre-defined tool schemas for file system and terminal operations

05Standardized Agent Loop Architecture (Think-Decide-Act-Observe)

06Human-in-the-loop approval workflow implementations

사용 사례

01Implementing multi-step reasoning workflows with safety guardrails

02Designing secure tool-calling APIs for LLM-powered applications

03Building a custom autonomous coding assistant or IDE plugin

주요 기능

010 GitHub stars

02Multi-model orchestration patterns for specialized task handling

03Granular permission systems and safety sandboxing patterns

04Pre-defined tool schemas for file system and terminal operations

05Standardized Agent Loop Architecture (Think-Decide-Act-Observe)

06Human-in-the-loop approval workflow implementations

사용 사례

01Implementing multi-step reasoning workflows with safety guardrails

02Designing secure tool-calling APIs for LLM-powered applications

03Building a custom autonomous coding assistant or IDE plugin