What is the Codex Agent SDK?

It is Anthropic's framework for building autonomous AI agents programmatically, allowing developers to use the same tools, agent loops, and context management that power Claude's core capabilities.

How do I handle tool permissions within the SDK?

The SDK offers five built-in permission modes including 'acceptEdits', 'dontAsk', and 'bypassPermissions', as well as the ability to implement custom 'canUseTool' callbacks and event hooks.

Does this skill support both Python and TypeScript?

Yes, it covers the complete API for both TypeScript (v0.2.76+) and Python (v0.1.48+), providing examples and implementation patterns for both languages.

Can I build multi-agent systems with this skill?

Absolutely. The SDK supports subagents with isolated contexts, restricted toolsets, and specific model overrides, enabling complex hierarchical orchestration and cost optimization.

Anthropic Codex Agent SDK

Name: Anthropic Codex Agent SDK
Author: liminal-ai

byliminal-ai

0•

API開発

Build and orchestrate autonomous AI agents using Anthropic's programmatic Codex framework for TypeScript and Python.

This skill empowers developers to design, implement, and debug sophisticated multi-agent systems using the Codex Agent SDK. It provides comprehensive guidance on the core query API, built-in system tools like Read, Edit, and Bash, and advanced features such as MCP tool integration, event hooks for behavior modification, and granular permission management. Whether you are building isolated subagents or complex automated workflows, this skill offers the implementation patterns and best practices needed to leverage Anthropic's agent loop programmatically within your own applications.

主な機能

0116+ lifecycle hooks for intercepting and modifying agent behavior

02Native Model Context Protocol (MCP) tool integration

03Hierarchical subagent orchestration with isolated contexts

040 GitHub stars

05Structured output support via Zod and JSON Schema

06Programmatic query() API for asynchronous agent loops

ユースケース

01Creating multi-agent workflows for complex software engineering tasks

02Building custom automated coding assistants and programmatic agents

03Developing secure, sandboxed tools for AI to interact with local filesystems

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add liminal-ai/code-steward claude-agent-sdk-skill

For use in Claude.ai and ChatGPT

主な機能

0116+ lifecycle hooks for intercepting and modifying agent behavior

02Native Model Context Protocol (MCP) tool integration

03Hierarchical subagent orchestration with isolated contexts

040 GitHub stars

05Structured output support via Zod and JSON Schema

06Programmatic query() API for asynchronous agent loops

ユースケース

01Creating multi-agent workflows for complex software engineering tasks

02Building custom automated coding assistants and programmatic agents

03Developing secure, sandboxed tools for AI to interact with local filesystems

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add liminal-ai/code-steward claude-agent-sdk-skill

For use in Claude.ai and ChatGPT