In this tutorial, we build a terrible version of Claude Code using the all-powerful Effect library. You’ll see how to set up an "agentic harness" (a.k.a., a while loop), connect to Claude with @effect/ai, define schema-based tools, handle multiple tool calls, and switch from safe stub handlers to fully live, “dangerous” filesystem operations—culminating in a demonstration of Claude autonomously laying waste to my file system.
00:00 Introduction to Building a Clawed Code
00:11 Setting Up the Environment
00:57 Implementing User Input and LLM Interaction
01:43 Creating the AI Chat Object
03:57 Defining and Using Tools
06:58 Handling Tool Calls and Responses
09:19 Adding More Tools
12:25 Testing and Final Implementation
13:49 Conclusion and Final Thoughts
