Woodshed Explained

Woodshed is a tool for building and testing AI skills that work with Claude, Anthropic's AI assistant. Think of it as a personal workshop where you can create a skill (like "summarize text in a specific style" or "extract data from messy input"), run it multiple times with different test cases, grade how well it works, and then tweak it based on the results. It's designed for people who want to experiment with Claude without needing a lot of coding knowledge.

Here's how it works in practice: You start by creating a workspace with a command, then you write out your skill as plain text instructions in a markdown file. You also write test cases (prompts) and evaluation criteria (how to judge if the skill worked well). When you run the tool, it feeds your skill and test prompts to Claude, records what happens, then uses another Claude instance to grade the results. All the outputs—the prompts, Claude's responses, grades, and logs—get saved in a results folder so you can review them and learn what went wrong or right.

The ideal user is someone exploring what Claude can do and wants to iterate quickly. For example, a founder might create a skill for customer service replies, test it 10 times with real customer emails, see how well Claude followed the instructions, then refine the instructions and test again. The tool automates the repetitive parts (running Claude, keeping organized records) so you can focus on the creative and strategic work of making the skill better.

One important caveat from the creator: this is early, experimental software built quickly without heavy testing. It can use a lot of Claude API tokens (costing money), and the author explicitly warns that there are rough edges. The tool is meant for people comfortable with alpha-stage software and willing to read through messy results folders to learn from their experiments.