Back
Post
Where it started
I started Ready for Agents while looking for a way to earn from worklore.dev, my library of developer stories that coding agents can run.
The most useful stories there turned out to be about things the docs don't say. Docs fall behind when a library ships an update, and even a perfect docs writer can't foresee every mistake people and agents make. So the moment a library ships a breaking change is the moment a worklore story about it helps most.
I went looking for libraries with recent breaking changes. While reading their docs, I got stuck on another question: what is the best way to get docs to a coding agent? An llms.txt file, a skill, a docs MCP server, or a few lines in AGENTS.md?
So I tested it: Claude Code, Codex and Gemini, three libraries with a real recent break, each channel switched on one at a time, 135 graded runs. A short brief in AGENTS.md was the only channel with a significant effect. The docs MCP helped only when an agent called it, and Codex never did.
The next question was whether library teams want the same check for their own package. Ready for Agents runs it on typical tasks for your library and shows which version the agents pick, whether the code passes hidden tests, which docs channel they opened and what each run cost.
The most useful stories there turned out to be about things the docs don't say. Docs fall behind when a library ships an update, and even a perfect docs writer can't foresee every mistake people and agents make. So the moment a library ships a breaking change is the moment a worklore story about it helps most.
I went looking for libraries with recent breaking changes. While reading their docs, I got stuck on another question: what is the best way to get docs to a coding agent? An llms.txt file, a skill, a docs MCP server, or a few lines in AGENTS.md?
So I tested it: Claude Code, Codex and Gemini, three libraries with a real recent break, each channel switched on one at a time, 135 graded runs. A short brief in AGENTS.md was the only channel with a significant effect. The docs MCP helped only when an agent called it, and Codex never did.
The next question was whether library teams want the same check for their own package. Ready for Agents runs it on typical tasks for your library and shows which version the agents pick, whether the code passes hidden tests, which docs channel they opened and what each run cost.
👋 Join WIP to participate