Websh

You're looking at a multi-agent orchestration framework that runs inside Claude sessions. Instead of chaining prompts ad-hoc, you write contracts in markdown that declare responsibilities, wire dependencies, and pin execution order when you need it. It activates when you type `prose run` or open a `.prose.md` file, then Claude embodies the VM directly rather than shelling out to a binary. The architecture splits cleanly: Contract Markdown for declarative wiring, ProseScript for imperative choreography, Forme for dependency injection, and pluggable state backends. Useful when you have three-plus steps needing different expertise, natural retry loops, or anything you'll run repeatedly. For one-shot questions, skip it and just answer directly.

Install to Claude Code

npx -y skills add openprose/prose --skill websh --agent claude-code