// the find
poteto/how
skill for explaining architecture
A Cursor plugin that adds a `/how` slash command for getting architectural explanations of code. You ask how something works, it explores the codebase and returns a structured writeup — or, if you ask for a critique, it spawns multiple critic agents across different models. Useful for onboarding onto unfamiliar codebases or pressure-testing a design.
The fan-out / synthesis pattern for complex questions is genuinely smart — decomposing into parallel explorer subagents and reconciling findings avoids the single-threaded context problem that kills most code-explanation tools. The automatic routing between simple (single-agent) and complex (fan-out) questions means you're not paying the overhead on trivial asks. The critique mode using multiple independent models rather than one is a real differentiator — you get adversarial disagreement, not consensus slop. The prompt files are plain Markdown and fully editable, so you can tune the rubric without touching any code.
The whole repo is prompt files — there is no code, no tests, and no way to verify any of this actually works except by trying it in Cursor. That's a significant adoption risk. The plugin is Cursor-specific; anyone not on Cursor gets nothing from this. It depends entirely on the quality of whatever model Cursor is routing to, and the prompts have no fallback or error handling — if an explorer subagent returns garbage, the synthesis agent silently incorporates it. At 38 forks and no contribution activity, this looks like a personal tool that was open-sourced but isn't actively maintained.