neeeophytee/finding-unknowns-skills

finding-unknowns-skills is a set of 8 installable skills for AI coding assistants like Claude Code and OpenAI Codex, built around one idea: your prompt is like a map, and the real codebase is the territory it describes. The gap between what you wrote and what is actually true is your unknowns, and the earlier you surface them, the cheaper they are to fix. The project turns that idea, taken from a public essay by an Anthropic team member, into ready to use commands. The 8 skills are grouped by when you would use them. Before you start building, blindspot-pass looks for things you have not considered in an unfamiliar area, brainstorm-prototypes generates throwaway variations so you can react to options instead of describing taste you cannot put into words, interview-me asks you one question at a time starting with the ones most likely to change your architecture, reference-hunt searches for existing code that already solves your problem, and implementation-plan produces a plan that leads with your riskiest decisions. While building, implementation-notes logs every place you deviate from the plan. After building, pitch-packager bundles your spec, prototype, and notes into a document for reviewers, and change-quiz gives you a comprehension quiz to pass before merging. Installing is meant to be simple. A single command using Vercel's skills CLI auto-detects whichever agent you use, including Claude Code, Cursor, Codex, Copilot, and Gemini, and installs the skills in the right place for each. It can also be installed as a Claude Code plugin, copied manually into a project's skills folder, or added as a CLAUDE.md or AGENTS.md file for agents that read passive instructions instead of individual commands. The project also documents specific setup steps for Codex and for Kimi Code CLI. It is released under the MIT License, with credit given to the original essay the ideas are drawn from.