What is preflight-spec-skill?

This project is a small add-on skill for Claude Code, the AI coding assistant, meant to fix a common problem: developers often type vague requests like fix duplicate drafts without spelling out what they actually mean, and the AI ends up guessing, sometimes solving the wrong problem or making changes that are broader than intended. The skill works as a step that runs before any real coding happens. First it figures out what kind of task is being asked for, such as a bug fix, a new feature, a refactor, a frontend change, an API change, a database migration, or an investigation. Then, only if genuinely needed, it asks up to three clarifying questions rather than a long questionnaire. From there it writes a short specification covering the goal, what is explicitly out of scope, criteria for knowing the work is done, which files are likely involved, and possible risks. Finally, it requires some way to verify the change actually worked, such as a test that fails before the fix and passes after. Importantly, the README is explicit about what this is not: it is not a library of prompts, not a wrapper that replaces Claude Code itself, not a substitute for writing tests, and not meant to interrogate the user before every tiny edit. Small, low-risk changes should not trigger this extra process. Installing it is simple: you copy a single file, SKILL.md, into wherever your Claude Code setup keeps its skills, whether that is project level or user level. The README includes worked examples covering a bug fix, a frontend layout issue, a refactor, an API change, and a database migration, each showing the kind of questions and spec the skill would produce. This tool is aimed at developers who use Claude Code regularly and want more disciplined, spec-first behavior out of their coding sessions, especially on tasks where a wrong guess could cost real rework. It is released under the MIT license, allowing free use including for commercial purposes.