What is shipcheck?

Shipcheck is a skill for Claude Code, the AI coding assistant, that acts as a stand-in for a senior QA engineer reviewing a code change before it ships. Once installed, you can say something like shipcheck this branch before I merge, and Claude will scope the size of the change, then run two reviewers at the same time: one that hunts for regressions, meaning things the change might break that used to work, and one that checks whether the change actually does what it claims, including edge cases. The two reviews are combined into a single verdict of SIGN OFF, SIGN OFF WITH NITS, or DO NOT SHIP. The tool is entirely read only. It never edits files, makes commits, runs database migrations, or touches outside services. It only reads your code and runs checks you already have, such as a type checker, a linter, your test suite, and git diff, to gather evidence. What sets it apart from a typical review my code prompt is that each reviewer must back up any claimed bug with a specific file and line number plus a concrete scenario showing how it would go wrong. A guess with no evidence is dropped rather than reported as noise. It also distinguishes pre-existing problems in the codebase from new ones introduced by the current change, so you are not blamed for failures that were already there. Every report also lists what was verified safe and what still needs a human to manually test, so an unclear result is marked as needing a manual check instead of being falsely reported as passing. To install it, you clone the repository and run its install script, or copy the skill folder into your Claude Code skills directory, either for all your projects or for a single repository. After that, Claude Code will notice natural language requests like is this safe to ship and trigger the skill automatically. It works in any git repository, and the optional GitHub CLI tool is only needed if you ask it to review a specific pull request number. The project is released under the MIT license.