turtleneck
A change can pass every test and still be broken when someone opens the app. turtleneck gives your coding agent a process for big projects that checks the real thing at each step, from the first analysis through to production.
The agent drives the real app twice, once on a local stack before anything lands and once on the deployed environment afterwards. Then it gives you a short list of what to check yourself, including anything it could not do, such as signing in as a second user.
Before any ticket is filed
The skill starts before a ticket exists. The agent reads what was asked, measures what is there and writes an RFC in your tracker, with the open decisions put to you in plain English. Tickets come after your answers, and each one says how it will be proven.
After that, everything is written to the tracker. The agent claims a ticket before it branches and comments with its evidence after each step, so a fresh session can carry on from the ticket alone.
Your project's own commands go in a turtleneck.md file, which the agent writes on its first run from your CI and branch rules.
Where it came from
The steps come from work on large enterprise software projects. Most of them exist because something went wrong without them. A command-line bundle once passed 825 tests and could not start. A change was reported landed while the run behind it stayed red for ten hours.