A playbook rots at the second project

The first Android skill was written from one project that worked. It stayed accurate for about as long as it took a second project to do something differently.

Pins moved. A database library started to need a newer compile SDK. A trap that cost an afternoon in one repo was missing from the playbook the next repo started from, so it cost an afternoon again.

The fix was not a better first draft. Each project writes its traps down the moment they bite, in its own CLAUDE.md and a shared log. At each milestone the log is merged and de-duplicated. A trap seen in two projects, or one that cost more than an hour, is promoted into the skill. Then a throwaway app is stamped from the skill and built, to prove the skill still makes something that works.

The log holds about 315 entries. Each one was paid for once. That is the point of writing it down.