The Stacks: a build-knowledge library that outlives any one tool
A git-backed library of what I learned building things, kept above and outside every AI tool, so a lesson from one project is reachable from any other. The pattern is open source.
- Status
- in use
- Stack
- Markdown, Git, Claude Code skills
- Links
- check-the-stacks on GitHub (MIT)
The problem
Every AI coding tool has its own memory, and none of them share. What I learned fixing a filter-syntax bug in one project on Tuesday is invisible to the tool working on a different project on Thursday. The lesson gets relearned, at full price, each time.
The constraint
The library has to live outside the tools. Not in a vendor’s memory feature, not in a project’s own docs, but somewhere every tool and every project can be pointed at. And an entry has to be worth reading a year later by someone with no context, which usually means me.
What I built
A catalog, atomic entries, and curated shelves, all markdown, all in git. Each entry records a problem, the dead ends, what worked, and what to watch for, with a verified date and the project it came from. Shelves group entries by theme. A skill teaches the AI tool to check the stacks before starting work and to write an entry when something is learned.
The three essays on this site are three of those entries, cleaned up for a reader who was not there.
The starter template and the skill are public under an MIT license as check-the-stacks.
What it cost or taught
An entry is only useful if it records the dead ends, and dead ends are the part you least want to write down. The discipline that makes the library work is writing the entry on the day the lesson is learned, while the failure is still embarrassing enough to be specific.