Decisions
Architectural choices about how I build with AI. Each one is a short record of why a path was taken, what was rejected, and what I'd revisit. Stronger than opinions, narrower than principles.
Skill rules belong in three layers — manifesto, cross-project behaviors, per-project specifics. Each tier has a different lifecycle, a different audience, and a different promotion rule. Mixing them produces unmaintainable Claude instructions.
Chat history is disposable. The plan file is durable. Every multi-step session writes its evidence into a markdown file that any future session can resume from cold — no handover notes, no "remember to" messages. Resets become cheap.
If a repo is meant to stay generic, defend it with a pre-commit hook, not a willpower-dependent rule. Mechanical guards beat governance — they don't depend on the author remembering what's allowed at commit time.