Building

Building articles

Turning a scoped problem into prompts, tools and a test suite. The recurring argument is that an agent is only as good as the cases you can prove it handles, so most of the detail here is about evals, golden datasets, and the deployment gate that stops a regression reaching anyone.

All Fundamentals Building Architecture Operations Go-to-Market Monetization

Migrate to a New Model Without Breaking ProductionWhat changes when the model changes, how to shadow it on real traffic, and how to keep an eval set meaningful across the switch.4 min
Build an Eval Set That Catches Real FailuresWhere cases come from, what coverage means, and the leakage that makes a suite look better than the agent it tests.5 min
Write a SKILL.md That Claude Actually FollowsThe description is the trigger, not documentation. How to write both halves so the skill fires when it should and gives specific guidance when it does.4 min
Bundle Files, Templates, and Scripts with a SkillProgressive disclosure in practice: what belongs in SKILL.md, what belongs in the folder beside it, and how to write the pointer so the file actually gets read.