ART
ART is the git repo that runs our agency. It holds the skills, scripts, and accumulated lessons from years of client work, and every agent session we start loads it before doing anything else. When one of us figures out how to do something, it gets written down in ART. From then on the agents know it too.
That changes what a small team can take on. With ART underneath, each of us works more like a department. The same repo that plans content also fixes Core Web Vitals, triages a rank crash, audits a new client, and packages a WordPress plugin release, because the procedures for all of it live in one place and the agents can execute them.
In practice it's our ops deployment layer. Dev ops and marketing ops both run through it. When we want to change how something is done, we commit a change to a runbook, and every future session on every machine works the new way. Git is the deployment mechanism.
What's in the repo
- SkillsRunbooks that say when and how to do things. Decision trees, not essays. This is the procedural memory.
- MethodologyHow we score an audit, what a niche needs, what good looks like. The semantic memory, the domain facts.
- LearningsDated notes written during real work. What broke, what fixed it, what we'd do differently. Episodic memory.
- PolicyOne file every session reads first. It spells out what agents may do on their own and what needs a human.
- ScriptsDeterministic code for everything code can do. If a bash script handles it, we don't ask the model.
The model brings the intelligence and Claude Code brings the runtime. ART is everything we know. A new session starts with all of it, and so does a new teammate.
*ART was named after a very capable spaceship from Martha Wells' Murderbot Diaries.