Claude Code skills

AR Labs runs its work through agent skills: packaged instructions, with any scripts, that a coding agent such as Claude Code loads only when a task needs them. These pages describe the skills we use, for engineers and teams who build with coding agents and want working examples. A skill is a folder: a SKILL.md file that says when the skill applies and what to do, plus whatever workflows, templates, rubrics and scripts the job needs. The agent reads the short description of every installed skill, and loads the full instructions only when a request matches. That keeps the agent's context small until the knowledge is needed.

We work through skills rather than through long one-off prompts. When a way of doing something has proved itself, we write it down as a skill, and from then on every agent session does it the same way. When a skill produces a bad result, the fix goes into the skill, so the correction carries forward instead of being remembered by one conversation. Many of our skills call each other: a design interview feeds a planning document, which feeds a task breakdown, which feeds the coding pipeline.

Each page covers what a skill is for, how it works, the judgement it encodes, and an example of what it produces. We built most of them. Where a skill began as someone else's public work, such as Matt Pocock's skills, Daniel Miessler's Personal AI Infrastructure, OpenClaw's Dreaming or Simon Willison's walkthrough pattern, its page names the author and links to the original. Two are public on GitHub today, Break It Down and an earlier version of Coding Planner. The rest are described here but not released.