a working design practice packaged as an agent skill. an agent loads it before any design work, in any project, canvas or code, and follows one house law instead of improvising per session.
- SKILL.md, the entry point: the reading order and the loop the agent enforces
- rules.md, the system: rule grades and the deposit mechanism. the graded rows themselves (invariants, defaults, experiments) live in house.md, gitignored and private per install; house.template.md seeds a fresh one
- playbook.md, the path through a session: fresh project, existing project, component work, shipping, stuck
- inventory.md, every tool, skill, and mcp in the practice with a verdict on each
- doctrine/, component intake, codebase scaffold, and the design-system package format
copy this folder into your agent's skills directory, .claude/skills/studio
for one project or ~/.claude/skills/studio for all of them. the skill fires
on design work and tells the agent what to read and when. nothing else to
install, the package is self-contained.
borrow freely, but note the practice's own law: rules here were earned from real incidents, so anything you take enters your version as an experiment, not an invariant, until it survives two real uses of your own.
- operator intent first. references give anatomy, never skin
- deliverables are state graphs, not happy frames
- motion numbers live in the margins of the law, tuned by hand, promoted by use
- second-use gate: nothing enters the law until it survives a second real use
- pain-only deposits: rules cite the incident that earned them
the canonical copy lives in cortex
at design/workflows/studio/, mirrored standalone at
tommylower/studio. client projects
appear as neutral aliases (project-a, project-b, ...); dates and incidents
are real. a few pointers, like the learning-journal path in SKILL.md, are
specific to the author's machine.