The lab

The agents, MCP servers, skills, and knowledge tools I use every day.

Skills

asd-ste100

Two layers, always on. The words become Simplified Technical English, so docs and commit messages read like a person wrote them. The shape puts the next action on line one, for a reader with ADHD who will not find it in paragraph four.

/ste

design-taste-frontend

Anti-slop frontend design, for landing pages, portfolios and redesigns. It reads the brief before it picks an aesthetic, reaches for a real design system where one fits, and finishes on a ban list of the tells that give a generated page away.

/tst

better-interface

Reviews a screen across accessibility, color, layout, typography, polish and writing. It routes each problem to the right skill, then returns one ranked verdict.

/btr

break

Stress-tests one component on a temporary page with the hostile inputs it can really receive, then reports what visibly fails.

/brk

explain-interface

Answers "how was this built?" from a URL or screenshot. It reads the layers behind an effect, or reconstructs them when source code is unavailable.

/xpl

interface-review

Reviews the interface in uncommitted work, the current branch or a pull request. It checks the change itself, not the whole product.

/rev

variant

Builds several real versions of one UI piece behind a picker, so I can compare them in context and keep the winner.

/var

elliptic

Teaches an agent the Elliptic workspace over MCP: the domain model, the org discipline, and the mistakes that quietly land a write in the wrong workspace.

/ell

herdr

Drives Herdr, the terminal multiplexer for coding agents. An agent reads neighbouring panes and starts work in them, and refuses to try from outside a Herdr session.

/hrd

Herdr Gram

Agents send me push notifications and files through Herdr. They also receive work I queued for them.

/grm