Skills

Skills fire automatically when Claude encounters matching phrases in conversation. You do not need to invoke them — they activate on their own. You can also trigger them explicitly if you want.

Discipline

These enforce good engineering habits without you having to ask. Together, atomic-tdd and atomic-verify are the loop's objective gate: a test that passes or fails, not a second opinion. That is the difference between a loop that catches its own mistakes and one that talks itself into "done".

Skill	Fires when you say...	What it does
atomic-tdd	"let's implement X", "add feature Y", "fix bug Z"	Writes a failing test before touching production code.
atomic-verify	"done", "fixed", "passing", "ready to merge"	Runs verification before letting Claude claim completion. No evidence, no claim.
atomic-debug	pastes an error, "broken", "doesn't work", "failing"	Drives hypothesis-driven debugging instead of symptom-patching.

Workflow

These handle the craft of committing, reviewing, and documenting.

Skill	Fires when you say...	What it does
atomic-git-discipline	"write a commit", "commit message", "open a PR", "PR description", or automatically from ship commands	Generates a Conventional Commits message. Subject under 50 chars, body only when the "why" is not obvious. Also owns PR titles and bodies: only what the diff can't show, about 120 words max.
atomic-review	"review this PR", "code review", "review the diff"	Produces compressed review comments. One line per finding: location, problem, fix.
atomic-visual-options	"show me options", "mock up variants", "let me see this side by side", "compare these layouts"	Renders 2–4 visual variants per decision dimension as a throwaway self-contained HTML file; user picks by typing codes (e.g. A2 B3); chosen option recorded in the design doc. Invoked by /atomic-plan when a design question is visual.

Awareness

These keep Claude and your docs in sync with the project.

Skill	Fires when you say...	What it does
atomic-writing	"draft the README", "write the docs", "edit the guide", "write the spec"	The one voice for every file the repo ships, from READMEs to specs to agent prompts. Length follows the surface; a diagram beats a paragraph wherever the content has a shape. No marketing language, no AI-tell phrases.
atomic-documentation	"doc this change", "what surfaces does this touch"	Figures out which docs need updating based on a diff and routes each to the right voice.
atomic-wiki	"I want a place for notes/tickets/research", "add a bucket", "add this to my wiki", "wiki this", "what does my wiki know about X", "is my wiki stale", "set up a karpathy wiki"	Routes capture-folder intent to atomic wiki bucket add (never a bare mkdir), handles karpathy-realm setup, and answers staleness queries for registered wikis.
atomic-bus	"join the bus", "connect to the bus", "message the backend session", "delegate this to the frontend session", "coordinate with the other agent", "watch the room", "halt the room"	Joins a named room (atomic bus join) and starts a recv Monitor so peer messages arrive as prompts. Carries the reaction policy: act on messages addressed to you, note FYI traffic, never act unaddressed.