# The Claude Code Skills I Use

Claude Code *skills* are packaged instructions for recurring tasks. I use them to keep
agent-assisted work on-pattern — the same conventions I'd apply by hand, encoded once and
reused.

## Scaffolding domains & components

Generators over hand-rolled files: a skill drives the Nx generators and applies the
post-generation corrections so a new domain or component is born on-convention.

## Placing NgRx state

A skill that decides where reactive state belongs — a component-scoped signal store for
local UI state, the global store via a facade for domain data, or pure functions for
algorithms.

## Verifying theming & accessibility

Skills that check (and auto-fix) CSS-variable token usage and accessibility selectors, so
components stay themeable and testable.

Skills encode the *actions* an agent takes; the layered instruction files those actions are
read against — the always-loaded core, the colocated `AGENTS.md` pointers, and the deep-dive
docs — are documented separately in
[The Agent-Files Architecture](https://jerryorta.com/concepts/claude-agent-files). Beyond the repository's own files,
the same agent reads the framework's pinned upstream source to keep work against newer APIs
accurate — the subject of [Framework Source as Agent
Reference](https://jerryorta.com/concepts/oss-clones-ai-reference).

A skill encodes one action, but skills also compose in sequence. [From Backlog to Merged
PR](https://jerryorta.com/concepts/epic-workflow) describes how a family of them — the `/epic-*` skills — carries a
feature from a planned ticket to a merged pull request, with status kept in Jira and the plan
kept in version control.