Skip to content

docs: ship an agent reference in the package, plus a Claude Code skill - #7

Merged
HaydenBruin merged 1 commit into
mainfrom
docs/agent-reference
Aug 25, 2026
Merged

docs: ship an agent reference in the package, plus a Claude Code skill#7
HaydenBruin merged 1 commit into
mainfrom
docs/agent-reference

Conversation

@HaydenBruin

Copy link
Copy Markdown
Contributor

AGENTS.md, shipped in the package

Lands at node_modules/@engineio/ui/AGENTS.md and versions with the code.
Consumers point at that path rather than copying rules into their own
CLAUDE.md — a copy goes stale on the next bump, and stale brand rules are
worse than none because they still read as authoritative.

Contents are what an agent needs and the README buries:

  • The @source requirement, and why its absence is silent (half-styled app)
  • The three import forms and which names each exposes
  • Variant roster per component
  • Exact token names, as a lookup rather than prose
  • The hard rules, as rules
  • The tv({ extend }) seam
  • Known gaps, so they get flagged rather than worked around

.claude/skills/engine-design-system/SKILL.md

Skills must live in .claude/skills/, ~/.claude/skills/ or a plugin — Claude
Code will not load them from node_modules. So the skill is a thin pointer to
the package's AGENTS.md plus the handful of rules worth knowing before reading
it. Two consequences, both deliberate:

  • The copied file contains nothing that can rot; the substance stays versioned
    in the package.
  • A skill body loads lazily, so the reference costs no context until a UI task
    actually needs it.

Consumer side

engine/front/AGENTS.md and rgs/services/front/AGENTS.md each gain a short
pointer section (committed separately in those repos).

Adds AGENTS.md to the published files, so it lands at
node_modules/@engineio/ui/AGENTS.md and versions with the code. Consuming repos
point at that path instead of copying rules into their own CLAUDE.md, where a
copy goes stale on the next version bump — and stale brand rules are worse than
no brand rules, because they read as authoritative.

The content is what an agent actually needs and a human README buries: the
@source requirement and why its absence is silent, the three import forms, the
variant roster per component, the exact token names, the hard rules as rules
rather than prose, the tv({ extend }) seam, and the known gaps. It deliberately
repeats almost nothing from README.md, which is written for a person deciding
whether to adopt this.

Also adds .claude/skills/engine-design-system/SKILL.md. Skills must live in
.claude/skills/, ~/.claude/skills/ or a plugin — Claude Code does not load them
from node_modules — so the skill is a deliberately thin pointer at the package's
AGENTS.md plus the half-dozen rules worth having in mind before reading it.
Keeping the substance in the package means the copied file contains nothing that
can rot, and a skill body loads lazily so the reference costs nothing until a UI
task needs it.
@HaydenBruin
HaydenBruin merged commit 862874b into main Aug 25, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant