Verify-team agents

Two read-only Claude Code plugin agents — design-reviewer + a11y-auditor — that run in a fresh context for independent design + accessibility review.

Updated 2026-06-25

Two read-only Claude Code plugin agents form the ui-craft parallel verify team. They complement the slash commands — they don’t replace them. Where a command runs inline in your build session, an agent runs in a fresh context with read-only tools, so its judgment isn’t contaminated by the conversation that produced the code.

Agent	Invocation	Does
design-reviewer	`ui-craft:design-reviewer`	Adversarial design critique — loads review rules, anti-slop signals, and Nielsen / design-law heuristics. Returns severity-tagged findings (Critical / Warning / Suggestion, `file:line`). No edits.
a11y-auditor	`ui-craft:a11y-auditor`	Accessibility audit — keyboard, focus-visible, APCA contrast, ARIA, touch targets, reduced-motion. Returns severity-tagged findings. No edits.

Both are read-only (tools: Read, Grep, Glob) and auto-discovered when you install the plugin. They never write code.

Parallel verify team

Delegate both at once on the same diff or file for a full design + a11y sweep in a single pass:

Delegate ui-craft:design-reviewer and ui-craft:a11y-auditor together on [target]. Run both simultaneously. Each returns an independent severity-tagged findings table.

Because each runs in its own fresh context, you get two independent verdicts rather than one session’s accumulated bias.

Agents vs. commands

	Agents	Commands (`/critique`, `/audit`)
Context	Fresh, isolated	Inline, your session
Concurrency	Parallel	Sequential
Writes code	Never	Review passes don’t; build passes do
Best for	Final review, PR audits, CI-style verification	Quick lens mid-build

Reach for the agents when you want independent judgment on a finished diff. Reach for the commands when you want a fast lens while you’re still building.

Availability

The agents are Claude Code plugin agents. In other harnesses (Codex, Cursor, Gemini, OpenCode) they may not be reachable — use the /critique and /audit commands inline, or the portable ui-craft-detect CLI for deterministic checks.

Spotted something out of date? Open an issue on GitHub →