review-ui-full

name	review-ui-full
description	Comprehensive 5-specialist + verifier UI review. Dispatches all review agents (visual, responsive, motion, accessibility, runtime) plus verification pass for maximum coverage. Use when the user wants the full treatment across all quality dimensions. For standard 2-3 specialist review, use review-ui. Triggers: "full UI review", "comprehensive UI review", "review everything about this UI", "thorough UI audit", "full quality audit", "review-ui-full", "complete UI review".
argument-hint	[path | file | directory | screenshot | "diff" | "staged" | "pr" | "all"]
allowed-tools	Bash, Read, Grep, Glob, TodoWrite, Agent

Full UI Review

Coordinate a comprehensive 5-specialist + verifier UI quality review via the team lead.

Step 1: Determine scope

Same as review-ui -- resolve the argument to a concrete file or evidence list.

Argument	Meaning
(empty) or diff	Uncommitted + staged changes, filtered to UI-relevant files
staged	Only staged files
pr	Diff vs main/master
<file>	Single file
<directory>	All UI files in dir
<screenshot>	Screenshot file(s) for screenshot-only review
all	Entire project

Step 2: Gather context

Same pre-flight as review-ui:

• Detect platform (web/iOS/Android/screenshot-only)
• Gather project context (framework, tokens, config)
• Determine evidence availability (Playwright, code, screenshots)

Step 3: Dispatch the team lead

The full review dispatches the ui-review-lead team lead logic, which internally dispatches all 5 specialists + verifier.

Dispatch via general-purpose, not the plugin namespace -- plugin-namespaced dispatch silently strips the Agent tool the lead needs for its 5 sub-specialists. Canonical rationale and failure behavior: the RUNTIME DISPATCH NOTE at the top of agents/ui-review-lead.md; do not restate it here or elsewhere. Read ${CLAUDE_PLUGIN_ROOT}/agents/ui-review-lead.md and inline its full body as the prompt prefix. Omit model -- the team lead and every specialist it dispatches inherit the session model (always the strongest available Claude):

Agent({
  subagent_type: "general-purpose",
  prompt: "<full body of agents/ui-review-lead.md> + <scope, platform, project context, evidence availability>",
  description: "Full UI quality review (5 specialists + verifier)"
})

The team lead handles:

1. Dispatching visual + responsive + accessibility in parallel
2. Dispatching motion + runtime in parallel
3. Running the verifier on all findings
4. Merging into the unified report with per-dimension verdicts

Execution mode

The team lead and its specialists inherit the session model; never dispatch to, or wait on, a specific named model. The 5-specialist parallel dispatch is the default for this workflow because it is what buys maximum coverage -- independent context per dimension, true parallelism, and a dedicated verification pass. Only if the Agent tool is unavailable in the current environment may the orchestrator fall back to running the review dimensions sequentially inline in the main context; this never weakens the read-only guarantee (Hard rule 1 in agents/ui-review-lead.md) or skips the verifier pass (Hard rule 6).

Step 4: Present the report

The team lead returns a structured report. Present it verbatim to the user:

## UI Quality Review Report

**Scope:** <files/screenshots/pages>
**Platform:** <web / iOS / Android / cross-platform / screenshot-only>
**Review mode:** full (5 specialists + verifier)
**Evidence level:** <code + browser / code-only / screenshot-only>

### Dimension Verdicts

| Dimension | Verdict | Blockers | Key finding |
|---|---|---|---|
| Visual quality | ... | ... | ... |
| Responsive quality | ... | ... | ... |
| Motion quality | ... | ... | ... |
| Accessibility | ... | ... | ... |
| Runtime smoothness | ... | ... | ... |

### Summary

**Findings:** N CRITICAL, N HIGH, N MEDIUM, N LOW, N TASTE
**Verdict:** <ship / fix blockers / needs pass / rework>

### All Findings

[numbered list]

### Verification Notes

[removed/downgraded/merged findings with explanation]

Acceptance criteria (check before presenting)

Check	Pass condition
Verdict table	5 dimension rows, each with a verdict from that dimension's verdict set
Verifier ran	"Verification Notes" section present, listing removed/downgraded/merged findings (or explicitly "none")
Finding format	Every finding carries [SEVERITY] [CONFIDENCE] plus Surface, Issue, Why it matters, Evidence, Recommended change
Geometry rule	No spatial claim above "Possible issue" confidence without geometry evidence (${CLAUDE_PLUGIN_ROOT}/references/09-evidence-pipeline.md, "Geometry evidence rule")
Ordering	Findings numbered, CRITICAL first, TASTE last

If the lead's report fails a check, re-dispatch once naming the concrete failure; on a second failure, apply the verifier's rules (${CLAUDE_PLUGIN_ROOT}/agents/ui-verifier.md) to the raw specialist findings directly before presenting -- never present an unverified report.

Step 5: Write learnings

If the review surfaced non-obvious patterns, write to GoodMem Learnings.

When to use this vs review-ui

Use review-ui when	Use review-ui-full when
Quick quality check	Pre-launch quality gate
Specific concern (alignment, a11y)	Comprehensive audit
Small scope (1-5 files)	Large scope or full project
Standard review	Maximum coverage needed
Cost/time matters	Quality is the priority