Runs a full QA engagement for an app: e2e test plan and execution, UI/UX walkthrough review, and static frontend review — up to three docs in the project's docs/. Use when the user asks to "QA this app", "test and review the app", or wants the full treatment; for a single activity invoke e2e-testing, ui-ux-review, or frontend-review directly.
Plans and runs end-to-end tests against a running app — web, API, CLI, TUI, or mobile — producing docs/E2E-TEST-PLAN.md with executable scenarios and walkthrough results. Use when the user asks to e2e-test an app, verify user flows end to end, do a test walkthrough, or write an e2e test plan.
Static source review of an app's interface layer — API client and auth plumbing, state management, component quality, i18n, type safety, accessibility, test coverage, build hygiene — written to docs/FRONTEND-REVIEW.md. Use when the user asks for a frontend code review or a source-level complement to live testing. Runs without the app running.
Reviews a running app's user experience by walking through every screen (or command/endpoint surface) and writing severity-tagged findings to docs/UI-UX-REVIEW.md. Use when the user asks for a UI review, UX walkthrough, design QA, or a "does the app feel right" pass on a live app.
The structural model of an autonomous loop — the four loop types (turn-based, goal-based, time-based, proactive), the six components, and inner/outer layering — for designing a loop rather than one-off prompting. Consult when turning a repeated task into a loop, picking the right loop type, deciding what a loop must contain to run unattended, or when the user asks to "set up a loop / make this run on its own". For the native Workflow tool's orchestration patterns see dynamic-workflows; for the operational safety checklist see "Safe autonomous loops" in CLAUDE.md.
Self-writing vault — a single-inbox markdown vault for the USER'S raw project thinking (voice transcripts, quick notes, half-ideas), which Claude files, cross-links, and synthesizes on a schedule without the user. Operations: init (scaffold), process (inbox → linked notes + daily digest), synthesize (weekly themes/contradictions), health (graph pulse). Use when the user wants to set up a vault, capture a thought mid-session, process the inbox, or run note upkeep unattended. Distinct from knowledge-wiki (digests an EXTERNAL source corpus) and basic-memory (cross-project decision graph).
Control layer for frontend visual design work — landing pages, hero sections, portfolios, marketing sites, redesigns, "make it look good / premium / not generic". Reads the brief, routes to the right globally-installed taste-skill variant (design-taste-frontend, minimalist-ui, industrial-brutalist-ui, high-end-visual-design, redesign-existing-projects), and composes a section-by-section master design prompt from the archetype templates in references/. For API/module interface shape use design-an-interface instead; for dashboards and data-heavy product UI this skill mostly does not apply.
Catalog of the native Workflow tool's 6 orchestration patterns (classify-and-act, fan-out-and-synthesize, adversarial-verification, generate-and-filter, tournament, loop-until-done) plus operational controls. Consult when designing multi-agent orchestration — how to fan out, verify at scale, loop until done, pick parallel vs pipeline, or compose patterns into a workflow.