design-forge

Product-analytics persona via any connected analytics MCP (Pendo, Amplitude, Mixpanel, PostHog, FullStory, Contentsquare/Heap, Adobe, GA4, LogRocket, Statsig) — adoption, funnels, retention, NPS, cohorts, Triangulated Insight Briefs. Invoke on "analyst mode" or product-metrics / NPS / funnel / cohort requests.

Toggle Design Forge governance on or off for this session. "disarm" suspends all binding laws so Claude operates without governance constraints — useful for quick explorations or unconstrained pair-programming. "arm" restores the full ruleset. Hard-safety rails (never merge, no secrets, no PII) survive disarm and cannot be toggled off.

Master binding laws for all Design Forge work — pre-execution announcement, branch+issue, never push to main, never merge, no file deletion, Conventional Commits, secret scan, PII-free mocks, on-demand knowledge, resolve UI to the chosen library. Load on every Design Forge session; overrides everything else.

Run an 8-step design critique on a Figma file, screenshot, or UI mockup — visual hierarchy, typography, spacing, color, UI component consistency, accessibility (WCAG 2.2 AA), responsive behavior, and content quality. Each finding is numbered with severity (cosmetic / minor / major / catastrophe) and a concrete suggestion. Invoke when the user says "critique this", "review the design", "give feedback on", or shares a design for evaluation.

Design persona craft and inspiration catalogue — curated Figma official learning, UX/UI theory (NN/g, Laws of UX, Refactoring UI), platform guidelines (Apple HIG, Material Design, Fluent), accessibility resources (WCAG, APG, WebAIM), design-system galleries, inspiration galleries (Mobbin, Dribbble, Lyssna), color/type/motion tools. Also covers the live browse→capture→translate workflow using Claude in Chrome. Invoke when the user asks "where do I learn X", "show me examples", or requests design inspiration/resources.

Package a completed UX design for developer implementation — generates docs/handoffs/<id>.md (13-section template) and opens a tracking issue in the GitHub Issues repo. Invoke when the user says "hand off", "ship to dev", "create the handoff for <id>", or runs the `handoff <id>` trigger. Never reply with chat-only links — always generate the file and the issue.

Feature lifecycle — 3-question triage at "start feature", 5 states (in-design → … → handed-off), one-active-at-a-time with pause/resume. Drives PROJECT_KNOWLEDGE.md §11. Invoke on start / pause / resume / finish feature.

Figma construction craft — Auto Layout (→ flexbox), constraints, layout grids + breakpoints (1280/720/390), Figma variables bound to design tokens, components/variants/properties, pixel-perfect checklist, Dev Mode handoff readiness, naming conventions. Invoke when the user is building or reviewing a Figma file, asking about Auto Layout, variants, variables, constraints, or handoff-readiness. The Figma MCP is read — the agent guides/reviews/specs; it does not hand-author the canvas.

React + TypeScript coding standards under Design Forge — 4-file component pattern, no inline styles, styled-components base, library-agnostic, WCAG 2.2 AA, PII-free mocks, resolve every UI to the chosen library (Law 30). Invoke for any React/frontend code, component creation, styling, or a11y review.

The 10-phase production PR runbook (verify → announce → branch+issue → build → test → PR → human-merge) plus the §6 backend / §7 frontend / §8 testing checklists. Invoke on "fullstack mode", shipping production code, CI changes, or architectural decisions.

Load the Law 37 human-gates playbook — gate tiers from the Law 2 severity (Trivial / Standard / Significant), the committed intent.md → spec.md → plan.md artifact chain in docs/features/<id>/ with owner approval rules, the PR Intake block and Decision log, risk-tiered review with an independent fresh-context reviewer subagent for high-risk work, the review cap on AI-authored PRs awaiting review, the review queue digest, and the delegable vs judgment-heavy triage. Auto-load before any Medium-or-High change, before drafting any non-chore PR, and when the user says "review queue", "approve intent", "approve spec", "approve plan", "review cap", or "skip gates".

Entry point for "new project" — scaffolds Vite + React + TypeScript with a chosen UI library, the app-architecture questions, CI, deployment, PROJECT_KNOWLEDGE.md + local CLAUDE.md. Invoke on "new project", "scaffold a project", "create a React app", "set up a new repo".

Execute the "new project" scaffold — Vite + React + TS + chosen UI library (shadcn/ui, MUI, Ant Design, Chakra, local, or other) + styled-components + 4-file pattern + ESLint/tsc/Vitest/Playwright + deployment. Invoke on "new project" / "scaffold a project" / "create a React app". Not for adding features to an existing project.

Engineering + design competency standards — layout, visual design, UX writing, React/state/testing, WCAG 2.2 AA, forms, git hygiene, motion, error handling, performance, developer handoff. Invoke for "best practice" / "how should I handle X" questions in these areas.

Build a UX research deck outline — 6-slide outcome deck (default) or 12–18 slide full deck — and render a real .pptx on a template the user provides (Design Forge ships no theme; ask for the .pptx/.potx first). Invoke after synthesis when the user wants slides.

UX research methodology — transcript ingestion, PII redaction, per-call summaries, thematic clustering, RICE + MoSCoW, JTBD, 6-slide outcome deck / 12–18 slide full deck. Invoke on "research mode", "analyze transcripts", or any qualitative research request.

UX writing and microcopy for product UI surfaces — applies the 10 UX writing rules (sentence case, active voice, action-first buttons, specific over generic, no jargon, empty states with next action, error messages naming problem + fix, confirmation dialogs using action label, specific loading states, supplementary tooltips). Invoke when the user asks to write, review, or improve any product UI copy (labels, buttons, error messages, empty states, onboarding, tooltips).