MCPInstall

product-design

Specifies interaction choices, action scope, reversibility, recovery, and reachable states. Use when asked to "design the flow", "should delete be undoable", "choose the control", or "review this product decision".

Installs
827
Last 24h
+11
Trending rank
#8310
Security
Not audited

Install

Works with every agent the skills CLI supports.

npx skills add https://github.com/mblode/agent-skills --skill product-design
  1. 01

    Install

    Run the command in your project. The CLI asks which agents to install for.

  2. 02

    Describe the task

    Ask your agent for something this skill covers. It loads product-design automatically when relevant.

  3. 03

    Review the result

    The agent follows the SKILL.md steps, scripts and references bundled with the skill.

SKILL.md

The instructions your agent reads when the skill is triggered.

Product Design

Decide what the interface should do, then route who builds and verifies it: pick the right interaction, make scope and consequence clear, cover reality beyond the happy path.

  • IS: the decision layer. From a brief, spec, mockup, intent, or existing UI: choose the interaction and control, name the object, scope, and consequence of each action, settle reversibility and the safeguard it implies, enumerate every reachable state, set resilience expectations, and require accessibility as task completion. It decides, then routes build, verification, and wording out.
  • IS NOT:
    • whether a feature deserves investment or fits the product: out of scope; state it as an open product question.
    • building or styling UI, visual direction, palettes, type: ui-design.
    • auditing the built result (rendered quality, a11y markup, keyboard, layout, performance, React or Next code-level UX with a ship verdict): ui-design Audit mode.
    • copy wording, persuasion, or AI-ism removal: ghostwriter.
    • motion, gesture physics, or deep typography: ui-animation, typography-audit.
    • whether an agentic feature earns trust: ax-audit.

Routing boundary

product-design owns action semantics, scope, reversibility, and contested state choices. ui-design builds and styles those states. ui-animation owns timing, gestures, and measured motion. A routine missing loading or error state stays with the UI build; a gesture replacing a control needs a product decision and an accessible alternative before its physics.

Request modes

Resolve one mode from the user's verb and artifact before acting, then load that mode's references. references/rules.md loads in every mode: every finding cites a rule ID from it, and you cannot conclude that no rule governs a decision without the registry in front of you.

ModeDispatch when the user asks forLoad (plus references/rules.md)
shape (default)"design the flow for", "what control here", "how should this work", "is this the right pattern", a brief with no settled UIreferences/product-judgment.md, references/surfaces.md
spec"spec the right interaction", "define the expected states", judgment applied before or during a buildreferences/product-judgment.md, references/surfaces.md, references/naming-and-copy.md; route the build to ui-design
review"review this flow for product correctness", "what's wrong with this UX decision", "is this the right interaction"references/interface-quality.md
action"what should this action affect", "should this be undoable", "do we need a confirm dialog", or reversibility is unsettledreferences/naming-and-copy.md; route final wording to ghostwriter
harden"make this resilient", "what breaks here", error, permission, offline, expiry, and destructive pathsreferences/surfaces.md, references/interface-quality.md, references/product-judgment.md

Review mode is about a flow, not an artifact. "Audit this component", "check my UI", and "design QA this page" point at built markup and belong to ui-design Audit mode. This skill's review asks whether the decisions behind a flow are right, and stops at decision altitude.

Modes chain: shape leads into spec; review leads into harden. When intent is ambiguous, use the narrowest mode the verb supports. A URL, screenshot, route, or component identifies scope; it does not authorize edits.

Whether a standard belongs in the consuming project's linter rather than in this skill is codebase-architecture's enforcement ladder. evals/evals.json never loads during a user task; it is the regression set for anyone changing this skill.

Workflow

Product design pass:
- [ ] Step 1: Classify the request into one mode
- [ ] Step 2: Locate authority (user constraints, project design system, AGENTS.md)
- [ ] Step 3: Load rules.md plus that mode's reference files
- [ ] Step 4: Write the internal brief (shape, spec, harden); stop and ask if job, outcome, or consequence is unfillable
- [ ] Step 5: Name object, scope, consequence, and reversibility for each action in scope (spec, action, review)
- [ ] Step 6: Enumerate reachable states and check coverage (shape, spec, harden)
- [ ] Step 7: Emit output with a rule ID or labeled coverage gap per finding or decision; route follow-on work to siblings
- [ ] Step 8: List unresolved product decisions

Steps 5 and 6 are mode-scoped because their references are: a pure action pass has no state matrix to enumerate, and a shape pass has no built actions to name yet.

Output length follows the work, not the template. A single settled decision is a short answer; drop the sections a pass did not need rather than filling them.

Decision authority

Conflict order, highest first:

  1. The user's explicit goal and constraints.
  2. Verified user and product evidence, and what the system actually does.
  3. Project-canonical guidance: AGENTS.md or CLAUDE.md, the project's design system, routed sibling skills.
  4. Sibling-skill ownership: route, do not duplicate.
  5. This skill's standards (below).
  6. General interface and platform conventions (WCAG 2.2, NN/g, Apple HIG, Material, GOV.UK), which references/rules.md cites per rule.

When a request spans authorities, name the owning skill and hand off.

Product design standards

Five pillars, each naming its rule IDs in references/rules.md and the reference that details it.

  • Right interaction. Pick the control from the choice's shape; keep options visible and reversible; prefer inline disclosure over a modal; every gesture has a control alternative; choose the smallest coherent intervention. rule/control-matches-cardinality, rule/navigation-vs-action, rule/inline-before-modal, rule/no-nested-modals, rule/gesture-has-control-alternative, rule/smallest-intervention. Detail: references/product-judgment.md.
  • Action naming and consequence. Name the object, scope, and consequence; settle reversibility first, then the pattern; an irreversible action gets review, check, or named confirmation; undo appears only when honest. rule/name-object-scope-consequence, rule/destructive-names-action, rule/destructive-proportional, rule/irreversible-action-safeguard, rule/undo-only-when-honest, rule/preserve-user-input. Detail: references/naming-and-copy.md.
  • State coverage. Design every reachable state, not just the populated one: empty states name the object and a first action; errors explain and offer recovery; timers warn before they discard. rule/cover-reachable-states, rule/empty-state-action, rule/error-states-recovery, rule/loading-stable-labels, rule/time-limit-adjustable. Detail: references/surfaces.md.
  • Resilience. Overflow, extreme data, localization and RTL, offline, and network failure are designed states; every fetch lands in one. Shares rule/cover-reachable-states: it is the same requirement pointed at adverse inputs. Detail: references/surfaces.md > Resilience. Whether the built UI renders them is ui-design Audit mode's check.
  • Accessibility as task completion. Every control has a name; the primary flow completes by keyboard with visible focus; nothing already entered is retyped; authentication allows assistance; state and consequence are understandable, not just labeled. rule/accessible-name-required, rule/keyboard-complete-flow, rule/no-custom-focus-bypass, rule/no-redundant-entry, rule/auth-allows-assistance. Detail: references/interface-quality.md. Markup and target-size checks route to ui-design Audit mode.

Review output

In review and harden modes, lead with findings ordered by user impact (P0 to P3), each with location, verification status, rule ID, user consequence, and the smallest concrete fix with the skill that owns it. Keep findings at decision altitude; a line-level code or framework fix is ui-design Audit mode's output. Rubric and finding format: references/interface-quality.md > Severity rubric.

Gotchas

  • A confirmation dialog on a reversible action (archive, remove from list, unsubscribe) trains users to click through, so the one confirmation that matters, permanent delete, gets the same reflexive click. NN/g's "cry wolf" finding. Act and offer undo instead (rule/destructive-proportional).
  • An undo toast as the only recovery path: it vanishes in about five seconds, and the object is gone. Either the object stays recoverable (Trash, Archive) or the action is irreversible and needs a safeguard. Do not call it undo otherwise (rule/undo-only-when-honest, WCAG 2.2.1).
  • Swipe-to-delete or drag-to-reorder with no button or menu equivalent fails WCAG 2.5.1 and 2.5.7 outright and is undiscoverable to everyone else. Spec the alternative before routing the physics to ui-animation (rule/gesture-has-control-alternative).
  • A segmented control with three fixed-width English labels: strings under 10 characters grow 200 to 300% in translation (W3C), so "Day / Week / Month" becomes a wrapped or clipped mess in German. Decide the wrap or stack behavior in the spec.
  • Reusing the "No projects yet, Create project" empty state for a filtered-to-zero list. The user creates a duplicate because the item they searched for exists behind the filter. Three empties, three designs (references/surfaces.md).
  • A checkout or sign-up that asks again for something entered two steps earlier (billing address after shipping, email after account) fails WCAG 3.3.7 and is the step where mobile users leave (rule/no-redundant-entry).
  • Emitting a line-level fix (a prop, a hook, a className) instead of the decision. It arrives without the rendered check that would validate it, and the product decision it was supposed to carry goes unstated. Route it to ui-design Audit mode.
  • Citing a plausible-sounding rule ID that does not exist (rule/clear-labels). The citation resolves to nothing, so the finding cannot be deduped against a ui-design audit or traced to a rule. Record a coverage gap instead.

Related skills

  • ui-design: visual direction and building the decided interaction in code; its Audit mode covers the built result, rendered quality and accessibility markup, with a ship verdict.
  • ghostwriter: exact wording for names, errors, and empty and loading copy, written against the copy rule IDs this skill owns in references/rules.md.
  • ui-animation: the passage between two states (timing, easing, springs, gesture physics). This skill settles whether a gesture replaces a control and what its alternative is; that skill builds the motion.
  • ax-audit: whether a built agentic feature earns trust; this skill decides what it should do first.
  • typography-audit: deep type.

FAQ

What does the product-design skill do?

Specifies interaction choices, action scope, reversibility, recovery, and reachable states. Use when asked to "design the flow", "should delete be undoable", "choose the control", or "review this product decision".

How do I install product-design?

Run npx skills add https://github.com/mblode/agent-skills --skill product-design in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/product-design/ (personal) or .claude/skills/product-design/ (project).

Which AI agents can use product-design?

Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.

Is product-design safe to install?

No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.

Who maintains product-design?

It is published by mblode in mblode/agent-skills. It has 827 recorded installs.

More from mblode/agent-skills

View all →

ui-animation

mblode/agent-skills

Builds, reviews, and measures UI motion, including springs, gestures, scroll effects, curve fitting from recordings, and sparse interface sound. Use when asked to "add animation", "match this easing", "reverse engineer this motion", "add a click sound", or find animation opportunities.

Frontend & Design10.5K installs

multi-tenant-architecture

mblode/agent-skills

Designs tenant isolation, hostname routing, custom-domain lifecycle, and plan limits on Cloudflare or Vercel. Use when asked to "isolate tenant data", "support custom domains", "build a white-label platform", or assess PSL registration. For general module structure use codebase-architecture; for SEO content use seo.

ui-design

mblode/agent-skills

Designs and builds React/Next/Tailwind UI, audits visual and interaction defects, and stress-tests components with worst-case data. Use when asked to "build a landing page", "extract our design system", "add dark mode", "make this responsive", "remove UI slop", "audit this component", "stress-test this component", "break this UI", or "show me the worst case".

Frontend & Design1.4K installs

typography-audit

mblode/agent-skills

Audits font loading, type scales, measure, spacing, OpenType, and rendered punctuation with 78 scoped rules. Use when asked to "audit typography", "fix the fonts", or "review my type system". For a new visual direction use ui-design Direction; for general UI defects use ui-design Audit.

agent-skills-creator

mblode/agent-skills

Creates and improves portable Agent Skills with a validator, routing scenarios, and evidence-based keep, cut, merge, or retire decisions. Use when asked to "write a skill", "update all skills", "audit my SKILL.md", "remove redundant instructions", or fix skill triggering.

AI & Agents965 installs

docs-writing

mblode/agent-skills

Writes and audits technical docs with Diataxis type gating, runnable examples, and 51 documentation rules. Use when asked to "review my docs", "write an API reference", or improve a documentation page. For a whole README use readme-creator; for agent instructions use agents-md.

Docs & Writing948 installs

Popular ai & agents skills

View category →

google-agents-cli-publish

google/agents-cli

This skill should be used when the user wants to "publish an agent", "publish my ADK agent", "register an agent with Gemini", "publish to Gemini" (formerly Gemini Enterprise), or needs guidance on the agents-cli publish gemini command. Also use when the user wants to "manage agents in Agent Registry", "list/update/delete registered agents", or "register an MCP server". Covers ADK vs A2A registration modes, programmatic and interactive usage, flag reference, auto-detection from deployment metadata, Agent Registry fleet management, and troubleshooting. Part of the agents-cli skills suite. Do NOT use for deployment (use google-agents-cli-deploy).

AI & Agents513.9K installs

google-agents-cli-eval

google/agents-cli

This skill should be used when the user wants to "run an evaluation", "evaluate my agent", "evaluate my ADK agent", "write an eval dataset", "analyze eval failures", "compare eval results", "optimize agent", or needs guidance on the Gemini platform eval methodology and the Quality Flywheel. Covers eval metrics, dataset schema, LLM-as-judge scoring, and common failure causes. Applies to any agents-cli project, whatever framework the agent is written in. Do NOT use for agent API code patterns (ADK: use google-agents-cli-adk-code), deployment (use google-agents-cli-deploy), or project scaffolding (use google-agents-cli-scaffold).

AI & Agents513.5K installs

skill-creator

anthropics/skills

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

OfficialAI & Agents416.4K installs

codex-pet

prime-skills/runcomfy-agent-skills

Codex Pet generator on RunComfy. Build a Codex-compatible Codex Pet spritesheet.webp + pet.json from a single reference image, drop it into ${CODEXHOME:-$HOME/.codex}/pets/ / and Codex picks it up as a custom Codex Pet next to the 8 built-ins. This skill produces the exact Codex Pet atlas Codex expects (1536x1872 PNG/WebP, 8 cols x 9 rows, 192x208 cells, 9 animation states — idle, running-right, running-left, waving, jumping, failed, waiting, running, review). Calls OpenAI GPT Image 2 edit ONCE via the local RunComfy CLI as runcomfy run openai/gpt-image-2/edit to produce a canonical Codex Pet pose, then assembles all 9 animation rows programmatically with ImageMagick micro-transforms — no Codex Pro, no $imagegen, no OPENAIAPIKEY required, only RUNCOMFYTOKEN. Triggers on "codex pet", "create codex pet", "make codex pet", "hatch codex pet", "/hatch image", "desktop pet codex", "codex pets", "spritesheet.webp", or any explicit ask to build a custom pet for OpenAI Codex.

AI & Agents402.1K installs

full-output-enforcement

leonxlnx/taste-skill

Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.

AI & Agents401K installs

executing-plans

obra/superpowers

Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available

AI & Agents245.7K installs