Designpowers
Direct a 10-agent team through an inclusive design workflow, from discovery to review.
- Source repo
- Owl-Listener/designpowers
- Stars
- ★ 262
- Last updated
- 5mo ago
- License
- MIT
- Primary language
- Shell
- FA score
- 62/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformClaude Code
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A design lead wants a guided process from problem discovery and user research through an accessible product interface.
- Not a fit if
- Users who only need one-off UI generation without a design workflow
- Teams requiring a documented native ChatGPT or API integration
- Maintenance
- Commits in the last 6 monthslast commit 2026-06-23+0.6 ★/day over 7 days
- Source review
- 62/100 · Some gaps
What does this agent do, and when should you use it?
Designpowers is a model-agnostic design workflow made of Markdown skills, agents, and hooks, with 10 specialist agents and 36 skills. It covers discovery, research, strategy, design, building, review, and handoff, with accessibility included throughout. Choose Direct mode to approve each handoff or Auto mode to run the pipeline and review at the end; Auto still pauses for critical accessibility findings, reviewer conflict, and other documented safeguards. The system keeps an observational record of how you design across projects, which the README says is for you to read and is never used to steer current work. The repository ships as a Claude Code plugin and Gemini CLI extension, and says its Markdown workflow can also be used with other AI coding tools that read instruction files.
When you describe what you want to design, the using-designpowers router activates the workflow and design-state holds shared decisions and handoff notes. Work can begin with design-discovery, then move through research-planning, inclusive-personas, design-strategy, design-taste, inspiration-scouting, and writing-design-plans before design, content, motion, and builder skills implement it. After the build, accessibility-reviewer and design-critic review in parallel, with heuristic-evaluator available for a Nielsen-style pass; findings are reconciled for fixes and handoff. At each handoff, you can approve, correct, add scope, redirect, or skip, and you can review a screenshot and give taste feedback during the build. The Review lane accepts an existing screenshot, URL, or code for assessment. figma-bridge can turn specs or code into Figma frames, bring Figma material in, or render an HTML prototype.
- A design lead wants a guided process from problem discovery and user research through an accessible product interface.
- A designer needs accessibility review, design critique, and heuristic evaluation of an existing screenshot, URL, or codebase.
- A small team wants to approve every agent handoff at first, then switch to Auto once it trusts the workflow.
- A team working from an established design system wants agents to follow the tokens and guidance in a DESIGN.md file.
- A designer wants to catch visual mismatches during implementation by reviewing intermediate output and giving taste feedback.
How do you install or deploy this agent?
The README recommends working inside the repository for Claude Code:
git clone https://github.com/Owl-Listener/designpowers.git
cd designpowersYou can also copy skills, agents, hooks, CLAUDE.md, and .claude/ into your project. For Gemini CLI, work from the repository or install the extension:
git clone https://github.com/Owl-Listener/designpowers.git
cd designpowers
geminigemini extensions install https://github.com/Owl-Listener/designpowers.gitFor other AI coding tools that read Markdown instructions, copy skills/ and agents/ and point the tool to using-designpowers/SKILL.md. The README does not specify model accounts, API credentials, or usage costs.
How do you use this agent?
Start a Claude Code or Gemini CLI session and describe what you want to design; the README says Claude Code shows a welcome screen and other tools can begin discovery. First-time users can opt into the guided walkthrough. Direct mode is the default and pauses at each handoff for approval; say “go auto” to switch modes and “pause” to stop. At a handoff, you can also correct, add scope, redirect, or skip, or address an agent by name. To assess existing work, use the Review lane with a screenshot, URL, or code.
What are this agent's strengths and limitations?
- Ten named specialists cover strategy, visual design, content, motion, implementation, and multiple review roles.
- Accessibility is woven through the workflow, with documented pauses for critical findings and reviewer conflicts.
- Direct and Auto modes support both hands-on approvals and a more automated pipeline, with mid-build visual feedback.
- It supports DESIGN.md and includes a Figma bridge skill.
- Claude Code hooks are specific to Claude Code; adapting the workflow to other tools may require changing file paths and conventions.
- The README does not specify model providers, account or key requirements, or actual usage costs.
- No standalone API or server interface is documented; the documented delivery is Markdown instructions, a plugin, or an extension for coding tools.
How does this agent compare with similar options?
The README presents Designpowers as complementary to Superpowers: Superpowers handles engineering workflows such as TDD, debugging, and code review, while Designpowers handles design workflows such as discovery, strategy, accessibility, and critique.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Designpowers This agent | 62 · Some gaps | Agent plugin / skill | ★ 262 | 5mo ago | Shell | Claude Code |
| Dembrandt UX Skills | 60 · Some gaps | Agent plugin / skillFree | ★ 197 | 1d ago | JavaScript | Claude Code |
| Agent Skills for Claude Code | 42 · Major gaps | CLIFree + model costs | ★ 106 | 3d ago | HTML | Claude Code · Claude.ai |
| XIAS — UX/UI Agent Skills | 74 · Some gaps | Agent plugin / skillFree + model costs | ★ 3.2k | 3d ago | JavaScript | Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Direct mode defaults to approval at each handoff, and users can correct, skip, or redirect work. Auto mode and visible Figma, research, and file workflows are also described, so privilege and external-effect controls receive partial credit. The README describes design-state and a taste-profile under the home directory, but not retention, deletion, or access controls for sensitive material, and data transmission is not fully explained. MIT metadata and a package.json without runtime dependencies help, but dependency auditing is not evidenced. Rollback appears as design advice; recovery for product operations is unclear. Acknowledgements name W3C and other knowledge sources and credit Superpowers for the plugin architecture, but individual design claims are not consistently sourced.
The README, skill directory, plugin manifests, and package metadata support a coherent description of 10 agents, multiple skills, and two platform entry points. A consistency workflow checks orphaned skills, router references, and counts, earning substantial but not full credit. The Markdown product has few runtime dependencies, but still relies on the host AI tool and its loading conventions. Specific error messages and recovery paths for failed loading, unavailable tools, or interrupted workflows are not adequately documented.
The materials cover build and review lanes, POCs and full products, varied user preferences, and accessibility needs. The debate skill gives clear use and non-use cases, so scenarios and triggers are well specified. Claude Code, Gemini CLI, and other tools have documented entry points, and the README says hooks are Claude-specific; other tools still require path and rule adaptation, and actual capability depends on the host.
The README is organized around the product, workflow, skills, installation, and support tasks, with clear tables and examples. Claude and Gemini installation steps are provided, though copy-based installation and adaptation to other tools require user judgment. Skill names are generally stable and have trigger descriptions; examples are plentiful, but no FAQ is shown. The MIT license text is complete. No changelog or explicit maintenance commitment is provided; manifest version numbers alone do not establish update history. The README names an author and issue channel, but maintenance responsibility remains vague.
The workflow turns research, strategy, building, accessibility review, fixes, and handoff into reusable steps. Plans, checklists, parallel reviews, and mid-build feedback support usable outputs. Specialist agents, design memory, and mode-based coordination may add value beyond generic prompts, but the claimed benefit is not demonstrated by static source material. The full process and large skill set also add setup and usage overhead, partly offset by quick-entry options and progressive tips.
README feature descriptions can be checked against skills, installation notes, manifests, and the consistency workflow, so some architectural claims have cross-source support. Many claims about quality, accessibility outcomes, and real platform compatibility lack corresponding result evidence. The debate skill asks agents to cite the brief, personas, and evidence but does not define a consistent source-recording method. Intended behavior and verified facts are not always clearly distinguished.
- Auto mode runs handoffs without step-by-step review; use it only when that level of oversight fits the task, and note the documented pause conditions.
- Design records accumulate across projects. Although the materials say they are observational, check that the local record and its location meet your privacy expectations.
- Compatibility with tools beyond Claude and Gemini depends on manual adaptation; cross-platform claims do not establish that every tool has been verified.
FAQ
Do I have to use Claude Code?
Can it review a design that already exists?
Does Auto mode remove human review?
Does it require a paid model or API key?
Does the design record set the style for future projects?
Is Designpowers open source?
Is Designpowers safe to use?
Is Designpowers still maintained?
Who makes Designpowers?
Thanks — we use this to improve the page.