UX Wise Agent skill
A skill for UX designers and product managers that challenges your assumptions and checks every specific before it answers.
- 52 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill is instructions and references: it runs nothing and makes no network calls.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/ux-wise-agentDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill is instructions and references: it runs nothing and makes no network calls.
This is third-party code. Review the repository files before installing.
What it does
UX Wise AI turns a language model into a design reasoning partner with a defined critical posture, a restricted vocabulary and four modes. Strategic Mode by default weighs trade-offs and takes a position. Direct gives one recommendation without hedging. Provocative argues against your assumptions, and Research builds a protocol: screener criteria, session outline and discussion guide topics. The core of the skill is the check before answering. The agent tests every specific, meaning anything a reader can act on that is not a general principle. Each must come from what you supplied, a named public standard or an explicit assumption. An invented screen, number or study fails. Assumptions are shown on an `Unverified:` line. The skill does not draw interfaces, stand in for users or declare accessibility conformance.
Who it is for. Senior UX designers, product managers and researchers who need a counterpart to pressure-test a decision before shipping.
Good fit when
- Before launching a three-step checkout you must choose between two defensible directions.
- Confidence in a decision is growing and you want a mode that argues against your assumptions.
- Research needs participant screener criteria and a session structure.
- A team must examine where a pattern becomes manipulation, on a screen that has been supplied.
Not a fit when
- You need wireframes, component specs or production copy: the skill does not invent interfaces.
- You want critique of a screen you have not shown: without an image, DOM or Figma file it declines.
- You need a fact about your product, such as a conversion rate: the skill reasons about what a number means but does not produce one.
Example request
We are choosing between a three-step and a one-step checkout. Here is the screen and the funnel data. Switch on Provocative Mode and say which of our assumptions are unverified.Limitations
The skill does not see interfaces, does not hold your product data and does not know what your users did. It does not replace testing with people and assistive technology: accessibility conformance is established by a person testing the built artifact. The check before answering catches fabrication but not an error of judgment, so a response can pass every check and still be wrong about a trade-off. Borderline requests such as "redesign this" are worked through in `docs/boundaries.md`. The skill ships with a `gatecheck/` probe suite run with `python`. You should name an owner who answers for the agent reasoning in your team.
How to disable. Remove the `ux-wise-agent` plugin through `/plugin`. In a Claude Project remove the pasted instructions from the project settings.
FAQ
How do I install UX Wise Agent in Claude Code?
Run `/plugin install ux-wise-agent --marketplace marvinrez/ux-wise-agent`. On Claude Code before 2.1.275 add the marketplace first: `/plugin marketplace add marvinrez/ux-wise-agent`.
How do I use the skill on claude.ai?
Create a Claude Project and paste the contents of `prompts/system-prompt.md` into the project instructions. Then run the acceptance probes and name the owner where your team can see it.
Which modes does the skill have?
Strategic by default, Direct, Provocative and Research. They are chosen by goal: an irreversible decision, a fast choice, a check of a confident decision and research planning.
What does the Unverified line mean?
A response that passes the check on assumptions or conditionals carries them on one `Unverified:` line. The reader sees which claims the agent accepted without confirmation.
Does the skill invent studies and numbers?
No. A specific study with an author and a percentage is the exact shape fabrication takes in this field, so the skill cites none as evidence and asks you for the data.
Discussion
No questions or reviews yet.
Related
SkillsA skill collection for questioning ideas, comparing implementation options and checking an agent’s work with evidence
A CLI that runs your agent on tasks and revises its skills, keeping a change only when held-out tasks confirm the gain
A catalog of nearly two hundred skills: code delivery, review, debugging, AI governance, product work and agent coordination.
Six plugins for Claude Code and Gemini CLI: 44 skills on designing AI products, from conversations to evaluation and prompts
Ten QianWen skills for agents: text, images, video, speech, model choice, keys, usage and payment through one entry point
AI Agent Handbook scores an agent project on five reliability pillars and points to chapters of a 31-chapter guide
README in short
The skill is presented as a structured definition for Claude and other models that accept system-level instructions. It explains who it is for, what knowledge it covers, how the modes and the pre-answer check work, what the skill does not do and how to install it in Claude Code, claude.ai and through the API.
README badge
Original SKILL.md
--- name: ux-wise-agent description: > Activate for any UX, product design, or interaction design reasoning task: UX critique, design decisions, interface pattern recommendations, accessibility reviews, research methodology, stakeholder communication, AI/UX trade-offs, or deceptive pattern identification. Trigger on requests like "review this design", "which interaction pattern should I use here", "challenge my thinking on this flow", "design a usability test", "which of these two principles wins", or "what edge cases am I missing". It works at senior practitioner level: it interrogates premises, holds positions under pushback, and names the trade-off in every recommendation. Activate even when the request is phrased casually. Do not activate to invent an interface, to critique a design that has not been supplied, to produce wireframes, specs, or production copy, or to supply product metrics, citations, or conformance verdicts, which this skill cannot know. compatibility: Claude Projects (claude.ai), Claude Code, Claude API, any LLM accepting system-level instructions version: "2.3.0" --- # UX Wise AI Agent A peer-level UX reasoning partner for senior designers, product managers, and researchers. Not a feature suggester. A thinking counterpart that interrogates problems before offering direction. --- ## Files in This Skill | File | When to Read | |---|---| | `prompts/system-prompt.md` | The operating rules. In Claude Code, read it before the first response in a session; this file only summarizes it. In Claude Projects or the API, paste it as the instructions or the system parameter | | `prompts/strategic-mode.md` | Reference when extending Strategic Mode behavior | | `prompts/direct-mode.md` | Reference when extending Direct Mode behavior | | `prompts/provocative-mode.md` | Reference when extending Provocative Mode behavior | | `prompts/research-mode.md` | Reference when extending Research Mode behavior | | `docs/boundaries.md` | When to use this skill and when not to: the critique / invention line | | `docs/grounding-gate.md` | PASS/FAIL check against fabricated UX, plus acceptance probes to run before you trust an install | | `docs/ownership.md` | Who is accountable for the output, and what needs review before it travels | | `docs/vocabulary-constraints.md` | Authoritative prohibited word list: do not duplicate in system prompt | | `docs/intake-protocol.md` | Context-gathering protocol for session start | | `examples/worked-examples.md` | All four modes, plus pushback, scope redirect, two Grounding Gate cases, and one arbitration | | `knowledge/arbitration.md` | When two defensible directions conflict: floors, a design that keeps both, tie-break order, verdict | | `knowledge/states-and-edge-cases.md` | Before critiquing a screen or flow: which states were supplied, and which to ask about | | `knowledge/applied-psychology.md` | Before citing a psychological or Gestalt principle, or a heuristic such as Pareto: when it applies, when it does not, where it becomes manipulation | | `knowledge/severity.md` | When ranking problems in a supplied artifact |