What is the critique-theater skill?
The critique-theater skill is a SKILL.md in nexu-io/open-design that makes an AI coding agent score its own output on five dimensions: craft, critique, brand, accessibility and copy. It fixes defects in the same turn, treats three passes as a sensible ceiling and reports in plain prose.
The critique-theater skill is a five dimension design review that runs before an artifact is handed over. The agent scores what it just produced against craft, critique, brand, accessibility and copy, then fixes what falls short.
The skill calls itself a quality bar, not a performance. It asks for the review to be done, the fixes applied and the outcome described in ordinary prose, with no scored panel or reviewer personas.
The five dimensions
- Craft covers layout, hierarchy, spacing, alignment and visual rhythm, and asks whether the work reads as deliberate rather than a template with the words swapped.
- Critique asks whether the artifact solves the stated problem, and looks for the weakest claim and the thinnest section.
- Brand checks tokens, typography and voice against the active design system, and treats a violation as a defect. The design system audit skill audits a component library itself.
- Accessibility covers contrast, touch target sizes, focus rings, labels, keyboard reachability and heading order. The accessibility review skill runs a full WCAG 2.1 AA audit.
- Copy checks that text is specific and readable, with placeholder phrasing and marketing padding counted as defects. The UX copy skill writes and reviews microcopy.
How the review loop ends
Each defect is recorded and fixed in the same turn, and a finding that cannot be fixed is stated with the reason. The loop stops when all five dimensions clear or extra rounds stop helping, with three passes as a sensible ceiling.
Who the critique-theater skill is for
Best for
- Agent generated pages, prototypes and components that need a self check before a person looks at them.
- Teams that want accessibility, brand and copy defects caught in the same pass as layout problems.
Not for
- Numeric score panels. This skill reports in plain prose, while the design review skill returns an overall score and flags Nielsen heuristic violations.
How to install Critique Theater Skill: Five Dimension Design Self Check
git clone https://github.com/nexu-io/open-design.git
od plugin install ./open-design/plugins/_official/atoms/critique-theaterInstall the plugin
Clone the repository, then run od plugin install with the path to the critique-theater folder. The README says od plugin install accepts a local folder.
Ask for a review
Finish a page or component, then ask the agent to review it before handoff. The skill's description covers scoring against craft, brand, accessibility and copy.
Check that it loaded
Ask the agent to review a finished page. Expect a plain prose summary of what it checked, what it fixed and what it left alone, with no scored panel or critique.json file.
Key features
- Five dimensionsCraft, critique, brand, accessibility and copy are judged separately, and a strength in one does not excuse a failure in another.
- Craft checkLayout, hierarchy, spacing, alignment and visual rhythm are checked for deliberate work instead of a template with the words swapped out.
- Weakest claim huntThe critique dimension looks for the weakest claim, the thinnest section and the thing a skeptical reviewer would attack first.
- Brand fidelityOutput is compared with the active design system's tokens, typography and voice, and a brand violation counts as a defect even if the result looks pleasant.
- Accessibility passContrast ratios, touch target sizes, focus rings, labels on controls and dialogs, keyboard reachability and heading order are all reviewed.
- Copy checkText must be specific, readable and free of filler, and both placeholder phrasing and marketing padding count as defects.
- Fix in the same turnEach concrete defect is recorded and fixed immediately, and a finding that cannot be fixed is stated plainly with the reason.
- Three passes as a ceilingThe agent iterates while defects remain, stops when all five dimensions clear or further rounds stop producing real improvements, and treats three passes as a sensible ceiling.
- Plain prose reportThe outcome is written as ordinary prose in the user's language, saying what was checked, fixed and deliberately left alone.
Frequently asked questions
What is the critique-theater skill?
It is a skill in the nexu-io/open-design repository that makes an AI coding agent review its own artifact before handoff. The agent scores the work on craft, critique, brand, accessibility and copy, fixes what falls short and describes the outcome in ordinary prose.
How do I install the critique-theater skill?
Clone the nexu-io/open-design repository and run od plugin install with the path ./plugins/_official/atoms/critique-theater, which the README says the command accepts as a local folder. The folder holds SKILL.md and open-design.json. Afterward, finish a page or component and ask your AI coding agent to review it before handoff.
What are the five dimensions the skill reviews?
Craft covers layout, hierarchy, spacing, alignment and rhythm. Critique asks whether the artifact solves the stated problem. Brand checks tokens, typography and voice. Accessibility covers contrast, touch targets, focus rings, labels, keyboard reachability and heading order. Copy checks for specific, readable text without filler.
How many review rounds does the skill run?
It keeps iterating while defects remain and the agent still has room to work. It stops when the artifact clears all five dimensions or when further rounds stop producing real improvements. The skill names three passes as a sensible ceiling for the review.
How does the skill report its findings?
As plain prose in the user's language, saying what was checked, what was fixed and what was deliberately left alone. The skill forbids a structured review transcript, a scored panel, named reviewer personas, tag based envelopes and a critique.json file, since none of them are inputs to anything.
What happens to a defect the agent cannot fix?
The agent states it plainly and gives the reason, so the user can decide what to do. Every defect it can fix is recorded and fixed in the same turn, before the artifact is handed over, and the final prose says what was checked, what was fixed and what was deliberately left alone.
Which license does the critique-theater skill use?
It is Apache-2.0 licensed. The nexu-io/open-design repository uses the Apache-2.0 license, and the critique-theater folder sits inside that repository. The README notes that some bundled skills carry their own LICENSE files, and the critique-theater folder lists only SKILL.md and open-design.json.