break-ui
Stress-test changed UI with realistic worst-case data, empty and large collections, loading and error states, and narrow layouts. Use before final verification of UI changes or when asked to find visual edge cases. Does not redesign the UI or replace acceptance tests.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- break-ui
- license
- MIT
- description
- Stress-test changed UI with realistic worst-case data, empty and large collections, loading and error states, and narrow layouts. Use before final verification of UI changes or when asked to find visual edge cases. Does not redesign the UI or replace acceptance tests.
02 资源文件 2
- 参考文档CATALOG.md8.1 KB
- 许可证LICENSE1.0 KB
03 SKILL.md 正文
Break UI
Adapted from Emil Kowalski's break-ui. The upstream copyright and license are in LICENSE. CATALOG.md keeps the upstream realistic-value catalog.
Scope and authorization
Start with the named component or the UI changed by the active task. Do not wait for a second prompt or announce readiness. Read the repository's UI, browser, fixture, and test instructions first. Reuse the running task-owned browser and server; never restart the user's stack.
An audit reports findings without fixes. During an authorized implementation, fix clear defects within that scope. Ask only for genuine product choices, such as whether primary identifying text should wrap or truncate. This skill grants no commit, publication, deployment, or production-data permission.
Map the rendered data
List each rendered field, its source, type, optionality, and accepted limit. Include names, identifiers, numbers, statuses, labels, dates, images, and collection length. Read the actual form and API contracts. If no limit exists, record "unbounded"; do not invent a limit or alter the backend to make the test easier.
Read CATALOG.md. Choose plausible synthetic values or the actual accepted boundary. Invalid values belong in rejection/error tests, not in supposedly valid typed fixtures. Use reserved example domains and no customer data.
Exercise the actual component
Create typed fixtures at the existing props, mock, or API boundary. Never edit markup or CSS to manufacture a failure. Mix applicable cases across visible rows: long and short names, unbreakable identifiers, optional fields absent, every supported status, large and singular counts, Unicode and emoji, and missing or failed images.
Also cover empty, one, page-boundary, loading, error, partial-data, and realistic large collections. Select only cases the changed component can render. Do not run unrelated screens or a full catalog sweep for a small fix.
Use the repository's existing story, fixture switcher, test route, or isolated preview. Add a Demo / Worst case selector only when it helps inspection. Keep it outside production routes and bundles, use existing primitives, and retain selection in the URL when supported. Do not add an optional environment variable or leave test controls in the shipped UI.
Inspect and fix
Use the real running component at its actual container width, its supported narrow and wide layouts, and 200% zoom or enlarged text. Check dark mode, RTL, and touch access only where supported. For terminal UI, use the actual TUI with narrow/wide columns, wide characters, resize, themes, and keyboard navigation instead of browser-only checks.
Look for clipped or unreadable text, shrinking icons, unreachable actions, ambiguous truncation, raw missing values, wrong plurals, bad number/date formatting, broken media fallbacks, layout jumps, and large-list stalls. Capture the same scenario before and after any fix. Record observed behavior separately from CSS inference. Without a usable runtime, report NOT VERIFIED rather than treating inference as visual proof.
Keep existing design tokens and components. Choose wrap, clamp, or truncation per field; full identifying information must remain accessible. Do not hide a meaningful number or date to make the layout fit. Recheck normal fixtures as well as worst cases after each coherent fix batch.
Report and retain proof
Give one row per finding with severity, field/value, observed failure, proposed fix, and file:line. Separate real product decisions from defects, then name what already held up. Include the fixture route or command and the tested revision and layouts. A short report with no failures is valid.
Keep useful fixtures or tests that catch the actual break. Prefer the repository's existing regression framework; do not add production debug controls or screenshot pins that duplicate stronger tests. The full requested acceptance flow and repository-required checks still run afterward.
At task end, stop task-owned test servers and browsers and remove transient controls. Leave a demo running only if explicitly requested, with its teardown command. Do not finish with an unsolicited "say fix all" offer during authorized implementation.
04 同仓库更多 Skills
redesign-existing-projects
Upgrades existing websites and apps to premium quality. Audits current design, identifies generic AI patterns, and applies high-end design standards without breaking functionality. Works with any CSS framework or vanilla CSS.
linear
Manage issues, projects & team workflows in Linear. Use when the user wants to read, create or updates tickets in Linear.
design-taste-frontend
Anti-slop frontend skill for landing pages, portfolios, and redesigns. The agent reads the brief, infers the right design direction, and ships interfaces that do not look templated. Real design systems when applicable, audit-first on redesigns, strict pre-flight check.
skill-installer
Install Codex skills into $CODEX_HOME/skills from a curated list or a GitHub repo path. Use when a user asks to list installable skills, install a curated skill, or install a skill from another repo (including private repos).
skill-creator
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Codex's capabilities with specialized knowledge, workflows, or tool integrations.
openai-docs
Use when the user asks how to build with OpenAI products or APIs, asks about Codex itself or choosing Codex surfaces, needs up-to-date official documentation with citations, help choosing the latest model for a use case, or model upgrade and prompt-upgrade guidance; use OpenAI docs MCP tools for non-Codex docs questions, use the Codex manual helper first for broad Codex self-knowledge, and restrict fallback browsing to official OpenAI domains.