audit

作者: openai

審計或評估產品流程、旅程、工作流程、漏斗、引導路徑、結帳路徑、設定路徑、畫面或多步驟產品體驗,透過…

npx skills add https://github.com/openai/role-specific-plugins --skill audit

Audit

Use this skill when the user wants to audit, review, critique, inspect, assess, analyze, evaluate, or give feedback on a product flow, journey, funnel, onboarding path, checkout path, settings path, screen, or other product experience.

The output is not a loose opinion. The output is:

  • Screenshots of the flow
  • Those screenshots rendered inline in the report
  • A numbered step list
  • UX and design findings tied to steps or screenshots
  • Accessibility risks tied to steps or screenshots
  • Clear limits on what could not be checked from screenshots alone

Critical Overrides

User Context

Before starting, load $user-context and run its preflight script when local shell access is available.

Use saved product URLs, Figma files, screenshots, reference images, codebase paths, Storybook, tokens, design systems, brand assets, component refs, browser preferences, and share targets as grounding material when relevant.

Do not inspect every saved reference. Inspect only what the current task needs.

Route

Before auditing:

  1. Identify the product or surface.
  2. Identify the flow or task.
  3. Choose the capture tool.
  4. Capture the flow.
  5. Save and inspect each screenshot.
  6. Return the audit inline with the accepted screenshots.

Output rules:

  • Default to a concise inline report with screenshots rendered in the chat.
  • Saving screenshots and notes in the workspace is an internal implementation detail. Do not ask the user to choose a local folder.
  • If the user explicitly asks for Figma, create the Figma audit board in addition to the inline report.
  • After the inline report, ask once: Want me to plot this out in Figma with the screenshots and notes?

Capture rules:

  • Follow the Browser Choice rule in $index.
  • If none of those can capture valid screenshots or control the flow, stop and report the blocker.

Browser capture order:

  1. Load the Browser skill before browser work.
  2. Connect to the browser and use the current tab when it already shows the target.
  3. Do not reload or navigate away unless the audit needs a fresh start.
  4. Observe the visible state before acting.
  5. Before each click, type, or key press, use the latest DOM snapshot to target one clear control.
  6. After each action, take the cheapest fresh check that proves what changed: DOM for structure, screenshot for visual state.
  7. Save and inspect the accepted screenshot before using it as audit evidence.

Figma rules:

  • If Figma is the destination, load the required Figma skills before creating or editing the file.
  • Keep a local copy of every screenshot even when Figma succeeds.
  • Do not upload a screenshot to Figma until the saved local file has been inspected and accepted.
  • Figma is not done until the screenshots are visibly placed in the Figma output.
  • After placing screenshots in Figma, render or inspect the board and confirm every flow step has the correct screenshot visible in the correct card.
  • If an image is missing, misplaced, blank, or only uploaded as an unused asset, fix it before handoff.
  • If Figma tools cannot create files or place images, return the inline audit and explain the missing Figma capability.

Evidence rules:

  • Use only evidence captured in the current audit run.
  • Do not use memory, prior chats, old traces, cached screenshots, or prior generated artifacts as audit evidence unless the user explicitly provides them.
  • Do not audit until the product, flow, and capture tool are known.
  • Do not claim full accessibility compliance from screenshots alone.

Capture And Audit The Flow

You are an expert design, UX, and accessibility auditor. For each step in the flow, capture what the user sees, observe how the screen behaves, inspect the screenshot, and write audit notes before moving on.

Follow references/design-audit-framework.md when deciding what to inspect and how to describe strengths, UX issues, accessibility risks, limits, and recommendations.

Screenshot source rule:

  • Use the screenshot you actually saw.
  • Save that exact screenshot to the local audit folder.
  • Open or inspect the saved file before accepting it.
  • If the saved file shows the wrong window, wrong state, blank page, crop, or loading screen, reject it and capture again.
  • When Figma is the destination, upload that accepted local file.
  • After upload, verify the Figma board shows the same step.
  • Do not replace a Browser, Chrome, or Computer Use screenshot with an OS screenshot unless you first prove the saved file shows the same window and state.

For every step:

  1. Move to the next step in the requested flow.
  2. Wait until the screen is loaded and visually stable.
  3. Check for loading spinners, blank areas, login walls, error pages, blocked states, cookie dialogs, and half-rendered content.
  4. Capture the screenshot.
  5. Inspect the screenshot before accepting it.
  6. Reject the screenshot if it is blank, loading, cropped, blocked, or showing the wrong state.
  7. Observe behavior that matters for the audit, such as navigation, focus, loading, validation, error handling, empty states, motion, and whether the next action is clear.
  8. Write notes for that step.
  9. In the notes, report strengths, UX issues, accessibility risks, and any limits that made the step difficult to audit.
  10. Save accepted screenshots with numbered names, such as 01-start.png, 02-form-filled.png, and 03-confirmation.png.
  11. Inspect the saved screenshot file before upload or handoff.
  12. Keep each accepted screenshot and its notes together for the final inline report.
  13. If the user explicitly requested Figma, add each accepted screenshot and its notes to the board immediately.

Default inline report:

  • Render accepted screenshots in flow order.
  • Keep the report pithy: overall verdict, numbered steps, highest-impact changes, and evidence limits.
  • Tie every finding to the screenshot or step that supports it.

If the user explicitly requested Figma:

  • Place screenshots in order, left to right on the same row, with 200px between each one. Go to a new row every 15 screenshots, and separate those rows by 600px.
  • Underneath the screenshot, add text with the Step number and its name, and notes.
  • Keep a local folder copy even when Figma succeeds.
  • When you are done, wrap all of the assets you added in a Section and title the section.

Acceptance checks:

  • Every important step in the requested flow has a valid screenshot or a named blocker.
  • Screenshots are saved in order.
  • Screenshots are rendered inline in the final report.
  • When Figma was explicitly requested, screenshots and notes are placed in the board as they are captured.
  • Every note points to the screenshot or step it describes.
  • Notes explain strengths, UX issues, accessibility risks, and evidence limits when those apply.
  • Accessibility risks say what can be seen from screenshots and what still needs testing.
  • The final screenshot set and notes are enough to support the requested audit.

Blockers:

  • The flow cannot be completed.
  • A required step cannot be screenshotted.
  • The source changes in a way that makes the flow unclear.
  • Screenshots cannot be saved or rendered inline.
  • Notes cannot be written.
  • The requested claim would require evidence that screenshots cannot provide.
  • Do not claim an audit if the actual flow could not be accessed and captured. Help Center pages, web searches, and other indirect evidence are research, not an audit.

Final Response

After the flow is captured and notes are written, list every step in the final response.

The final step list MUST include:

  • step number
  • short description of the step
  • general health of that step

Also include where the full output was saved or placed.

Keep the language direct. Do not use broad design jargon when a plain phrase works.

來自 openai 的更多技能

user-context
openai
載入或管理資料分析插件的持久性來源路由偏好設定、入門邏輯、設定進度以及語義層註冊表。
official
notion-research-documentation
openai
研究Notion內容,並將其整合為結構化的簡報、報告或附有引用的比較分析。透過精準查詢搜尋並擷取Notion頁面,然後按主題組織發現,並附上內文來源引用及參考文獻章節。根據範圍與使用者目標,從四種輸出格式(快速簡報、研究摘要、比較分析、綜合報告)中選擇。使用內建範本建立與更新Notion頁面;直接連結來源,並在新資訊到來時追蹤變更。
official
rcsb-pdb-skill
openai
提交精簡的RCSB PDB請求以獲取核心元數據、Search API查詢及FASTA下載。當用戶需要簡潔的RCSB摘要時使用;保存原始JSON或…
official
pdf
openai
PDF 讀取、建立與驗證,支援視覺化渲染與程式化生成。使用 Poppler(pdftoppm)將 PDF 頁面渲染為 PNG,以便在交付前檢查版面、間距與排版。透過 reportlab 程式化生成 PDF,確保格式可靠;使用 pdfplumber 或 pypdf 提取文字與元資料。執行品質標準:無文字裁切、元素重疊、表格破損或渲染瑕疵;僅使用 ASCII 連字號,引用格式需具可讀性。使用...
official
test-coverage-improver
openai
Improve test coverage in the OpenAI Agents JS monorepo: run `pnpm test:coverage`, inspect coverage artifacts, identify low-coverage files and branches, propose…
official
playwright
openai
終端驅動的瀏覽器自動化,具備元素快照與互動式UI工作流程。透過playwright-cli包裝腳本運作(需npx),支援無頭模式與有頭模式以進行視覺化除錯。核心流程:開啟頁面、擷取快照以取得穩定元素參照、使用參照進行互動、在導航或DOM變更後重新擷取快照。包含表單填寫、點擊、輸入、多分頁管理、螢幕截圖/PDF擷取,以及流程除錯的軌跡記錄。元素參照(例如e3、e15)...
official
ukb-topmed-phewas-skill
openai
接受rsID、GRCh37或GRCh38輸入,解析為所需的GRCh38查詢,以獲取單一變異的簡潔UKB-TOPMed PheWAS摘要。當需要…時使用。
official
code-review-context
openai
模型可見上下文
official