Menu
Use browser tools to open webpages, inspect page state, interact with elements, debug browser data, and extract content in the browser automation session.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Deep web search workflow using browser tools and multiple search engines such as Bing, Google, Baidu, Sogou/WeChat, 360, Toutiao, DuckDuckGo, Brave, Quark, or other available engines. Use when the user asks for deep_search, deep search, multi-engine research, source discovery, current web evidence, or when Codex needs to search, collect result links, de-duplicate sources, open candidate pages, and gather browser_snapshot/browser_extract evidence from each source.
Guide for creating effective skills. Use when users want to create a new skill, update an existing skill, inspect a prior session for failures, or improve one or more related skills from session evidence.
Generate and edit Draw.io-first diagrams for formal documents, reports, proposals, design documents, and implementation plans. Use Draw.io for flowcharts, sequence diagrams, class diagrams, Entity-Relationship (ER) diagrams, mind maps, architecture diagrams, UI wireframes, module dependency diagrams, and network topologies. Do not generate Mermaid for formal方案/设计文档; convert any existing Mermaid content into Draw.io diagrams and export PNG/SVG assets for insertion into Word/PDF/PPTX.
UI/UX design intelligence and implementation guidance for building polished interfaces. Use when the user asks for UI design, UX flows, information architecture, visual style direction, design systems/tokens, component specs, copy/microcopy, accessibility, or to generate/critique/refine frontend UI (HTML/CSS/JS, React, Next.js, Vue, Svelte, Tailwind). Includes workflows for (1) generating new UI layouts and styling, (2) improving existing UI/UX, (3) producing design-system tokens and component guidelines, and (4) turning UX recommendations into concrete code changes.
Generate reusable data charts from CSV files and chart specifications for reports, proposals, presentations, dashboards, and analytical documents. Use when Codex needs to create line charts, bar charts, scatter plots, correlation heatmaps, or chart image assets for insertion into Word/PDF/PPTX workflows.
Generate, rewrite, expand, or polish high-quality document text for solutions, reports, proposals, plans, presentations, or analysis before Word/PDF/PPT formatting. Use for text-first deliverables that may need planned figure insertions, including flowcharts, architecture diagrams, topology diagrams, process diagrams, design sketches, data charts, and analytical visualizations. Enforce Request → Plan → Draft → Audit → Refine → Deliver for complex writing tasks; allow streamlined handling for simple text-only requests.
| name | tasi-browser-automation |
| description | Use browser tools to open webpages, inspect page state, interact with elements, debug browser data, and extract content in the browser automation session. |
| category | browser |
Use this skill when the user needs web actions through Tasi Harness browser automation.
In external browser mode, these browser tools attach to the managed Chromium-family external browser through CDP. Treat the external page as the real automation target, and use the harness-managed preview opener only as a fallback or manual surfacing path.
If the app setting enables headless external CDP mode, no browser window is visible. Continue by observing pages with browser_snapshot, browser_extract, and browser_screenshot.
Use these browser tools:
browser_open to navigate to a URL.browser_snapshot to get the default accessibility/semantic tree with @e refs, roles, names, states, links, headings, images, and viewport data. Prefer this before interacting with unfamiliar pages.browser_find to locate elements by role, text, label, placeholder, alt text, title, test id, or CSS; it can return a ref or perform click, fill, type, hover, check, uncheck, select, or text.browser_click, browser_type, browser_hover, browser_select, browser_check, and browser_press to interact with controls. Use @e refs from browser_snapshot when available.browser_wait to wait for selectors, visible/hidden/detached state, text, URL glob, load state, JavaScript condition, or a short delay after navigation.browser_scroll to load lazy content; it can scroll the page or a scrollable selector/ref.browser_extract to capture text, HTML, or structured JSON from the full page or a selector/ref.browser_screenshot and browser_pdf to save visual evidence or a printable copy inside the workspace.browser_storage, browser_cookies, browser_console, and browser_network to inspect storage, cookies, console/page errors, and resource timing/network data.browser_eval for bounded page-side JavaScript inspection when the dedicated tools do not expose the needed data.browser_viewport to resize the browser for responsive checks.browser_state to verify current page location and title.browser_close to reset the browser session.browser_open.browser_state if navigation or redirects are ambiguous.browser_snapshot to read the accessibility tree, identify interactive elements, and collect @e refs.browser_find or @e refs over brittle positional CSS selectors.browser_extract and prefer format=json for page data you want to normalize.browser_console, browser_network, browser_storage, or browser_cookies when debugging failed interactions, auth state, SPA/API behavior, or missing data.max_chars and iterate if more detail is needed.When the final answer uses facts from opened webpages, include numbered inline Markdown citation links:
claim text[1](https://example.com/source) immediately after the supported claim.browser_snapshot.snapshot as the primary page observation for actions: it is an accessibility-first tree optimized for LLMs. Prefer semantic lookup and @e refs over brittle positional assumptions.browser_extract after browser_snapshot when you need article/body text, HTML, or normalized content rather than page controls.browser_open, browser_click, browser_find, and browser_extract operate on the external CDP target. If the external engine is set to Safari/WebDriver or falls back to shell.openExternal, the page may be preview-only and browser data tools can be unavailable.browser_eval sparingly. Prefer dedicated tools for state, storage, network, console, and extraction.tasi-browser-automation to avoid confusion with similarly named community skills.