Build and verify interactive HTML experiences in a sandboxed live preview.
npx skills add https://github.com/autonomous-ai/openharness --skill web-autonomous-ai
Please help me install this Agent Skill. Skill: web Source: https://github.com/autonomous-ai/openharness/tree/main/store/agents/web-studio/skills/web Command: npx skills add https://github.com/autonomous-ai/openharness --skill web-autonomous-ai
index.html as the entry point and no bundler required. - Automated Proof Verification: Define actions and assertions in proof.json (clicks, fills, text and attribute checks) that run against the real sandboxed HTTP viewer, saving screenshots and a dependency-hashed browser-proof.json. - Performance Measurement: Drive page controls while measuring rAF delays and long tasks, targeting at least 55 fps and under 2% slow frames. - Use Case: Create an interactive WebGL2 data visualization, then run the screenshot and perf scripts to confirm controls respond, no browser errors occur, and animation stays smooth on both desktop and narrow viewports. ## Quick Start Build an interactive HTML page in the workspace and verify it by running the screenshot and performance proof scripts against index.html.High-intent search queries and answers about installing and using this skill.
Define actions and assertions in proof.json using CSS selectors with click, fill, or select steps, then run the screenshot script against index.html. The proof runs the real sandboxed HTTP viewer, checks browser errors, and saves a screenshot plus browser-proof.json.
Run the perf script, which drives the page controls while measuring rAF delays and long tasks rather than timing an idle page. Target at least 55 fps and under 2% slow frames, and record the hardware and renderer used.
No bundler is required. The preview serves index.html directly with local JS, CSS, images, and ES modules through relative URLs, and it only needs Node 20 or later to run.
The proof helpers require Playwright plus Chromium, installed in the shared isolated-web-viewer package or located via the PLAYWRIGHT_MODULE environment variable. Missing tooling counts as a failed proof, not a passing result.
Yes, set PLAYWRIGHT_CHANNEL=chrome to measure using the installed Chrome browser. Note that a software WebGL renderer does not reflect the user's actual GPU, so record the renderer in your results.
sheawinklerConvert design briefs into self-contained HTML artifacts with embedded CSS and JavaScript.
liuzeming-yuxiGenerate a user-facing delivery summary from verification reports and release notes.
Hack23Automate end-to-end browser testing for static websites with Playwright.
ankaboot-sourceBuilds React and shadcn/ui web apps and bundles them into single HTML artifacts.