How htmlark works

Three steps. The agent writes a file. htmlark stores it. You open it on your machine.

Install

npx htmlark on Node 22.13+, or download a binary. Thenhtmlark setup --json.

Write

Any coding agent emits a single HTML or Markdown file. No CDN. Pin scripts as /vendor/pkg@1.2.3/file.js with an exact semver. Use the htmlark-authoring skill so the agent always passes --key and --json.

Put

htmlark put --key q3-sales --file ./q3.html --json creates the artifact or appends a version. The same key is the same id. If you get CONFLICT, get the head, re-apply your edit, retry with --base-version. Do not create a second artifact for the same deliverable.

Open

htmlark open --id art_… starts loopback serve if needed and opens the viewer. Preview runs in a CSP sandbox. Source, versions, copy, and download are in the chrome. Follow head, or pin ?v=N.

Share

htmlark publish --id art_… --json pushes one version to your remote. The public URL is https://a.htmlark.com/a/art_…. unpublish makes it 404. You can still export a file.