Hub bundle doyourmagic for github.com/browser-use/browser-use (v0.13.10, measured 09/2026)
and its sibling repo github.com/browser-use/browser-harness (v0.1.13).
Steps
- Read ARGUMENTS →
<slug>. No slug → print the table below then STOP. - Find the child file in order, take the FIRST file that exists, read EXACTLY ONE file:
(1)
../dym-browser-use-<slug>/SKILL.md(next to the hub — installed vianpx skills add rheinmir/dym); (2).overstack/doyourmagic/browser-use/skills/dym-browser-use-<slug>/SKILL.md(bundle in the project); (3).claude/skills/dym-browser-use-<slug>/SKILL.md. Follow that file's Steps/Rules. Do not read any other child file. If none of the 3 are found → say so clearly, stop.
| slug | purpose |
|---|---|
repl |
Drive the browser in a stateful REPL style — one action per turn, the tab persists between turns. Use instead of playwright-verify when the task needs MANY steps. |
qa |
Score a site/app 1–5 with evidence, including console/network errors that screenshots don't show. |
cloud |
Cloud browser + tunnel localhost to the outside, when you need a clean environment or to run in parallel. Costs credits. |
Rules
- Do NOT install
browser-use(agent package, 50+ deps including openai/anthropic/google-genai) just to drive the browser. What you need isbrowser-harness: 4 deps (cdp-use,fetch-use,pillow,websockets), no LLM SDKs. browser-usev0.13 does NOT use Playwright. Every occurrence of "playwright" in its source code is just a comment/doc link (measured:grep -ril playwright browser_use/→ only hits comments;pyproject.toml:244says# "pytest-playwright-asyncio>=0.7.0", # not actually needed I think). Don't read that repo looking for "Playwright tips" — there are none.