| name | browser-harness |
| description | Site-specific browser playbooks (domain-skills, interaction-skills) for Naver, GitHub, etc. Automate the browser with the browser-use CLI (state/indexed elements, eval, screenshots)—same stack as skills/browser-use. Upstream browser-harness Python repo is optional for advanced CDP. |
| allowed-tools | Bash(browser-use:*) |
browser-harness (reference pack) + browser-use CLI
This folder bundles domain-skills/ and interaction-skills/ from browser-use/browser-harness (site maps, UI mechanics). Automation in this project uses the browser-use command, documented in detail in ../browser-use/SKILL.md—same flow as that skill.
Do not use the separate browser-harness shell command as the default driver here unless the user explicitly asks for the upstream Python harness / helpers.py / new_tab+js workflow.
Browser automation: browser-use CLI (default)
Prerequisites: browser-use doctor — see browser-use skill README / upstream CLI docs.
- Navigate:
browser-use open <url> (add --headed to see the window, --connect to attach to running Chrome, or --profile "Default" for saved logins)
- Inspect:
browser-use state — indices for click targets
- Interact:
browser-use click <index>, browser-use input <index> "text", browser-use keys "Enter"
- Scripted DOM / one-off:
browser-use eval "JavaScript returning a value" (replaces a harness js("...") snippet)
- Screenshot:
browser-use screenshot [path.png] (or no path for base64)
- Session stays open between commands — for “same tab, don’t open again”, do not run
open for the same task; use state / eval / click on the current page
- Done:
browser-use close
Maps from harness docs → CLI (when reading domain-skills/)
js("...") / in-page script → browser-use eval "..." (escape quotes for shell; use browser-use python for multiline if needed)capture_screenshot(path) → browser-use screenshot path.pngnew_tab(url) / first visit → browser-use open <url>- “Which element?” →
browser-use state first, then click / input by index
- Remote cloud browser →
browser-use cloud connect and related commands in browser-use skill (not the harness start_remote_daemon + BU_NAME unless user insists on the harness)
Chaining (same session)
browser-use open https://search.shopping.naver.com/search/all?query=demo
browser-use state
browser-use eval "document.body.innerText.slice(0,2000)"
Read ../browser-use/SKILL.md for the full command table, profiles, CDP, cloud, and troubleshooting.
Search first (this directory)
rg --files domain-skills
rg -n "naver|shopping" domain-skills
Examples: domain-skills/naver/scraping.md, domain-skills/github/scraping.md, …
Paths: from repo root, under application/skills/browser-harness/.
When to use this skill
- Domain-specific navigation (Naver security, site APIs, URL patterns) — read
domain-skills/ then drive the page with browser-use.
- Reusable UI patterns (iframes, scrolling) —
interaction-skills/; implement with browser-use (eval, wait, state, scroll).
Always contribute back
Non-obvious site learnings belong in upstream domain-skills/ as PRs. This project’s domain-skills/ is a snapshot; refresh from upstream when appropriate.
What a domain skill should capture
Durable site shape: URL patterns, APIs, selectors, quirks, traps — not one-off run logs or secrets.
Optional: upstream browser-harness command (not default here)
browser-harness is a separate Python tool: new_tab / click_at_xy / helpers.py / CDP daemon. Use only if the user explicitly wants that flow; install and SKILL.md live in the upstream repo, not in this file.
- Docs: upstream
SKILL.md, install.md, helpers.py
- This project’s default for agents is
browser-use CLI as above
References
Skill location in this project
| Path | Role |
|---|
SKILL.md (this file) | Use browser-use CLI + where to read bundled skills |
domain-skills/ | Site-specific reference (bundled) |
interaction-skills/ | UI mechanics reference (bundled) |
