[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77MJ70CG7AMMVBJQD45R44G":3,"skill-same:01M77MJ70CG7AMMVBJQD45R44G":54},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":32,"updatedAt":33,"install":34,"resources":47},"01M77MJ70CG7AMMVBJQD45R44G","browser-harness",null,"Direct browser control via CDP. Use when the user wants to automate, scrape, test, or interact with web pages. Connects to the user's already-running Chrome.","davidondrej\u002Fskills","davidondrej","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F131768515?v=4","01M77MHARFKYVNT9FV22807T8F","David Ondrej","01M77MHASHNDV507F9NTVKR2CY",4310,"MIT","https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fbrowser-harness\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: browser-harness\ndescription: Direct browser control via CDP. Use when the user wants to automate, scrape, test, or interact with web pages. Connects to the user's already-running Chrome.\n---\n\n# browser-harness\n\nDirect browser control via CDP. For task-specific edits, use `agent-workspace\u002Fagent_helpers.py`. For setup, install, or connection problems, read install.md.\n\n**Routing check first:** if the task needs no interaction (no clicks, logins, or forms) and you just want page content, use DeepAPI `POST \u002Fv1\u002Fscrape\u002Fwebsite` instead of driving a browser — see the `deepapi` skill. Use browser-harness when the task needs a real browser: interaction, JS-heavy flows, logged-in sessions, or visual verification.\n\nDomain skills (community-contributed per-site playbooks under `agent-workspace\u002Fdomain-skills\u002F`) are off by default. Set `BH_DOMAIN_SKILLS=1` to enable them; see the bottom section.\n\n**If `BH_DOMAIN_SKILLS=1` and the task is site-specific, read every file in the matching `agent-workspace\u002Fdomain-skills\u002F\u003Csite>\u002F` directory before inventing an approach.**\n\n## Usage\n\n```bash\nbrowser-harness -c '\nnew_tab(\"https:\u002F\u002Fdocs.browser-use.com\")\nwait_for_load()\nprint(page_info())\n'\n```\n\n- Invoke as browser-harness — it's on $PATH. No cd, no uv run.\n- First navigation is new_tab(url), not goto_url(url) — goto runs in the user's active tab and clobbers their work.\n\n## Tool call shape\n\n```bash\nbrowser-harness -c '\n# any python. helpers pre-imported. daemon auto-starts.\n'\n```\n\nrun.py calls ensure_daemon() before exec — you never start\u002Fstop manually unless you want to.\n\n### Remote browsers\n\nUse remote for parallel sub-agents (each gets its own isolated browser via a distinct BU_NAME) or on a headless server. BROWSER_USE_API_KEY must be set. start_remote_daemon, list_cloud_profiles, list_local_profiles, sync_local_profile are pre-imported.\n\nWhen supervising those sub-agents, after each check send the user one very short status line: what they are doing and whether they are on track.\n\nClaude Code cmux note: after Claude finishes, it may prefill a predicted next user message; that draft is Claude, not the user speaking.\n\n```bash\nbrowser-harness -c '\nstart_remote_daemon(\"work\")                               # default — clean browser, no profile\n# start_remote_daemon(\"work\", profileName=\"my-work\")      # reuse a cloud profile (already logged in)\n# start_remote_daemon(\"work\", profileId=\"\u003Cuuid>\")         # same, but by UUID\n# start_remote_daemon(\"work\", proxyCountryCode=\"de\", timeout=120)   # DE proxy, 2-hour timeout\n# start_remote_daemon(\"work\", proxyCountryCode=None)      # disable the Browser Use proxy\n'\n\nBU_NAME=work browser-harness -c '\nnew_tab(\"https:\u002F\u002Fexample.com\")\nprint(page_info())\n'\n```\n\nstart_remote_daemon prints liveUrl and auto-opens it in the local browser (if a GUI is detected) so the user can watch along. Headless servers print only — share the URL with the user. The daemon PATCHes the cloud browser to stop on shutdown, which persists profile state. Running remote daemons bill until timeout.\n\nProfiles (cookies-only login state) live in interaction-skills\u002Fprofile-sync.md — covers list_cloud_profiles(), the chat-driven \"which profile?\" pattern, and sync_local_profile() for uploading a local Chrome profile.\n\n## Interaction skills\n\nIf you start struggling with a specific mechanic while navigating, look in interaction-skills\u002F for helpers. They cover reusable UI mechanics like dialogs, tabs, dropdowns, iframes, and uploads. The available interaction skills are:\n- connection.md\n- cookies.md\n- cross-origin-iframes.md\n- dialogs.md\n- downloads.md\n- drag-and-drop.md\n- dropdowns.md\n- iframes.md\n- network-requests.md\n- print-as-pdf.md\n- profile-sync.md\n- screenshots.md\n- scrolling.md\n- shadow-dom.md\n- tabs.md\n- uploads.md\n- viewport.md\n\n## What actually works\n\n- Screenshots first: use capture_screenshot() to understand the current page quickly, find visible targets, and decide whether you need a click, a selector, or more navigation.\n- Clicking: capture_screenshot() → read the pixel off the image → click_at_xy(x, y) → capture_screenshot() to verify. Suppress the Playwright-habit reflex of \"locate first, then click\" — no getBoundingClientRect, no selector hunt. Drop to DOM only when the target has no visible geometry (hidden input, 0×0 node). Hit-testing happens in Chrome's browser process, so clicks go through iframes \u002F shadow DOM \u002F cross-origin without extra work.\n- Bulk HTTP: http_get(url) + ThreadPoolExecutor. No browser for static pages (249 Netflix pages in 2.8s).\n- After goto: wait_for_load().\n- Wrong\u002Fstale tab: ensure_real_tab(). Use it when the current tab is stale or internal; the daemon also auto-recovers from stale sessions on the next call.\n- Verification: print(page_info()) is the simplest \"is this alive?\" check, but screenshots are the default way to verify whether a visible action actually worked.\n- DOM reads: use js(...) for inspection and extraction when the screenshot shows that coordinates are the wrong tool.\n- Iframe sites (Azure blades, Salesforce): click_at_xy(x, y) passes through; only drop to iframe DOM work when coordinate clicks are the wrong tool.\n- Auth wall: redirected to login → stop and ask the user. Don't type credentials from screenshots.\n- Raw CDP for anything helpers don't cover: cdp(\"Domain.method\", params).\n\n## Design constraints\n\n- Coordinate clicks default. Input.dispatchMouseEvent goes through iframes\u002Fshadow\u002Fcross-origin at the compositor level.\n- Connect to the user's running Chrome. Don't launch your own browser.\n- cdp-use is only for CDPClient.send_raw. Prefer raw CDP strings over typed wrappers.\n- run.py stays tiny. No argparse, subcommands, or extra control layer.\n- Core helpers stay short. Put task-specific helper additions in `agent-workspace\u002Fagent_helpers.py`; daemon\u002Fbootstrap and remote session admin live in the core package.\n- Don't add a manager layer. No retries framework, session manager, daemon supervisor, config system, or logging framework.\n\n## Hermes Agent integration\n\nInstalled as editable `uv tool install -e .` from the browser-harness checkout. Binary at `~\u002F.local\u002Fbin\u002Fbrowser-harness`. Skill at `~\u002F.hermes\u002Fskills\u002Fbrowser-harness\u002F`.\n\n**Frontmatter pitfall:** The upstream SKILL.md ships with `name: browser` in frontmatter, which collides with Hermes's built-in `browser` toolset. When copying into `~\u002F.hermes\u002Fskills\u002F`, rename to `name: browser-harness` in the frontmatter or Hermes will shadow\u002Fconflict with its own browser tools.\n\n**Brave Browser:** Works identically to Chrome. Enable remote debugging at `brave:\u002F\u002Finspect\u002F#remote-debugging` (same checkbox). The harness auto-discovers Brave's profile directory.\n\n## Authenticated content extraction (proven pattern)\n\nbrowser-harness connects to the user's real browser with their active sessions — ideal for extracting content from login-walled sites where `web_extract` or Hermes's built-in `browser_navigate` fail (e.g. X\u002FTwitter articles, LinkedIn, paywalled sites).\n\n**Pattern:**\n```bash\nbrowser-harness -c '\nnew_tab(\"https:\u002F\u002Fx.com\u002Fuser\u002Fstatus\u002F123456\")\nwait_for_load()\nimport time\ntime.sleep(5)  # let JS-heavy pages render\ntext = js(\"\"\"\n    const article = document.querySelector(\"article\");\n    if (article) return article.innerText;\n    return document.body.innerText;\n\"\"\")\nwith open(\"\u002Ftmp\u002Fextracted.txt\", \"w\") as f:\n    f.write(text)\nprint(\"Written\", len(text), \"chars\")\n'\n```\n\n- Write to a temp file to avoid shell escaping issues with large text\n- Use `time.sleep()` generously for JS-heavy SPAs (X, LinkedIn need 3-5s)\n- X\u002FTwitter articles render inline — just scroll\u002Fextract via DOM, no extra click needed\n- For very long pages, `js(...)` with `innerText` grabs everything including below-fold content\n\n## Hermes Agent integration\n\nInstalled as editable `uv tool install -e .` from the browser-harness checkout. Binary at `~\u002F.local\u002Fbin\u002Fbrowser-harness`. Skill at `~\u002F.hermes\u002Fskills\u002Fbrowser-harness\u002F`.\n\n**Frontmatter pitfall:** The upstream SKILL.md ships with `name: browser` in frontmatter, which collides with Hermes's built-in `browser` toolset. When copying into `~\u002F.hermes\u002Fskills\u002F`, rename to `name: browser-harness` in the frontmatter.\n\n**Brave Browser:** Works identically to Chrome. Enable remote debugging at `brave:\u002F\u002Finspect\u002F#remote-debugging` (same checkbox). The harness auto-discovers Brave's profile directory.\n\n## Authenticated content extraction (proven pattern)\n\nbrowser-harness connects to the user's real browser with active sessions — ideal for login-walled sites where `web_extract` or Hermes's built-in `browser_navigate` fail (X\u002FTwitter articles, LinkedIn, paywalled sites).\n\n```bash\nbrowser-harness -c '\nnew_tab(\"https:\u002F\u002Fx.com\u002Fuser\u002Fstatus\u002F123456\")\nwait_for_load()\nimport time\ntime.sleep(5)  # let JS-heavy pages render\ntext = js(\"\"\"\n    const article = document.querySelector(\"article\");\n    if (article) return article.innerText;\n    return document.body.innerText;\n\"\"\")\nwith open(\"\u002Ftmp\u002Fextracted.txt\", \"w\") as f:\n    f.write(text)\nprint(\"Written\", len(text), \"chars\")\n'\n```\n\n- Write to a temp file to avoid shell escaping issues with large text\n- Use `time.sleep()` generously for JS-heavy SPAs (X, LinkedIn need 3-5s)\n- X\u002FTwitter articles render inline — just scroll\u002Fextract via DOM, no extra click needed\n- `js(...)` with `innerText` grabs everything including below-fold content\n\n## Gotchas (field-tested)\n\n- **Brave Browser** uses `brave:\u002F\u002Finspect\u002F#remote-debugging` instead of `chrome:\u002F\u002Finspect\u002F...`. The harness auto-discovers Brave's data dir.\n- Login-walled content extraction (e.g. X\u002FTwitter articles): navigate with `new_tab(url)`, `wait_for_load()`, then extract via `js(\"document.querySelector('article').innerText\")`. Write to a temp file to avoid shell escaping: `with open('\u002Ftmp\u002Fout.txt', 'w') as f: f.write(text)`. The user's existing browser session handles auth automatically.\n- Omnibox popups are fake page targets. Filter chrome:\u002F\u002Fomnibox-popup... and other internals when you need a real tab.\n- CDP target order != Chrome's visible tab-strip order. Use UI automation when the user means \"the first\u002Fsecond tab I can see\"; Target.activateTarget only shows a known target.\n- Default daemon sessions can go stale. ensure_real_tab() re-attaches to a real page.\n- Browser Use API is camelCase on the wire. cdpUrl, proxyCountryCode, etc.\n- Remote cdpUrl is HTTPS, not ws. Resolve the websocket URL via \u002Fjson\u002Fversion.\n- Stop cloud browsers with PATCH \u002Fbrowsers\u002F{id} + {\"action\":\"stop\"}.\n- After every meaningful action, re-screenshot before assuming it worked. Use the image to verify changed state, open menus, navigation, visible errors, and whether the page is in the state you expected.\n- Use screenshots to drive exploration. They are often the fastest way to find the next click target, notice hidden blockers, and decide if a selector is even worth writing.\n- Prefer compositor-level actions over framework hacks. Try screenshots, coordinate clicks, and raw key input before adding DOM-specific workarounds.\n- If you need framework-specific DOM tricks, check interaction-skills\u002F first. That is where dropdown, dialog, iframe, shadow DOM, and form-specific guidance belongs.\n\n## Domain skills (opt-in)\n\nOnly applies when `BH_DOMAIN_SKILLS=1`. Otherwise ignore — `agent-workspace\u002Fdomain-skills\u002F` is dormant and `goto_url` won't surface skill files.\n\nWhen enabled, search `agent-workspace\u002Fdomain-skills\u002F\u003Chost>\u002F` before inventing an approach. `goto_url` returns up to 10 skill filenames for the navigated host.\n\nIf you learn anything non-obvious — a private API, stable selector, framework quirk, URL pattern, hidden wait, or site-specific trap — open a PR to `agent-workspace\u002Fdomain-skills\u002F\u003Csite>\u002F`. Capture the durable shape of the site (the map, not the diary). Don't write pixel coordinates (break on layout), task narration, or secrets — the directory is public.\n","# browser-harness\n\nDirect browser control via CDP. For task-specific edits, use `agent-workspace\u002Fagent_helpers.py`. For setup, install, or connection problems, read install.md.\n\n**Routing check first:** if the task needs no interaction (no clicks, logins, or forms) and you just want page content, use DeepAPI `POST \u002Fv1\u002Fscrape\u002Fwebsite` instead of driving a browser — see the `deepapi` skill. Use browser-harness when the task needs a real browser: interaction, JS-heavy flows, logged-in sessions, or visual verification.\n\nDomain skills (community-contributed per-site playbooks under `agent-workspace\u002Fdomain-skills\u002F`) are off by default. Set `BH_DOMAIN_SKILLS=1` to enable them; see the bottom section.\n\n**If `BH_DOMAIN_SKILLS=1` and the task is site-specific, read every file in the matching `agent-workspace\u002Fdomain-skills\u002F\u003Csite>\u002F` directory before inventing an approach.**\n\n## Usage\n\n```bash\nbrowser-harness -c '\nnew_tab(\"https:\u002F\u002Fdocs.browser-use.com\")\nwait_for_load()\nprint(page_info())\n'\n```\n\n- Invoke as browser-harness — it's on $PATH. No cd, no uv run.\n- First navigation is new_tab(url), not goto_url(url) — goto runs in the user's active tab and clobbers their work.\n\n## Tool call shape\n\n```bash\nbrowser-harness -c '\n# any python. helpers pre-imported. daemon auto-starts.\n'\n```\n\nrun.py calls ensure_daemon() before exec — you never start\u002Fstop manually unless you want to.\n\n### Remote browsers\n\nUse remote for parallel sub-agents (each gets its own isolated browser via a distinct BU_NAME) or on a headless server. BROWSER_USE_API_KEY must be set. start_remote_daemon, list_cloud_profiles, list_local_profiles, sync_local_profile are pre-imported.\n\nWhen supervising those sub-agents, after each check send the user one very short status line: what they are doing and whether they are on track.\n\nClaude Code cmux note: after Claude finishes, it may prefill a predicted next user message; that draft is Claude, not the user speaking.\n\n```bash\nbrowser-harness -c '\nstart_remote_daemon(\"work\")                               # default — clean browser, no profile\n# start_remote_daemon(\"work\", profileName=\"my-work\")      # reuse a cloud profile (already logged in)\n# start_remote_daemon(\"work\", profileId=\"\u003Cuuid>\")         # same, but by UUID\n# start_remote_daemon(\"work\", proxyCountryCode=\"de\", timeout=120)   # DE proxy, 2-hour timeout\n# start_remote_daemon(\"work\", proxyCountryCode=None)      # disable the Browser Use proxy\n'\n\nBU_NAME=work browser-harness -c '\nnew_tab(\"https:\u002F\u002Fexample.com\")\nprint(page_info())\n'\n```\n\nstart_remote_daemon prints liveUrl and auto-opens it in the local browser (if a GUI is detected) so the user can watch along. Headless servers print only — share the URL with the user. The daemon PATCHes the cloud browser to stop on shutdown, which persists profile state. Running remote daemons bill until timeout.\n\nProfiles (cookies-only login state) live in interaction-skills\u002Fprofile-sync.md — covers list_cloud_profiles(), the chat-driven \"which profile?\" pattern, and sync_local_profile() for uploading a local Chrome profile.\n\n## Interaction skills\n\nIf you start struggling with a specific mechanic while navigating, look in interaction-skills\u002F for helpers. They cover reusable UI mechanics like dialogs, tabs, dropdowns, iframes, and uploads. The available interaction skills are:\n- connection.md\n- cookies.md\n- cross-origin-iframes.md\n- dialogs.md\n- downloads.md\n- drag-and-drop.md\n- dropdowns.md\n- iframes.md\n- network-requests.md\n- print-as-pdf.md\n- profile-sync.md\n- screenshots.md\n- scrolling.md\n- shadow-dom.md\n- tabs.md\n- uploads.md\n- viewport.md\n\n## What actually works\n\n- Screenshots first: use capture_screenshot() to understand the current page quickly, find visible targets, and decide whether you need a click, a selector, or more navigation.\n- Clicking: capture_screenshot() → read the pixel off the image → click_at_xy(x, y) → capture_screenshot() to verify. Suppress the Playwright-habit reflex of \"locate first, then click\" — no getBoundingClientRect, no selector hunt. Drop to DOM only when the target has no visible geometry (hidden input, 0×0 node). Hit-testing happens in Chrome's browser process, so clicks go through iframes \u002F shadow DOM \u002F cross-origin without extra work.\n- Bulk HTTP: http_get(url) + ThreadPoolExecutor. No browser for static pages (249 Netflix pages in 2.8s).\n- After goto: wait_for_load().\n- Wrong\u002Fstale tab: ensure_real_tab(). Use it when the current tab is stale or internal; the daemon also auto-recovers from stale sessions on the next call.\n- Verification: print(page_info()) is the simplest \"is this alive?\" check, but screenshots are the default way to verify whether a visible action actually worked.\n- DOM reads: use js(...) for inspection and extraction when the screenshot shows that coordinates are the wrong tool.\n- Iframe sites (Azure blades, Salesforce): click_at_xy(x, y) passes through; only drop to iframe DOM work when coordinate clicks are the wrong tool.\n- Auth wall: redirected to login → stop and ask the user. Don't type credentials from screenshots.\n- Raw CDP for anything helpers don't cover: cdp(\"Domain.method\", params).\n\n## Design constraints\n\n- Coordinate clicks default. Input.dispatchMouseEvent goes through iframes\u002Fshadow\u002Fcross-origin at the compositor level.\n- Connect to the user's running Chrome. Don't launch your own browser.\n- cdp-use is only for CDPClient.send_raw. Prefer raw CDP strings over typed wrappers.\n- run.py stays tiny. No argparse, subcommands, or extra control layer.\n- Core helpers stay short. Put task-specific helper additions in `agent-workspace\u002Fagent_helpers.py`; daemon\u002Fbootstrap and remote session admin live in the core package.\n- Don't add a manager layer. No retries framework, session manager, daemon supervisor, config system, or logging framework.\n\n## Hermes Agent integration\n\nInstalled as editable `uv tool install -e .` from the browser-harness checkout. Binary at `~\u002F.local\u002Fbin\u002Fbrowser-harness`. Skill at `~\u002F.hermes\u002Fskills\u002Fbrowser-harness\u002F`.\n\n**Frontmatter pitfall:** The upstream SKILL.md ships with `name: browser` in frontmatter, which collides with Hermes's built-in `browser` toolset. When copying into `~\u002F.hermes\u002Fskills\u002F`, rename to `name: browser-harness` in the frontmatter or Hermes will shadow\u002Fconflict with its own browser tools.\n\n**Brave Browser:** Works identically to Chrome. Enable remote debugging at `brave:\u002F\u002Finspect\u002F#remote-debugging` (same checkbox). The harness auto-discovers Brave's profile directory.\n\n## Authenticated content extraction (proven pattern)\n\nbrowser-harness connects to the user's real browser with their active sessions — ideal for extracting content from login-walled sites where `web_extract` or Hermes's built-in `browser_navigate` fail (e.g. X\u002FTwitter articles, LinkedIn, paywalled sites).\n\n**Pattern:**\n```bash\nbrowser-harness -c '\nnew_tab(\"https:\u002F\u002Fx.com\u002Fuser\u002Fstatus\u002F123456\")\nwait_for_load()\nimport time\ntime.sleep(5)  # let JS-heavy pages render\ntext = js(\"\"\"\n    const article = document.querySelector(\"article\");\n    if (article) return article.innerText;\n    return document.body.innerText;\n\"\"\")\nwith open(\"\u002Ftmp\u002Fextracted.txt\", \"w\") as f:\n    f.write(text)\nprint(\"Written\", len(text), \"chars\")\n'\n```\n\n- Write to a temp file to avoid shell escaping issues with large text\n- Use `time.sleep()` generously for JS-heavy SPAs (X, LinkedIn need 3-5s)\n- X\u002FTwitter articles render inline — just scroll\u002Fextract via DOM, no extra click needed\n- For very long pages, `js(...)` with `innerText` grabs everything including below-fold content\n\n## Hermes Agent integration\n\nInstalled as editable `uv tool install -e .` from the browser-harness checkout. Binary at `~\u002F.local\u002Fbin\u002Fbrowser-harness`. Skill at `~\u002F.hermes\u002Fskills\u002Fbrowser-harness\u002F`.\n\n**Frontmatter pitfall:** The upstream SKILL.md ships with `name: browser` in frontmatter, which collides with Hermes's built-in `browser` toolset. When copying into `~\u002F.hermes\u002Fskills\u002F`, rename to `name: browser-harness` in the frontmatter.\n\n**Brave Browser:** Works identically to Chrome. Enable remote debugging at `brave:\u002F\u002Finspect\u002F#remote-debugging` (same checkbox). The harness auto-discovers Brave's profile directory.\n\n## Authenticated content extraction (proven pattern)\n\nbrowser-harness connects to the user's real browser with active sessions — ideal for login-walled sites where `web_extract` or Hermes's built-in `browser_navigate` fail (X\u002FTwitter articles, LinkedIn, paywalled sites).\n\n```bash\nbrowser-harness -c '\nnew_tab(\"https:\u002F\u002Fx.com\u002Fuser\u002Fstatus\u002F123456\")\nwait_for_load()\nimport time\ntime.sleep(5)  # let JS-heavy pages render\ntext = js(\"\"\"\n    const article = document.querySelector(\"article\");\n    if (article) return article.innerText;\n    return document.body.innerText;\n\"\"\")\nwith open(\"\u002Ftmp\u002Fextracted.txt\", \"w\") as f:\n    f.write(text)\nprint(\"Written\", len(text), \"chars\")\n'\n```\n\n- Write to a temp file to avoid shell escaping issues with large text\n- Use `time.sleep()` generously for JS-heavy SPAs (X, LinkedIn need 3-5s)\n- X\u002FTwitter articles render inline — just scroll\u002Fextract via DOM, no extra click needed\n- `js(...)` with `innerText` grabs everything including below-fold content\n\n## Gotchas (field-tested)\n\n- **Brave Browser** uses `brave:\u002F\u002Finspect\u002F#remote-debugging` instead of `chrome:\u002F\u002Finspect\u002F...`. The harness auto-discovers Brave's data dir.\n- Login-walled content extraction (e.g. X\u002FTwitter articles): navigate with `new_tab(url)`, `wait_for_load()`, then extract via `js(\"document.querySelector('article').innerText\")`. Write to a temp file to avoid shell escaping: `with open('\u002Ftmp\u002Fout.txt', 'w') as f: f.write(text)`. The user's existing browser session handles auth automatically.\n- Omnibox popups are fake page targets. Filter chrome:\u002F\u002Fomnibox-popup... and other internals when you need a real tab.\n- CDP target order != Chrome's visible tab-strip order. Use UI automation when the user means \"the first\u002Fsecond tab I can see\"; Target.activateTarget only shows a known target.\n- Default daemon sessions can go stale. ensure_real_tab() re-attaches to a real page.\n- Browser Use API is camelCase on the wire. cdpUrl, proxyCountryCode, etc.\n- Remote cdpUrl is HTTPS, not ws. Resolve the websocket URL via \u002Fjson\u002Fversion.\n- Stop cloud browsers with PATCH \u002Fbrowsers\u002F{id} + {\"action\":\"stop\"}.\n- After every meaningful action, re-screenshot before assuming it worked. Use the image to verify changed state, open menus, navigation, visible errors, and whether the page is in the state you expected.\n- Use screenshots to drive exploration. They are often the fastest way to find the next click target, notice hidden blockers, and decide if a selector is even worth writing.\n- Prefer compositor-level actions over framework hacks. Try screenshots, coordinate clicks, and raw key input before adding DOM-specific workarounds.\n- If you need framework-specific DOM tricks, check interaction-skills\u002F first. That is where dropdown, dialog, iframe, shadow DOM, and form-specific guidance belongs.\n\n## Domain skills (opt-in)\n\nOnly applies when `BH_DOMAIN_SKILLS=1`. Otherwise ignore — `agent-workspace\u002Fdomain-skills\u002F` is dormant and `goto_url` won't surface skill files.\n\nWhen enabled, search `agent-workspace\u002Fdomain-skills\u002F\u003Chost>\u002F` before inventing an approach. `goto_url` returns up to 10 skill filenames for the navigated host.\n\nIf you learn anything non-obvious — a private API, stable selector, framework quirk, URL pattern, hidden wait, or site-specific trap — open a PR to `agent-workspace\u002Fdomain-skills\u002F\u003Csite>\u002F`. Capture the durable shape of the site (the map, not the diary). Don't write pixel coordinates (break on layout), task narration, or secrets — the directory is public.\n",{"name":5,"description":7},"main","a9d4ad810cf3c303fde23bdcd5fc86df7c57daf3","skills\u002Fbrowser-harness",626,"https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills","https:\u002F\u002Fwww.cloudroom.dev","https:\u002F\u002Fgithub.com\u002Fdavidondrej","https:\u002F\u002Fwww.cloudroom.dev\u002F","founder of Cloudroom","2026-11-14 10:00:36",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":46},[36,39,42],{"tool":37,"command":38},"npx","npx skills add davidondrej\u002Fskills --skill browser-harness",{"tool":40,"command":41},"bunx","bunx skills add davidondrej\u002Fskills --skill browser-harness",{"tool":43,"command":44},"pnpm","pnpm dlx skills add davidondrej\u002Fskills --skill browser-harness","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fbrowser-harness\u002FSKILL.md",true,[48],{"path":49,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":52,"sourceHtmlUrl":53},"references\u002Finstall.md","REFERENCE","参考文档",9688,"https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fbrowser-harness\u002Freferences\u002Finstall.md",[55,62,69,76,83,90],{"skillNo":56,"name":57,"nameZh":6,"description":58,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":19},"01M77MPMR6DE13EZFSMY76S6YB","varied-search","Research a topic across the web, GitHub, and X\u002FTwitter with DeepAPI and produce a concise Markdown report. Use for varied-search or requests to compare web coverage, relevant repositories, and social discussion.","https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fvaried-search\u002FSKILL.md",[],[],{"skillNo":63,"name":64,"nameZh":6,"description":65,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":19},"01M77MPG5F9H2JBW4VFHBY1VSM","total-review","Run five independent max-effort reviews (Fable 5.1, Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, Muse Wonder), merge their findings, and fix only what the user approves. Use for \u002Ftotal-review, or when the user or another agent asks for a total review.","https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Ftotal-review\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M77MP8XZJ42ZJZQV8C76B353","stop-overthinking","Manually-invoked skill that stops overthinking and forces a short, practical decision. Use only when the user explicitly invokes stop-overthinking.","https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fstop-overthinking\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M77MP5EGQSNF1VS9SY8CK60Q","state","Manually-invoked skill that recaps what the current thread is doing and how far along it is. Use when the user says \"state\", \"where are we\", or \"what % are we done\".","https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fstate\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M77MP2H479SWAP7687RAVG7T","setup-help","Guide setup one step at a time and show what remains. Use only when the user explicitly asks for it by name.","https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fsetup-help\u002FSKILL.md",[],[],{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":96,"publishedAt":19},"01M77MNZV9X8NPC9JZMWP8WAF8","self-archive","Archive the current Cloudroom thread and release its runtime. Use only when the user explicitly asks to self-archive, archive yourself, or archive this thread.","https:\u002F\u002Fgithub.com\u002Fdavidondrej\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fself-archive\u002FSKILL.md",[],[]]