desktop-browser-chrome
Skill1 starsChanged 50 days agoMITView on GitHub
Use when opening URLs in the user's real Chrome.
In this repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- ai-hotspot-digestSkill
- convert-documents-to-markdownSkill
- cursor-cliSkill
- feishu-chat-style-reviewSkill
- feishu-ui-feedbackSkill
- figma-copy-proofreadSkill
- hermes-daily-reviewSkill
- hermes-upstream-checkSkill
- writing-with-cursorSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 46
If a tab title already is the target site → click that tab.
- Very specific: line 60
Default computer_use is background.
- Distinctive: line 67
Do not re-type a URL that already confirmed in the omnibox.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Desktop browser chrome (Chrome via computer_use)
Load computer-use first for the full capture / escalate ladder.
This skill only adds browser chrome navigation validated on the owner's
macOS Chrome (cua-driver background + foreground).
When to use
- User asks to open a site in their Chrome (not headless)
- Task needs logged-in profile / extensions / existing tabs
- Omnibox / tabs / window chrome — not page DOM (for page content
prefer
cua_browser_*when binding is exact, or headlessbrowser_*)
Success criteria (hard)
Do not claim navigation success from address-bar text alone.
| Signal | Enough? |
| Omnibox value = target URL | no |
type → effect: confirmed | no |
| Window title / web-area label matches site | yes |
| Address bar + main content both match | yes (best) |
Example: after https://example.com, wait until title looks like
Example Domain … (not still the previous page, e.g. … Jenkins).
Preferred open-URL sequence (macOS)
list_apps/focus_app(app="Google Chrome", raise_window=false).capture(mode="som", app="Google Chrome").- If a tab title already is the target site → click that tab; done.
- Else navigate with foreground chrome focus (one coherent sequence):
key(keys="cmd+l", delivery_mode="foreground")
type(text="https://example.com", delivery_mode="foreground")
key(keys="return", delivery_mode="foreground", capture_after=true)- Re-check title / web area. If still on old page → one more foreground
cmd+l→ type →return, or ask user to press Enter once.
Background-first still applies — then escalate
Default computer_use is background. For Chrome omnibox:
| Attempt | Typical result | Next |
Click address field + type + background return | URL written; page may not change; Enter often unverifiable | Escalate submit to foreground |
Foreground cmd+l + type + return | Page actually navigates | Verify title |
Do not re-type a URL that already confirmed in the omnibox; only re-submit (or re-run the full foreground jump).
Pitfalls
| Symptom | Cause | Fix |
| Omnibox shows URL, content still old site | Background Enter / AXPress did not submit navigation | Foreground cmd+l → type → return |
| Click address field feels flaky | Omnibox UX / focus | Prefer cmd+l over clicking the text field |
| Huge SOM trees on Jenkins-like pages | Dense AX | Scope app="Google Chrome"; ignore page body until chrome is done |
| User said "continue yourself" | Expect agent to escalate, not stop at half-state | Keep going until title/content verify or one clear blocker |
Related
- Full tool vocabulary + escalate ladder: skill
computer-use - Session notes:
references/chrome-omnibox-navigation.md
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.