Browser Task Runner
Treat browser work as fragile. Prefer deterministic, repeatable flows over improvising clicks.
Workflow
- Confirm the target site and exact outcome.
- Identify the available execution path:
- direct web fetch/search for read-only public pages
- local browser automation tooling if available
- AppleScript/UI scripting as a fallback for simple fixed flows
- Check whether the task needs authentication, MFA, CAPTCHA, or sensitive confirmation.
- For destructive or account-changing actions, ask before proceeding.
- Execute the smallest reliable flow and report where manual takeover may still be needed.
Good fits
- open a fixed set of URLs
- log into a routine internal tool
- navigate to a report page
- export a CSV or PDF
- fill a repetitive form
- capture a screenshot of a page state
- scrape visible page data into a summary
Fragile cases
Be cautious when the task involves:
- CAPTCHA
- MFA prompts
- anti-bot protections
- drag-and-drop heavy UIs
- unpredictable popups
- account settings or payment actions
In those cases, prefer a hybrid flow: automate the boring parts and let the user finish the sensitive step.
Execution policy
Prefer this order:
- Read-only web tools for public pages
- Existing local automation tools/frameworks
- macOS UI scripting for narrow, stable workflows
Do not pretend browser control exists if the tooling is absent. State what is missing and what can still be automated.
Output pattern
Report using:
- Goal
- Automation path
- What succeeded
- Where it may break
- Next improvement
Keep website-specific details in scripts or references if this skill grows beyond a generic runner.