human-paced-web-ops

# Human-Paced Web Ops Use this skill when the task involves: - Web search and browsing in dynamic pages - Multi-step page navigation that easily breaks with rigid scripts - Long-running read-and-collect workflows ## Interaction Pattern Apply these defaults unless the task needs exact deterministic clicking: 1. Before actions, wait for visible/interactive state first. 2. Use small randomized delays between actions (for example 300-1200ms). 3. Prefer `hover` before `click` on menus/buttons when possible. 4. Use small random click offset inside the same target element (for example 2-8px), not random page clicking. 5. Add occasional small scroll steps during long pages. 6. Avoid repeated fixed-interval requests; pace actions with jitter. 7. Every 5-10 interactions, re-check page state and URL before continuing. ## Guardrails - If blocked by CAPTCHA, login challenge, paywall, or anti-bot page, pause and report the blocker URL plus required manual step. - Keep identity and session settings stable and traceable for repeatable runs. - Respect robots/terms and prefer official APIs when available. ## Output Requirement For web collection tasks, include: - What was visited (titles + links) - What was extracted - What could not be accessed (and why) - Next recoverable step