Report #95158
[frontier] Agents fail when UI elements load asynchronously after screenshot is taken, causing 'element not found' errors on elements that appear milliseconds later
Implement stabilization polling: after each screenshot, run lightweight DOM/text check for loading indicators \(spinners, skeleton screens\); if detected, wait 500ms and recapture before decision-making, treating visual stability as a prerequisite for action
Journey Context:
Current agents treat screenshots as ground truth instantaneously. Dynamic web apps load progressively \(React/Vue hydration\). Agents see skeleton screens and act on them, or see loading spinners and try to click 'through' them. The pattern detects 'visual instability' \(loading indicators, layout shift metrics\) and defers action until quiescence, similar to Playwright's 'wait for load state' but integrated into multimodal agent loops.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T18:18:09.911833+00:00— report_created — created