Skip to content

Screenshot Sequence β€” Frame Burst

Android Supported iOS Supported

screenshot_sequence captures a burst of screenshots over a short window and caches them per device β€” for a later sub_agent vision call to analyze. The frames are not returned into the main agent’s context, so you can grab 30 frames of an animation without spending 30 images of tokens.

It uses the normal screenshot path (screencap on Android, WDA on iOS) frame by frame β€” deliberately not a video recorder, which can freeze on a playing SurfaceView.

screenshot_sequence: captured 8 frames over 4.0s @ 2.0fps (cached for sub_agent)
  • duration_seconds β€” default 4, clamped 2–180.
  • fps β€” default 1.0, clamped 0.1–4.0.

Frame count is duration Γ— fps.

screenshot_sequence is read-only, so it’s in the safe-tool allow-list and is chainable. The idiomatic use is to open something and capture it in one turn:

chain({"actions": [
{"tool": "launch_app", "args": {"package": "com.google.android.youtube"}},
{"tool": "screenshot_sequence", "args": {"duration_seconds": 6, "fps": 2}}
]})

The frame cache survives across tool calls, so the follow-up sub_agent step reads exactly these frames.

  • Sub-Agent β€” analyzes the cached frames
  • Chain β€” open + capture in one step