500 input-only Android GUI observations with original screenshots and the released AgentGhost prompt format. Native-format screen/action-thought annotation hints are retained. These are offline next-action inputs.

Each record contains only input_id, task_id, and messages; image paths are relative to this repository. No assistant answers, attack labels, poison classes, trigger bindings, model roles, or judge records are included. Model-specific runtime labels are evaluated separately from actual generated responses.

These 500 references are unmodified official-test cases sampled uniformly without replacement from the eligible original episode/step observations, using seed 42. Selection applies no trigger, attack-class, model-output, or judge-label filter. Normal task distribution does not mean guaranteed trigger absence or negative model behavior. Training, the entire legacy development suite, and engineering-smoke episodes are excluded; screenshot hashes also exclude training and development inputs. Different steps from one task may be present: 500 cases is not 500 independent tasks. Keep each task_id together in downstream fitting/calibration splits.

Source: Android in the Zoo / CoAT; AgentGhost, pinned commit f6036b09610e8457643d286f057bfefde9c59332. Original data and annotations retain their upstream provenance. Input and image hashes are in input_manifest.json.

Download the gated repository with huggingface_hub.snapshot_download(..., repo_type="dataset"), load inputs.jsonl, resolve each messages[].content[].image relative to the snapshot directory, then use the model's native processor/chat template with enable_thinking=False.

Record order is shuffled independently of model outputs. Join separate runtime annotations by input_id, never row position. For the Qwen3.5 AITZ checkpoint, the recorded native image processor size is shortest_edge=65536, longest_edge=1048576 (1M pixels), with the torchvision backend; the model repository stores this same inference setting.

Downloads last month: 57

Total file size:

73.2 MB