The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/html_shot.py uses subprocess.Popen to launch system-installed headless browser binaries (such as Chrome, Chromium, or Microsoft Edge). This is used to render HTML files or URLs into images. The command is constructed using a fixed list of candidate paths and validated viewport parameters.\n- [COMMAND_EXECUTION]: The script scripts/long_screenshot_ocr.py uses subprocess.run to execute the glance CLI utility, which is a core component of the vision toolkit used for processing image chunks.\n- [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection because it processes untrusted external data that is subsequently interpreted by a vision model.\n
Ingestion points: scripts/html_shot.py renders HTML files or remote URLs; tools such as glance, ground, and detect process image files.\n
Boundary markers: The instructions do not specify the use of delimiters or provide warnings to the agent to ignore instructions embedded within the visual data or HTML.\n
Capability inventory: The skill allows for file system writes (scripts/crop.py, scripts/extract_fg.py), execution of local binaries via subprocess, and network activity through browser rendering and Vision API communication.\n
Sanitization: There is no evidence of sanitization or filtering applied to the rendered HTML or the image content before it is passed to the vision-capable agent.

vision-skills