The Agent Skills Directory

[COMMAND_EXECUTION]: The capture.py script invokes system-level utilities such as osascript (macOS), xdotool (Linux), and screencapture to manage window geometry and perform screen captures. It also interacts with a local utility named thegrid. All subprocess calls are implemented using argument lists rather than shell strings, which prevents shell command injection.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of well-known Python libraries mss (for screen interaction) and Pillow (for image processing). These are standard dependencies for the claimed functionality.
[PROMPT_INJECTION]: As the skill captures and analyzes visual content from the user's screen, it is susceptible to indirect prompt injection. Malicious instructions visible within a captured window (e.g., text on a website) could potentially influence the agent's analysis or responses.
Ingestion points: The agent reads image files generated by the capture.py script as described in agents/shot.md.
Boundary markers: No explicit boundaries or warnings are used to separate user-provided visual content from the agent's instructional context.
Capability inventory: The agent can trigger the capture.py script to perform further captures and output metadata.
Sanitization: There is no evidence of filtering or sanitizing the visual content before it is processed by the model's vision capabilities.

shot