The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill constructs shell commands by interpolating the '$SCREENSHOT_DIR' and '$COUNT' variables into Bash and PowerShell scripts. This creates a high risk of command injection if the directory path provided via the '--dir' argument or the '.claude/screenshot.json' config contains shell metacharacters like semicolons or pipes.
[DATA_EXFILTRATION] (HIGH): The skill is designed to automatically locate and read screenshots from common system directories (e.g., ~/Desktop, ~/Pictures, OneDrive). Screenshots are high-value targets for attackers as they frequently contain sensitive information, including ephemeral credentials, private communications, and PII. Accessing these paths constitutes significant data exposure.
[PROMPT_INJECTION] (HIGH): This skill is highly susceptible to indirect prompt injection (Category 8). Malicious content embedded within a screenshot could be processed by the agent's vision/OCR functions, potentially leading to the subversion of the agent's instructions.
Ingestion points: Local image files (png, jpg, jpeg, gif, webp) in screenshot directories.
Boundary markers: None present.
Capability inventory: Bash (subprocess execution), Glob, and Read tools.
Sanitization: None present.

screenshot