The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow requires the agent to execute bash commands to manage files. The specific command mv ~/Downloads/filename.png "/desired/destination/filename.png" incorporates a user-supplied destination path. This pattern is highly susceptible to command injection if the agent does not strictly validate and sanitize the input, potentially allowing for arbitrary code execution by including shell metacharacters in the path.
[DATA_EXFILTRATION]: The skill explicitly interacts with the ~/Downloads/ directory to retrieve generated images. This directory frequently contains sensitive personal files, credentials, and financial documents. Providing an agent with the ability to list and move contents in this directory represents a significant data exposure and exfiltration risk.
[INDIRECT_PROMPT_INJECTION]: Because the skill operates by reading and interacting with the live web interface of Google Gemini, it is vulnerable to indirect prompt injection. Malicious content generated by the external AI or present on the page could influence the agent's subsequent actions, which is particularly dangerous given the agent's access to shell commands and the file system.
[DYNAMIC_EXECUTION]: The skill employs dynamic shell operations where the result of one command (ls -t) is used as input for a subsequent mv operation. This logic is inherently brittle and can be manipulated if the state of the Downloads folder changes during the generation process, potentially causing the agent to operate on incorrect or sensitive files.

gemini-image-gen