The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs a shell command in SKILL.md that uses $(cat /tmp/image-studio-prompt.txt) to pass arguments to the generate_image.py script. If the prompt file contains shell metacharacters (such as backticks, semicolons, or dollar signs), it could lead to arbitrary command injection when executed in the agent's shell environment.
[DATA_EXFILTRATION]: The scripts/generate_image.py script transmits prompt data and potentially sensitive user-provided descriptions to an external third-party API endpoint (api.labnana.com). While this is central to the skill's stated functionality, it involves sending data to an external provider that is not identified as a standard trusted service.
[PROMPT_INJECTION]: The skill processes untrusted user input to build complex instructions for an image generation model (Indirect Prompt Injection). A malicious user could provide input designed to manipulate the final prompt written to the temporary file, potentially triggering the command injection vulnerability mentioned above or bypassing content constraints. The skill includes a manual 'Prompt Audit' step (Step 4) which serves as a human-in-the-loop mitigation, but the technical vulnerability in the execution flow persists.

image-studio