The Agent Skills Directory

EXTERNAL_DOWNLOADS (LOW): The script scripts/remove_background.py performs a local installation of rembg, onnxruntime, and pillow using pip install --target if they are not present. While these are legitimate packages, runtime installation is a potential attack vector.
EXTERNAL_DOWNLOADS (LOW): The references/review.md file directs the agent to fetch UI auditing guidelines from https://raw.githubusercontent.com/vercel-labs/web-interface-guidelines/main/command.md. Per the [TRUST-SCOPE-RULE], this is classified as LOW severity because vercel-labs is a trusted organization.
COMMAND_EXECUTION (LOW): The skill executes python3 .agent/skills/designer/scripts/remove_background.py to process images and uses subprocess.check_call within that script to manage local dependencies.
DYNAMIC_EXECUTION (LOW): The remove_background.py script dynamically modifies sys.path and uses importlib.invalidate_caches() to integrate the locally installed libraries into the runtime environment.
INDIRECT_PROMPT_INJECTION (LOW): The skill possesses a vulnerability surface for indirect prompt injection.
Ingestion points: references/review.md instructs the agent to retrieve external guidelines via read_url_content.
Boundary markers: Absent; the fetched content is treated as authoritative instructions for the auditing task.
Capability inventory: The skill can execute subprocesses, write files, and use the generate_image tool.
Sanitization: No sanitization or validation of the fetched URL content is performed before processing.

designer