The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill is designed to process user-uploaded images, which serves as a potential ingestion point for indirect prompt injection attacks where malicious instructions are embedded in image content or metadata.
Ingestion points: uploaded = Image.open('file.png') in SKILL.md.
Boundary markers: Absent. The skill does not provide delimiters or warnings to the agent to ignore instructions found within user-provided image data.
Capability inventory: File system writing (builder.save), extensive image manipulation via Pillow, and frame optimization.
Sanitization: Absent. There is no logic provided to sanitize metadata or validate that the image does not contain textual instructions meant for the LLM.
[Unverifiable Dependencies] (SAFE): The skill specifies standard, well-known Python packages (pillow, imageio, numpy) for installation. These are considered trustworthy dependencies for the stated purpose of image processing.
[Data Exposure & Exfiltration] (SAFE): No patterns for accessing sensitive system files, environment variables, or performing unauthorized network requests were detected. File access is limited to the intended workflow of processing images.
[Command Execution] (SAFE): The skill uses standard Python libraries for its operations. There is no use of os.system, subprocess, or other methods to execute arbitrary shell commands.
[Dynamic Execution] (SAFE): No use of eval(), exec(), or dynamic code generation was found in the provided workflow or utility descriptions.

slack-gif-creator