The Agent Skills Directory

[DATA_EXFILTRATION]: The Python script scripts/apimart_image_gen.py contains functionality to read local files provided via the --image-file command-line argument. The local_file_to_data_uri function reads the full content of the file and encodes it in base64 to be sent via a POST request to https://api.apimart.ai/v1/images/generations. This could be leveraged to exfiltrate sensitive files (e.g., SSH keys, AWS credentials) if the agent is instructed to use them as reference images.
[EXTERNAL_DOWNLOADS]: The skill script includes a download_images function that performs GET requests to URLs provided in the APIMart API response. It saves the retrieved data directly to the local filesystem in a user-specified output directory. This allows for the download of arbitrary content if the remote API provides a malicious URL.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the processing of untrusted API responses.
Ingestion points: API responses from api.apimart.ai processed in scripts/apimart_image_gen.py.
Boundary markers: None detected; the script assumes the structure and content of the API response are safe.
Capability inventory: File system read (Path.read_bytes), file system write (Path.write_bytes), and network operations (request.urlopen).
Sanitization: The script performs basic JSON parsing but does not validate the reputation or safety of the URLs returned for image downloads.

apimart-image-gen