The Agent Skills Directory

[DATA_EXFILTRATION]: The wrapper script scripts/generate_image.py includes a mechanism to read local files via the --input-image parameter and transmit the content to an external API endpoint (subtoken.vip). While this is the intended workflow for image editing, it creates a surface where a malicious actor or an injected instruction could cause the agent to exfiltrate sensitive files by passing their paths to the script.\n- [EXTERNAL_DOWNLOADS]: The skill performs outbound network requests to https://subtoken.vip/v1 to interact with the image API and downloads binary data from generated URLs to save images to the local filesystem.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) due to the processing of external API data.\n
Ingestion points: The script scripts/generate_image.py ingests JSON responses and image binary data from the remote subtoken.vip API.\n
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the processed API data (e.g., the revised_prompt field).\n
Capability inventory: The skill allows for local file writing and arbitrary outbound network requests via the Python script.\n
Sanitization: The skill does not sanitize text returned by the API before it is potentially used in subsequent agent turns.

subtoken-imagegen