The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded API keys were identified in the release documentation (release/bananapro-image-gen/SKILL.md) and within the help text of the main execution script (scripts/generate_image.py). Specifically, the keys sk-HoyC9xe8aB0CRC6j4EvuM0zRcN9TTYnIMjACMCeGvd95tuhQ and sk-WY9Rx5caemo39xIMN2OCNQ4cD3L0RDtKmC9Q1rot9rcKxgow are exposed.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection and path traversal.
Ingestion points: The --prompt and --filename arguments in scripts/generate_image.py take direct user input from the agent's command line.
Boundary markers: No boundary markers or validation logic are present to distinguish between instructions and data.
Capability inventory: The generate_image.py script possesses capabilities to perform network requests and write files to the local file system.
Sanitization: There is no sanitization or path validation for the --filename argument. The script uses Path(filename) and Path.mkdir(parents=True) which allows a malicious prompt to specify relative paths (e.g., ../../.ssh/id_rsa) to overwrite sensitive system files.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to https://apipro.maynor1024.live and other dynamic URLs returned by the API to fetch and download image content.
[COMMAND_EXECUTION]: The skill operates by executing the scripts/generate_image.py Python script, which handles network communication and file system operations.

Banana Pro Image Generation