The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md file defines shell command templates (e.g., node scripts/generate_image.js -p "{prompt}") that interpolate raw user input directly into a command string. This pattern introduces a command injection risk if the agent's execution environment processes these templates in a shell without proper escaping or sanitization of metacharacters (e.g., backticks, semicolons, or pipes).
[DATA_EXFILTRATION]: The skill requires an APIYI_API_KEY and transmits it, alongside user-provided prompts and image data, to the third-party endpoint https://api.apiyi.com. This domain is not among the whitelisted or well-known services, meaning user credentials and content are sent to an external proxy infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted user data and pass it to scripts with network and file-system capabilities.
Ingestion points: Raw user input is used as the primary prompt and editing instruction in SKILL.md.
Boundary markers: Absent; the instructions explicitly tell the agent to use the "original complete input" directly in the command.
Capability inventory: The scripts scripts/generate_image.js and scripts/generate_image.py perform network POST requests and write image files to the local filesystem (fs.writeFileSync / open(..., "wb")).
Sanitization: Absent; the prompts are used as-is in API payloads and command construction without content validation or escaping.
[EXTERNAL_DOWNLOADS]: The Python implementation scripts/generate_image.py depends on the requests library. This is a common but non-standard third-party library that must be pre-installed in the execution environment.

nano-banana-pro-image-gen