The Agent Skills Directory

[COMMAND_EXECUTION]: The Google provider implementation (scripts/providers/google.ts) utilizes execFileSync to execute the curl command. This is used as a legitimate workaround for proxy-related connectivity issues when using the default fetch implementation.
[DATA_EXFILTRATION]: The skill transmits user-provided text prompts and reference images to various third-party AI providers (e.g., Google, OpenAI, Replicate). This behavior is the documented and intended primary function of the tool.
[DATA_EXPOSURE]: Authentication for the supported AI services is handled via environment variables and local configuration files (.env, EXTEND.md). This aligns with industry standards for secure credential management in development environments.
[INDIRECT_PROMPT_INJECTION]: The tool can ingest prompt data from local files via the --promptfiles flag. While this presents an attack surface for indirect prompt injection, it is a standard functionality for tools that process external text for LLM prompts.

baoyu-imagine