The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file config/secrets.md contains a hardcoded API key (sk-1p6RyJLH3e3TnfMEfN9clcwZru2V513pAIwK8h5fMVUTDnEv) for the AI Router service. Additionally, scripts/upload-image.ts and references/图床上传.md contain a hardcoded API key for the freeimage.host service.
[COMMAND_EXECUTION]: The skill scripts contain multiple instances of unsafe shell execution where user-controlled or filesystem-derived paths are directly interpolated into command strings. Specifically:
scripts/merge-to-long-image.ts uses exec to run convert (ImageMagick) with file paths from the directory listing.
scripts/upload-image.ts uses execSync to run curl commands for multiple image hosting services with the image path argument.
[DATA_EXFILTRATION]: The skill includes functionality to upload local images to external public hosting services (sm.ms and freeimage.host) via command-line tools. This could result in the unintended exposure of local sensitive images if used improperly.
[EXTERNAL_DOWNLOADS]: The scripts/analyze-image.ts script downloads images from arbitrary URLs provided as arguments to the local filesystem for processing.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from markdown files in the prompts/ directory. This content is interpolated into API requests for image generation without sanitization. The skill environment also possesses powerful capabilities including network access, file system writes, and subprocess execution, which could be exploited through malicious prompt data.

pw-image-generation