The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Python subprocess module to orchestrate its workflow and perform image optimization.
Evidence: scripts/batch_article_images.py invokes other local Python scripts (generate_prompts.py, render_local_images.py) and scripts/compress_images.py calls system utilities pngquant and magick (ImageMagick) to compress output files.
[EXTERNAL_DOWNLOADS]: The skill downloads generated images from external URLs.
Evidence: scripts/siliconflow_generate.py performs HTTP GET requests via the requests library to download PNG files from URLs provided by the SiliconFlow API response.
[DATA_EXFILTRATION]: The skill transmits data to an external API provider as part of its primary function.
Evidence: When the --allow-api flag is used, scripts/siliconflow_generate.py sends prompts and local image data (in Base64 format) to api.siliconflow.cn to generate or edit illustrations.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted article content to generate image prompts.
Ingestion points: Untrusted data enters the agent's context through WeChat article links or pasted text as described in SKILL.md.
Boundary markers: The skill lacks explicit boundary markers or "ignore embedded instructions" warnings when interpolating article concepts into the final image generation prompts.
Capability inventory: The skill has the capability to perform network POST requests to an external API (scripts/siliconflow_generate.py) and write files to the local disk.
Sanitization: scripts/generate_prompts.py includes a visual_only_text function that rewrites certain terms to prevent text rendering in images, but this is not a security-focused sanitization mechanism.

handdrawn-article-illustrator