The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automates its workflow by executing local Python scripts provided within the skill package, such as 'scripts/inject_style.py', and referencing a utility script 'linyuebanzi-image-gen/scripts/generate.py'. These scripts are used to manage prompt styling and interface with the image generation backend.\n- [EXTERNAL_DOWNLOADS]: For specific artistic styles like 'cozy-handdrawn' and 'cartoon-infographic', the skill references a remote image asset at 'https://r2.cloudnative101.net/assets/katong.png' which serves as a reference for the generation model. This asset is hosted on the vendor's own infrastructure.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface (Category 8) as it ingests and processes untrusted user-provided articles to derive image content. Evidence chain: (1) Ingestion points: User articles provided via text or local markdown files (SKILL.md, README.md). (2) Boundary markers: No explicit delimiters or instructions are used to separate user content from prompt templates in 'prompt_template.md'. (3) Capability inventory: The skill has the capability to execute local scripts and perform network operations via the generation API. (4) Sanitization: No explicit filtering of the article content was observed during the prompt construction phase. However, the risk is assessed as low due to the narrow output domain (image generation).\n- [SAFE]: No malicious behaviors such as credential theft, persistent access mechanisms, or obfuscated payloads were identified. The skill's functionality is consistent with its stated purpose of enhancing technical articles with illustrations.

linyuebanzi-inline-diagram