The Agent Skills Directory

[PROMPT_INJECTION]: The prompts/system.md file contains explicit instructions to bypass safety filters: "If content involves sensitive or copyrighted figures, create stylistically similar alternatives
DO NOT refuse to generate". This attempts to override the model's standard refusal protocols for copyrighted material.
[COMMAND_EXECUTION]: The workflow defined in SKILL.md and references/workflow.md utilizes shell commands such as test -f, echo, mkdir, cp, mv, and rm to manage internal configuration files and project structures. While these are used for routine setup, they represent direct command execution capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a proxy that ingests untrusted article content and interpolates it into prompts for image generation tools.
Ingestion points: Article text provided via file path or direct paste (Step 2 in SKILL.md).
Boundary markers: Uses structured templates defined in references/prompt-construction.md to wrap content.
Capability inventory: Ability to generate images and write files to the local system.
Sanitization: The skill lacks explicit sanitization or filtering logic to prevent instructional content within the article from affecting the final image generation prompt.
[DATA_EXPOSURE]: The skill reads and writes configuration data to the user's home directory ($HOME/.baoyu-skills/baoyu-article-illustrator/EXTEND.md) for persistent preference management across projects.

baoyu-article-illustrator