The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions to bypass AI safety guidelines regarding sensitive or copyrighted content. In references/base-prompt.md, it directs the agent: 'If content involves sensitive or copyrighted figures, create stylistically similar alternatives
DO NOT refuse to generate'.
[INDIRECT_PROMPT_INJECTION]: The skill workflow is vulnerable to indirect prompt injection due to its processing of untrusted external content into high-priority instructions for downstream tools.
Ingestion points: External article content and user-provided reference descriptions (SKILL.md Step 1, references/workflow/reference-images.md).
Boundary markers: The template in references/workflow/prompt-template.md lacks delimiters or specific warnings to ignore instructions embedded in the user content.
Capability inventory: The skill can write files, execute shell commands (e.g., test, cp), and call image generation skills.
Sanitization: No input sanitization is defined; instructions prioritize 'exact' reproduction of titles and 'MUST INCORPORATE' directives for reference analysis elements.
[COMMAND_EXECUTION]: The skill uses shell commands for environment discovery and file management. For example, SKILL.md and references/config/first-time-setup.md use test -f to check for configuration files and cp to store reference images in the project structure.
[DYNAMIC_EXECUTION]: The skill generates a dynamic prompt script (prompts/cover.md) at runtime using a template, which is then interpreted by an image generation tool to produce the final output.

tuzi-cover-image