The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes authoritative instructions like "CRITICAL OUTPUT RULE," "HARD OUTPUT RULE," and "This rule overrides any model default" to ensure the AI agent follows specific formatting requirements, such as generating exactly one image per section. While these are used for stylistic control, they represent a targeted attempt to override standard model behavioral defaults.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a vulnerability surface where untrusted user input from design briefs influences the generation of images and the naming of files in the project workspace.
Ingestion points: External user briefs and site descriptions are processed as instructions for the design engine.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore potentially malicious commands embedded within user-supplied briefs.
Capability inventory: The skill directs the agent to perform file system operations, specifically writing new Markdown files and appending data to existing logs (index.md, log.md) within the llmwiki/ directory.
Sanitization: There is no explicit sanitization for the user-supplied <ten> placeholder used in generated filenames, creating a minor risk of unintended file path manipulation depending on the agent's internal tool implementation.

imagegen-frontend-web