The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill integrates user-provided content via the $ARGUMENTS variable without utilizing delimiters or protective instructions to distinguish data from commands. This architectural pattern allows for potential manipulation of agent behavior if a user provides instructions disguised as a startup idea.
Ingestion points: User-supplied input via $ARGUMENTS in SKILL.md.
Boundary markers: The skill lacks explicit separators (like triple quotes or XML tags) or instructions to ignore embedded commands within the startup idea.
Capability inventory: The skill is granted Read and Edit permissions for the founder/ directory, along with WebSearch and WebFetch tools.
Sanitization: There are no instructions or mechanisms defined to sanitize or validate the content of the $ARGUMENTS variable before processing.

product-brief