The Agent Skills Directory

[PROMPT_INJECTION]: The skill generates templates that process untrusted user input without implementing boundary markers or sanitization, creating a vulnerability to indirect prompt injection.
Ingestion points: The generated files question.md, plan.md, and plan-build-improve.md all interpolate the $ARGUMENTS variable directly into the prompt context.
Boundary markers: The generated instructions lack delimiters (e.g., XML tags or triple quotes) or specific system instructions to ignore potential commands embedded within user-provided text.
Capability inventory: The generated workflows enable powerful capabilities, including filesystem modification (Edit, Write) and command-line execution (Bash), which could be abused if an injection occurs.
Sanitization: No input validation or escaping mechanisms are provided in the scaffolding logic to sanitize the arguments before they are written to the target files.

create-expert