The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes data from external files to generate design elements.
Ingestion points: The skill reads page specifications from the specs/pages/ directory (SKILL.md).
Boundary markers: There are no instructions or delimiters defined to prevent the agent from executing instructions potentially embedded within these specification files.
Capability inventory: The agent has access to the use_figma tool and can execute shell commands via the memi CLI (prototype, spec, ia) (SKILL.md).
Sanitization: No sanitization or validation of the content within the specs/pages/ directory is performed before the agent uses it to build Figma components.
[COMMAND_EXECUTION]: The skill utilizes a custom command-line interface, memi, to perform various actions such as generating prototypes and creating information architecture specs.
Evidence: Commands such as memi prototype, memi spec page, and memi ia create are used throughout the workflow (SKILL.md).

figma-prototype