The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and act upon potentially untrusted data from user comments and component configuration schemas.
Ingestion points: The agent is directed to use canvas-comments-list and canvas-comments-retrieve to gather feedback from users, as well as canvas-layout-get to read existing placement configurations in SKILL.md.
Boundary markers: The instructions lack explicit guidance on using delimiters or ignoring embedded instructions when processing data from comments or configuration fields.
Capability inventory: The agent has extensive capabilities to modify the environment, including canvas-layout-patch for layout modifications, canvas-layout-publish for restructuring, and canvas-create for generating new components.
Sanitization: There is no requirement mentioned for sanitizing or validating user-provided strings before they are used to influence the agent's authoring or configuration tasks.

composing-grid-canvases