canvas

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The skill explicitly instructs the agent to connect via the Membrane CLI to external Canvas instances and run actions like list-discussions, get-submission, and list-assignments that ingest user-generated/untrusted Canvas content (discussion posts, submissions, announcements) as part of its workflow, so third-party content can influence agent decisions.