The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-generated content from LMS discussion forums and has write permissions, establishing an indirect prompt injection surface.
Ingestion points: Discussion post bodies and replies are ingested through the list_discussion_entries and get_discussion_entry_details tools as described in SKILL.md.
Boundary markers: No structural prompt encapsulation or formatting boundaries are specified to keep data separate from instructions.
Capability inventory: The skill possesses powerful write capabilities, including reply_to_discussion_entry, post_discussion_entry, create_discussion_topic, create_announcement, and send_conversation tools.
Sanitization: External content is not sanitized or validated before presentation, although the risk is partially mitigated by mandatory user confirmation steps prior to executing write tools.

canvas-discussion-facilitator