The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (emails and calendar events), which is an inherent part of its functionality but presents a surface for potential indirect prompt injection attacks. \n
Ingestion points: Data is retrieved using Superhuman_Mail.list_threads and Superhuman_Mail.query_email_and_calendar calls as defined in SKILL.md. \n
Boundary markers: No explicit delimiters are used to separate user-controlled email content from the system prompts. \n
Capability inventory: The skill allows for drafting emails via Superhuman_Mail.create_or_update_draft and modifying thread states via Superhuman_Mail.update_thread. \n
Sanitization: No explicit sanitization or filtering of external content is defined.

eod-wrapup