The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from the OpenTable provider, which creates an attack surface for indirect prompt injection.
Ingestion points: Data enters the context via the opentable connector through commands like lookup-rid and search-availability as defined in SKILL.md.
Boundary markers: The instructions guide the agent to interpret and summarize data in plain language, which serves as a basic abstraction, but no strict structural boundaries (e.g., specific delimiters) for external content are defined in the skill logic.
Capability inventory: The skill possesses capabilities for significant state-changing actions, including book-reservation, modify-reservation-with-lock, cancel-reservation, and scheduling reminders via cron.add.
Sanitization: The skill incorporates data minimization instructions, specifically limiting reservation requests to user-provided dietary or seating needs and using authorized contact details from the platform.

opentable