The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted meeting transcripts and events through get_transcript and get_events tools. It possesses capabilities to speak (Text-to-Speech), send_chat in meetings, and send notifications to the requester based on this external content.
Ingestion points: Meeting transcript segments and events are retrieved from the Telnyx API (SKILL.md).
Boundary markers: The skill specifies literal and semantic matching logic with boundary-safe phrase checks, but does not explicitly define delimiters to prevent the LLM from obeying instructions embedded in the speech.
Capability inventory: The skill can execute speak, send_chat, leave_meeting, and post-meeting reporting (SKILL.md).
Sanitization: Instructions include explicit filtering for API keys and require 'clear transcript evidence' for triggers, but lack general data sanitization for instruction bypass.

telnyx-meeting-bot