The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, untrusted data (meeting transcripts, Slack threads, and notes) through the $ARGUMENTS variable.
Ingestion points: The skill accepts external content in the SKILL.md template via the $ARGUMENTS placeholder.
Boundary markers: The skill employs <context> and </context> XML-style tags to delimit the untrusted input from the system instructions, reducing the risk of the model following instructions embedded within the meeting notes.
Capability inventory: The skill has no associated scripts, performs no file system writes, and does not initiate network requests. It is a pure text-to-text transformation skill.
Sanitization: There is no programmatic sanitization of the input, though the lack of executable capabilities significantly limits potential impact.

meeting-synthesis