The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the process for retrieving and handling transcription data from an external API, which creates a surface for indirect prompt injection attacks if the audio being transcribed contains instructions designed to manipulate the agent.
Ingestion points: Transcription text is retrieved through the sentences field in the transcript GraphQL query defined in SKILL.md.
Boundary markers: None are defined in the instructions to help the agent distinguish between API data and its own system instructions.
Capability inventory: No script files or automated tool executions are included in the skill, reducing the immediate risk of exploitation.
Sanitization: The documentation does not provide guidelines for sanitizing or escaping the transcription content before it is processed by the model.
[NO_CODE]: The skill consists entirely of documentation and metadata without any accompanying scripts or executable code.

fireflies