The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content such as transcripts, voice recordings, and notes, which creates a surface for indirect prompt injection where embedded instructions could attempt to bias the agent's analysis.
Ingestion points: Specifically designed to process external accounts, transcripts, and notes as defined in SKILL.md and references/artifact-shapes.md.
Boundary markers: Instructions in SKILL.md advise the agent to distinguish between user reports and verified artifacts, providing a conceptual boundary, but no formal data delimiters or technical sanitization are present.
Capability inventory: The skill does not contain instructions for command execution, file system modification, or network requests across any of its scripts.
Sanitization: No technical sanitization of the input text is implemented before the information is processed by the agent.

friction-inquiry