The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted meeting transcripts and notes through the distill.py script. These notes are then interpolated into a follow-up email draft and a structured summary. An attacker could embed instructions within the meeting transcript (e.g., in a fake speaker tag or comment) that the AI agent might follow when it processes the resulting summary or draft.
Ingestion points: The distill.py script reads transcript data from a file path specified by the --in argument or via stdin.
Boundary markers: The skill does not implement explicit boundary markers or instructions to the model to ignore embedded commands within the transcript data.
Capability inventory: The skill uses argparse for command line input, pathlib for file reading/writing, and re for pattern-based extraction. It does not perform network operations or execute arbitrary system commands beyond basic file I/O.
Sanitization: The script performs cleaning by stripping timestamps and speaker tags via regex, but it does not sanitize the remaining content for potential injection attacks before rendering it into the final output.

meeting-notes-distiller