The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses sensitive local application data containing private user information.
The script scripts/extract-apple-voice-memos-metadata queries a private SQLite database located at ~/Library/Group Containers/group.com.apple.VoiceMemos.shared/Recordings/CloudRecordings.db to extract recording metadata (titles, dates, durations, and file paths).
The script scripts/extract-apple-voice-memos-transcript accesses and reads content from .m4a files within the user's private library to extract embedded transcripts.
[PROMPT_INJECTION]: Potential for indirect prompt injection via the processing of untrusted audio transcripts.
Ingestion points: Transcripts extracted from user recordings by scripts/extract-apple-voice-memos-transcript are ingested into the subagent prompt.
Boundary markers: The skill uses a ## Transcript heading in PROMPT.md as a delimiter but lacks explicit instructions for the subagent to ignore potentially malicious commands spoken within the audio.
Capability inventory: The skill uses Python scripts to read local files and can save processed output to the filesystem as markdown files.
Sanitization: No sanitization or filtering is performed on the transcript text before it is sent to the subagent for processing.

apple-voice-memos