transcribe-voice-memos
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill triggers local shell scripts and CLI tools using hardcoded absolute paths (e.g.,
/Users/mtm/pdev/taylormonacelli/valorousverdin/scripts/voice-memos.sh). This allows the agent to execute complex local workflows, including interacting with the clipboard and modifying environment variables likeMAX_FILE_SIZE_MBduring execution. - [INDIRECT_PROMPT_INJECTION]: The skill processes transcription data and logs derived from external voice memos. While it explicitly instructs the agent to ignore the content and only report on routing metadata, reading untrusted text files creates a potential vector for indirect prompt injection.
- Ingestion points: The agent reads transcription logs (
zephyr.log,fulcrum.log), audit records (audit.jsonl), and system logs (valorousverdin.log) located in the user's local share directory. - Boundary markers: The skill provides defensive instructions to "Never summarise, paraphrase, quote or characterise a memo," effectively attempting to treat the memo bodies as opaque data.
- Capability inventory: The agent has access to the
Bashtool, allowing it to execute local scripts, read sensitive log files, and manipulate the system clipboard. - Sanitization: No explicit sanitization or filtering of the transcript content within the logs is performed before the agent processes them.
Audit Metadata