obsidian-recording-to-meeting-minutes
Pass
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is designed to ingest and prioritize instructions from a local 'AGENTS.md' file within the user's vault, which can influence the agent's behavior. It also processes transcripts derived from audio recordings, which presents a surface for indirect prompt injection.\n
- Ingestion points: Local 'AGENTS.md' configuration and meeting audio recordings.\n
- Boundary markers: The skill employs modular chunking and subagents but lacks explicit instructions to ignore embedded commands in the source data.\n
- Capability inventory: Shell execution (scribe), file system read/write access, and Python script execution.\n
- Sanitization: Includes a dedicated 'QA / audit subagent' phase to verify the final note against evidence before writing to the vault.\n- [COMMAND_EXECUTION]: Executes shell commands to run the 'scribe' transcription utility and uses an inline Python script to perform structural validation on the final markdown output.\n- [EXTERNAL_DOWNLOADS]: Fetches and installs the 'scribe' tool from the trailofbits/scribe GitHub repository using the 'uv' package manager or tool runner.\n- [DATA_EXFILTRATION]: Accesses audio recording files and existing notes within the Obsidian vault. The skill instructions specifically highlight the sensitivity of this data and advise against sharing full transcripts in public tools.
Audit Metadata