wjs-voicedrop-post-processing

Pass

Audited by Gen Agent Trust Hub on Aug 9, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes article content retrieved from the VoiceDrop database via MCP tools, which introduces an indirect prompt injection surface where untrusted data could attempt to influence the agent.
  • Ingestion points: Article text is fetched using the voicedrop MCP tool based on identifiers provided in the $ARGUMENTS field.
  • Boundary markers: The instructions do not define explicit delimiters or 'ignore' directives to isolate the fetched article content from the agent's instructions.
  • Capability inventory: The agent has permissions to read articles and create new article versions through the VoiceDrop MCP interface.
  • Sanitization: No specific sanitization or validation logic for the fetched article content is implemented beyond basic presence checks.
  • [SAFE]: The skill incorporates 'Iron Laws' that explicitly prohibit high-risk and destructive operations, such as publishing, deleting, or modifying community settings, ensuring the agent operates within a restricted safety scope.
  • [SAFE]: All infrastructure components mentioned, such as the launchd poller (com.jianshuo.voicedrop-postprocess) and the voicedrop MCP tools, are associated with the vendor's own platform and represent standard automation patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 9, 2026, 05:57 PM
Security Audit — agent-trust-hub — wjs-voicedrop-post-processing