skills/mr-kelly/skills/kelly-mv/Gen Agent Trust Hub

kelly-mv

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local Python scripts using the child_process.spawn method to perform machine learning inference for video and audio drafting. Arguments for these processes are derived from user-editable data in the Busabase database.
  • Evidence: scripts/gen_draft_video.mjs spawns a local Python environment to run an inference.py script.
  • Evidence: scripts/generate_song_draft.mjs spawns scripts/gen_song.py for audio generation.
  • [EXTERNAL_DOWNLOADS]: The skill is designed to communicate with external AI model APIs and a Busabase cloud instance for its core workflow.
  • Evidence: scripts/execute_generation_requests.mjs makes POST requests to a configurable image generation endpoint (defaulting to https://moonrouter.dev/v1).
  • [DATA_EXFILTRATION]: Scene descriptions, character reference images, and prompt metadata are transmitted to external third-party model providers for generation services.
  • Evidence: scripts/execute_generation_requests.mjs sends prompt strings and image buffers to the configured image API base URL.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection as it ingests data from external Busabase records and interpolates them into model prompts without sanitization.
  • Ingestion points: The cast and shots record fields in the associated Busabase Bases.
  • Boundary markers: Absent; prompts are constructed via string concatenation of description fields.
  • Capability inventory: The skill has the capability to spawn local processes and perform network exfiltration.
  • Sanitization: Not implemented for the interpolated description or role fields.
  • [REMOTE_CODE_EXECUTION]: The skill facilitates the execution and usage of remote machine learning models and configurations.
  • Evidence: The skill configuration points to external model weights such as mlx-community/SongGeneration-v2-large.
  • [DYNAMIC_EXECUTION]: The skill dynamically assembles execution commands for subprocesses based on data retrieved from its database environment.
  • Evidence: Argument array assembly in scripts/gen_draft_video.mjs and scripts/generate_song_draft.mjs.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 10:36 AM
Security Audit — agent-trust-hub — kelly-mv