document-to-narration

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied markdown files to determine scene breaks and spoken adaptation. Maliciously crafted documents could potentially influence the agent's interpretation of scene boundaries.\n
  • Ingestion points: Markdown files read by scripts/split-to-scenes.ts and scripts/full-pipeline.ts.\n
  • Boundary markers: None; the agent is instructed to use rhetorical judgment.\n
  • Capability inventory: File system writes (audio and metadata) and local command execution (Python, ffmpeg).\n
  • Sanitization: The skill relies on natural language interpretation rather than strict schema validation.\n- [COMMAND_EXECUTION]: Orchestration scripts use Deno.Command to execute Python and external binaries (ffmpeg, whisper-cpp). These commands use array-based argument passing, which prevents shell injection vulnerabilities.\n- [EXTERNAL_DOWNLOADS]: The skill documentation references downloading the Whisper model and installation tools from the Remotion ecosystem, which is a well-known service for programmatic video production.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 12:30 PM
Security Audit — agent-trust-hub — document-to-narration