video-shots

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface because it processes external video files and structured JSON data that could be manipulated by an attacker.
  • Ingestion points: Untrusted data enters the agent context via video files processed by ffmpeg/ffprobe and through the shots.json file read using readFileSync in scripts/video-shots.mjs.
  • Boundary markers: Absent for LLM context. No explicit delimiters or instructions are used to separate user-supplied metadata strings from prompt instructions when the agent reads the breakdown log.
  • Capability inventory: The skill possesses subprocess execution capabilities via execFileSync (invoking ffmpeg and ffprobe), and its manifest allows Bash, Read, Write, and Edit tools.
  • Sanitization: HTML entity escaping via the esc() function is applied during the HTML report generation to prevent cross-site scripting (XSS), but no prompt sanitization or natural language instruction filtering is performed on the parsed text fields.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 01:35 AM
Security Audit — agent-trust-hub — video-shots