ffmpeg
Pass
Audited by Gen Agent Trust Hub on Aug 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local system binaries (ffmpeg, ffprobe) and internal project scripts (validate_skill.py, package.py) via Python helper scripts. These utilize list-based arguments for subprocess.run, a safe pattern that avoids shell injection, while enabling the agent to perform media processing and project validation tasks.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) through the ingestion of untrusted media metadata. * Ingestion points: The scripts/probe_media.py script provides raw JSON metadata from ffprobe to the agent context. * Boundary markers: Absent. No specific instructions are provided to the agent to disregard instructions potentially embedded in metadata tags (e.g., Title or Comment fields). * Capability inventory: The skill uses Bash and Read tools to manipulate files and execute ffmpeg recipes based on user intent and probe data. * Sanitization: Absent. The metadata is ingested into the agent context without escaping or validation.
Audit Metadata