video-color-grade
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill frequently executes external commands (
ffmpegandffprobe) viasubprocess.run. This behavior is expected and necessary for its primary purpose of video processing. The implementation uses argument lists rather than shell strings, mitigating common command injection risks. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data files such as
looks.jsonand video metadata. This represents a potential attack surface where maliciously crafted configuration files could attempt to influence the agent's behavior or execute unexpected filter chains in FFmpeg. - Ingestion points: External configuration files (
looks.json,grade-plan.json) are loaded inscripts/gradelib.pyand processed by various scripts. - Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are used when processing these files.
- Capability inventory: Extensive use of
subprocess.runacross multiple scripts (e.g.,gradelib.py,apply_grade.py) to execute FFmpeg commands with parameters derived from the JSON input. - Sanitization: While the skill performs schema validation in
gradelib.normalize_spec(), it does not sanitize or escape the FFmpeg filter strings contained within the configuration files.
Audit Metadata