FFmpeg Video Editor
Warn
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The instructions in
SKILL.mdcontain hidden zero-width space characters (U+200B) used to obfuscate code block delimiters in the response format section. - [COMMAND_EXECUTION]: The skill provides a command template for concatenation (
ffmpeg -f concat -safe 0) that uses the-safe 0flag, which bypasses security restrictions and allows FFmpeg to read files from absolute or out-of-bounds paths. - [PROMPT_INJECTION]: There is a discrepancy between the author 'aaaaqwq' and the owner metadata 'mahmoudadelbghany' in
_meta.json, which constitutes deceptive metadata. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. 1. Ingestion points: Natural language user requests described in
SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: Generation of FFmpeg shell commands. 4. Sanitization: Absent. This surface allows untrusted input to influence generated command parameters.
Audit Metadata