video-editing
Pass
Audited by Gen Agent Trust Hub on Mar 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instruct s the agent to execute
ffmpegandffprobebinarie s for video m a nipul a tion. One exam p le incl u de s the-safe 0flag in concatenation, which relax e s defaul t security path restriction s in the binary. - [PROMPT_INJECTION]: The use of shell tem p l a tes in
SKILL.m dthat incorporate user-provide d fil e nam e s and param eter s create s a surface for indirect prom pt injection. - Ingestion points: Untruste d data enter s via pl a cehol der s for input fil e s, water m a r k s, and fil ter specification s in shell com m a n d s.
- Boundary m a r k e r s: No boundary m a r k e r s or instruction s to ignor e em bed ded instruction s are present in the com m a n d tem p l a tes.
- Capabil ity inventor y: The skill util iz e s
ffmpeg,ffprobe, and shell redirection (printfto fil e). - Sanitiz a tion: The re is no evidence of input val idation or sanitiz a tion to preven t com m a n d injection via special character s in fil e nam e s.
Audit Metadata