video-editing
Pass
Audited by Gen Agent Trust Hub on Jun 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill generates and instructs the agent to execute FFmpeg commands for processing video files, such as trimming segments, concatenating clips, and normalizing audio. This is the intended core functionality for the video editing pipeline.
- [REMOTE_CODE_EXECUTION]: The skill uses Python scripts with the
requestslibrary to interact with external APIs from ElevenLabs and fal.ai. This is used for generating voiceovers and assets. The implementation correctly manages credentials by referencing environment variables (os.environ["ELEVENLABS_API_KEY"]) rather than hardcoding them. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes potentially untrusted video transcripts in Layer 2 (Organization) to influence the generation of edit decision lists and shell commands.
- Ingestion points: Video transcripts and recording logs analyzed by the agent in Layer 2.
- Boundary markers: Absent; no specific delimiters or warnings are provided to prevent the agent from following instructions embedded within the processed transcripts.
- Capability inventory: Shell command execution via FFmpeg and network operations via API requests to ElevenLabs and fal.ai.
- Sanitization: Absent; the skill does not include instructions for validating or sanitizing the transcript content before it is used to scaffold commands.
Audit Metadata