edit-video
Pass
Audited by Gen Agent Trust Hub on Nov 15, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes several external binaries including
ffmpeg,ffprobe, andyt-dlpto perform media processing and metadata extraction. These commands are invoked using secure patterns (passing arguments as arrays tospawnandspawnSyncvialib/process.mjs), which prevents shell injection vulnerabilities. - [EXTERNAL_DOWNLOADS]: The
scripts/reference-analyze.mjsutility usesyt-dlpto download video content from external URLs provided by the user. This is a core feature for analyzing reference media and is performed within the project workspace. - [INDIRECT_PROMPT_INJECTION]: As the skill ingests external data (video metadata and content), it is potentially susceptible to indirect prompt injection. However, the skill implements boundary markers in its instructions (e.g., distinguishing
analysis-onlyfromreusablecontent) and uses a robustassertInsideutility inlib/paths.mjsto ensure file operations do not escape the designated workspace. - [CREDENTIALS_UNSAFE]: The skill demonstrates proactive security by including a 'Never store credentials' warning in its
assets/preferences.template.mdfile, guiding users to avoid placing sensitive information in project metadata.
Audit Metadata