download-media

Warn

Audited by Gen Agent Trust Hub on Jul 19, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The helper script scripts/download-media.sh appends all extra user-provided arguments directly to the yt-dlp command. This enables the use of flags such as --exec, which can execute arbitrary shell commands on the host after a download completes.
  • [CREDENTIALS_UNSAFE]: The skill's instructions specifically suggest using the --cookies-from-browser flag to access restricted content. This exposes the user's private browser session data to the tool and the agent.
  • [PROMPT_INJECTION]: The skill allows for indirect prompt injection by reporting untrusted media titles (fetched from the internet) as part of the result file paths. Ingestion points: Media titles and metadata from external URLs in scripts/download-media.sh. Boundary markers: None. Capability inventory: File system access and shell execution via bash and yt-dlp. Sanitization: No validation or sanitization is performed on the titles used in the output paths.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 19, 2026, 02:04 PM
Security Audit — agent-trust-hub — download-media