video-qa

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The runtime path ingests outsider-authored free text only insofar as the producer uploads generation-log.md, shot-list.md, and the locked/reference sheets/clips they choose, then performs prompt rule scans and continuity logic by reading those user-provided markdown files.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly tells the agent it can run system-level install commands (e.g., "the agent can check which ffmpeg, run brew install ffmpeg itself"), includes a sudo apt install line and a remote Homebrew install via curl, which pushes the agent to perform privileged package installs and execute remote install scripts on the host.