The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill is vulnerable to shell command injection via user-provided file paths. \n
Evidence: In SKILL.md sections 3 and 4, the instructions direct the agent to execute ffprobe and ffmpeg using unvalidated variables like VIDEO_PATH and INPUT (e.g., ffprobe -v quiet ... "VIDEO_PATH"). \n
Risk: A malicious user could provide a filename containing shell metacharacters (e.g., ; rm -rf / ;) which could lead to arbitrary code execution on the host system. \n- [PROMPT_INJECTION] (HIGH): The skill has a significant indirect prompt injection surface due to its core functionality. \n
Ingestion points: Video file frames in SKILL.md Step 4 and 5. \n
Boundary markers: None present to distinguish video content from system instructions. \n
Capability inventory: Subprocess execution via ffmpeg/ffprobe, file system writes/reads. \n
Sanitization: None. \n
Risk: The sub-agent prompt explicitly asks the AI to extract prompts/commands the user typed from the video. An attacker could embed visual instructions in a video (e.g., on a slide) that the agent then transcribes and obeys during the final synthesis phase (Step 6), potentially leading to unauthorized actions. \n- [EXTERNAL_DOWNLOADS] (MEDIUM): The skill uses an external installer from an unverified source. \n
Evidence: README.md recommends npx skills add fabriqaai/ffmpeg-analyse-video-skill. \n
Risk: The author fabriqaai is not in the trusted sources list, posing a risk of supply chain attack or execution of malicious skill code during installation.

ffmpeg-analyse-video