The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines multiple shell command templates using ffmpeg and ffprobe to perform technical analysis on video renders, including frame extraction at specific intervals, transition strip generation, and audio loudness measurement.\n- [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection because it ingests untrusted data from the user and interpolates it into shell commands without specified sanitization.\n
Ingestion points: SKILL.md (Standalone mode accepts arbitrary video files; In-studio mode receives project paths and revision hashes); evidence-packet.md (accepts $VIDEO, $OUT, and $SFX as shell variables).\n
Boundary markers: No delimiters or safety instructions are provided to the agent to isolate the user-provided paths or filenames from the command logic.\n
Capability inventory: evidence-packet.md performs subprocess calls to ffmpeg and ffprobe using the provided templates.\n
Sanitization: The instructions do not define any validation or escaping mechanisms for variables used in shell commands, relying on the agent to handle them safely.\n- [DYNAMIC_EXECUTION]: The skill generates and executes shell commands at runtime to produce the evidence packet required for the review process.

motion-critique