The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the node interpreter to measure reference videos and render templates (e.g., node <hyperframesCli> measure).
[EXTERNAL_DOWNLOADS]: The skill is directed to download reference media from user-provided URLs using host tools like curl or wget for local analysis.
[PROMPT_INJECTION]: The instructions contain directives that override default interactive behaviors, such as "Work from start to finish without asking questions," prioritizing autonomous completion over human-in-the-loop safety checks.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes and recreates external, untrusted content (videos and links), establishing a surface where instructions embedded in the reference material could potentially influence the agent's logic.
Ingestion points: User-provided video files and reference URLs (SKILL.md).
Boundary markers: The skill contains a directive to "Learn from observable behavior rather than guessing the original author's code" (SKILL.md).
Capability inventory: Execution of shell commands via node, file system writes to asset and project directories, and use of network download tools.
Sanitization: No specific sanitization or validation logic is defined for the input media content before processing begins.
[DYNAMIC_EXECUTION]: The executable tool path (hyperframesCli) is retrieved dynamically at runtime from external project metadata or MCP resources rather than being statically defined, leading to command execution from a computed location.

motion-video-to-template