The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external media and project data which could host malicious instructions. Ingestion points include finished renders and project databases. Boundary markers are present as the skill mandates human-in-the-loop review for all QC results and forbids automated gate clearing. Capabilities include render job management and media file operations. Sanitization is performed by normalizing codec and format inputs against a live matrix.
[COMMAND_EXECUTION]: The skill depends on external system utilities for file analysis. Evidence includes the requirement for ffmpeg and ffprobe to be available on the system PATH for the deliverable and media modules.

resolve-delivery