The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches additional composition logic from the vendor's GitHub repository (github.com/okou-ai/okou-skills.git) if the required resources are not locally present. This is documented as a standard vendor resource update mechanism for the okou-ai author.- [DYNAMIC_EXECUTION]: The skill utilizes npx to execute specific, pinned versions of the hyperframes library (e.g., version 0.8.26). It also invokes a Python script for image processing during the quality assurance phase to generate tiled previews of video frames using standard libraries like json and subprocess.- [COMMAND_EXECUTION]: A comprehensive set of managed CLI tools is used for video operations, including okou video render for cloud-based rendering and okou __intro-video-agent for HeyGen API interactions. These tools rely on platform-managed credentials rather than hardcoded secrets. The skill also invokes local media utilities such as ffmpeg and ffprobe for file validation.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted user inputs (text and attachments) to create video content.
Ingestion points: User-supplied prompts, screen recordings, and various file attachments (PDF, DOCX, etc.) processed in SKILL.md Step 1 and references/input-preparation.md.
Boundary markers: The skill includes explicit instructions to treat attachment contents as source material and never as instructions, and employs highly structured prompt skeletons in references/prompt-compiler.md to maintain control.
Capability inventory: Capabilities include cloud rendering (okou video render), network API interaction, and local media processing using ffmpeg and python3.
Sanitization: Inputs are normalized into a strict brief structure before being interpolated into final prompts, with facts extracted and verified to prevent the interpretation of data as instructions.

intro-video