The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external CLI tools to perform media operations.
Evidence: resources/mpt/driver.py invokes ffmpeg via subprocess.run to generate synthetic video clips. Arguments are constructed from fixed values and resolution parameters.
Evidence: resources/strudel/render.mjs spawns a headless Chrome/Chromium process using node:child_process to host the music rendering environment.
[DYNAMIC_EXECUTION]: The music generation component evaluates code at runtime.
Evidence: resources/strudel/render.mjs reads pattern code from an external file and executes it within the headless browser context using the Chrome DevTools Protocol (Runtime.evaluate).
Evidence: resources/strudel/page.html contains the logic to receive and evaluate Strudel patterns using window.strudel.evaluate.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to drive a complex generation pipeline.
Ingestion points: User-provided text "briefs" are accepted via the oma video generate command as described in SKILL.md.
Boundary markers: The skill relies on formatting guidance in resources/prompt-tips.md and uses render-spec.json as a deterministic bus to separate asset generation from rendering.
Capability inventory: The skill can write files to the local workspace, execute subprocesses (ffmpeg, Chrome), and invoke other provider tools for voice, images, and slides.
Sanitization: resources/execution-protocol.md specifies that output paths must be restricted to the current working directory ($PWD) and that URL query/hash tokens must be masked in logs.

oma-video