The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes ffmpeg and ffprobe via node:child_process.execFileSync to perform media operations, including audio-video muxing, metadata correction, and format conversion. These operations are performed using structured arguments rather than raw shell strings.
[EXTERNAL_DOWNLOADS]: The skill instructions facilitate the use of npx create-video@latest for scaffolding Remotion projects, which involves downloading and executing trusted tooling from the official npm registry.
[DYNAMIC_EXECUTION]: The scripts record.mjs and smoke.mjs dynamically load the playwright module using paths provided via the PLAYWRIGHT_REQUIRE environment variable. This is a standard developer pattern for resolving local project dependencies in modular environments.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data via $ARGUMENTS and a generated beats.json configuration to drive browser automation. Risk is mitigated by a mandatory evidence chain: 1) Ingestion point: SKILL.md and beats.json; 2) Boundary markers: Usage of structured JSON config; 3) Capability inventory: Browser automation (click, type, navigate) via Playwright; 4) Sanitization: Implements a forbiddenLabel regex filter in record.mjs to block interactions with sensitive UI components like account settings or logout buttons.
[DATA_EXPOSURE]: The scripts/smoke.mjs file contains a dedicated checkPrivacy function that scans project files for absolute system paths, email addresses, and developer-specific identifiers to ensure no sensitive information is included in the output or committed to the repository.

create-video