The Agent Skills Directory

[COMMAND_EXECUTION]: Multiple rule files (e.g., rules/3d.md, rules/audio.md, rules/display-captions.md, rules/fonts.md, rules/lottie.md) instruct the agent or user to execute shell commands such as npx remotion add <package> to install framework-specific plugins and dependencies.
[EXTERNAL_DOWNLOADS]: The skill recommends the installation and use of several Node.js packages from external registries to support features like 3D rendering, media processing, and typography.
[INDIRECT_PROMPT_INJECTION]: The skill implements patterns for ingesting and processing untrusted data from external sources, which constitutes a vulnerability surface for indirect prompt injection.
Ingestion points: Data enters the context via rules/calculate-metadata.md (fetching from props.dataUrl), rules/import-srt-captions.md (parsing SRT subtitle files), and rules/lottie.md (fetching Lottie JSON animation data).
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard natural language commands that might be embedded within these external data files.
Capability inventory: The skill environment allows for shell command execution (npx), network operations (fetch), and media rendering tasks.
Sanitization: The provided code examples do not include sanitization, validation, or schema-enforcement for the data fetched from remote URLs before it is utilized.
[DATA_EXFILTRATION]: Functional examples in rules/calculate-metadata.md and rules/compositions.md demonstrate fetching data from URLs provided via component properties (e.g., props.dataUrl). If an attacker can influence these properties, this pattern could be leveraged to exfiltrate local identifiers or state information to an external endpoint.

remotion-best-practices