The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): The skill is highly vulnerable to Indirect Prompt Injection through multiple ingestion points.\n
Ingestion points: rules/calculate-metadata.md and rules/compositions.md demonstrate fetching data from user-provided URLs (props.dataUrl, props.videoId) and incorporating the response into the rendering pipeline.\n
Boundary markers: Absent. The skill provides no instructions for sanitizing or isolating external content before processing.\n
Capability inventory: The skill possesses network access (fetch) and the ability to bundle/execute React code (@remotion/bundler) that is directly influenced by external inputs.\n
Sanitization: Absent. Data from external fetches is directly used to drive component props and rendering metadata.\n- [REMOTE_CODE_EXECUTION] (HIGH): The skill provides instructions for the deployment and execution of code on cloud infrastructure.\n
Evidence: rules/rendering.md outlines procedures for npx remotion lambda functions deploy and renderMediaOnLambda. In combination with untrusted inputs, this enables a path for executing malicious logic in high-privilege cloud-hosted environments.\n- [EXTERNAL_DOWNLOADS] (MEDIUM): The skill triggers downloads of numerous external packages and assets from unverified sources.\n
Evidence: Use of npx remotion add for various plugins and fetching Lottie animations from lottiefiles.com. None of these sources are within the trusted scope defined in the analysis framework.\n- [COMMAND_EXECUTION] (MEDIUM): The skill instructs the agent to run complex CLI commands and set up local servers.\n
Evidence: rules/rendering.md includes examples for setting up an Express server to process rendering requests, which creates an exposed execution surface on the host machine.

remotion-video-toolkit