The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The skill references an unverified external library mediabunny in rules/asset-lifecycle.md and rules/captioning-workflow.md. This package is recommended for audio extraction and metadata calculation, but its source and safety cannot be verified from the provided context.
[DATA_EXFILTRATION] (MEDIUM): The captioning-workflow.md describes a pipeline that extracts audio and uploads it to external cloud storage (S3). This creates a data egress path. If an agent implementing these rules processes untrusted remote video URLs, it could be leveraged to exfiltrate audio data to attacker-controlled storage.
[PROMPT_INJECTION] (MEDIUM): The skill handles untrusted external data (remote video assets) and provides patterns for network operations (S3 uploads, fetching assets), creating a surface for indirect prompt injection. 1. Ingestion points: Remote URLs processed in rules/asset-lifecycle.md and audio extraction in rules/captioning-workflow.md. 2. Boundary markers: Absent; the code snippets do not include delimiters or instructions to ignore embedded commands in external content. 3. Capability inventory: fetch() for data retrieval and uploads in rules/asset-lifecycle.md and rules/captioning-workflow.md. 4. Sanitization: Absent; no explicit sanitization or validation logic is provided for external URLs or transcription outputs.

remotion-editor