video-downloader — agentic threat model

Not certain from the listing — The listing does not specify which foundation model is used to parse user requests or drive the skill. Standard LLM risks like prompt injection could lead to tool misuse, such as downloading malicious files or writing to restricted paths.

Not certain from the listing — There is no mention of vector databases, RAG, or training data operations. The primary data handled is transient video streams and local disk writes.

The agent orchestrates a bundled downloader tool to fetch remote media and write files to disk. The primary threat is tool misuse or insecure tool integration, where malicious URLs or crafted metadata could exploit the underlying downloader (e.g., command injection in yt-dlp) or cause arbitrary file writes.

Not certain from the listing — The hosting environment, sandboxing, and network isolation controls are not specified. Without strict sandboxing, writing files to disk and fetching remote media poses a severe risk of host compromise, local file overwrite, or SSRF.

Not certain from the listing — No logging, monitoring, or guardrails are described. There is a risk of blind spots regarding what URLs are being accessed and what files are being written to the host system.

Not certain from the listing — No authentication, authorization, or compliance policies are mentioned. There is a risk of unauthorized users abusing the agent to download copyrighted material or perform denial-of-service attacks via large file downloads.

Not certain from the listing — There is no mention of multi-agent coordination or marketplace interactions. However, if integrated into a larger ecosystem, other compromised agents could exploit this skill to exfiltrate data or write malicious payloads to disk.