The Agent Skills Directory

[SAFE]: The skill contains no malicious code, obfuscation, or credential exposure. It demonstrates good security posture by explicitly instructing the agent to refuse requests to read local filesystem paths (e.g., C:\ or /Users/...) and explaining that the agent cannot reach localhost, thereby preventing unauthorized local access or SSRF attempts.
[INDIRECT_PROMPT_INJECTION]: The skill defines a process for ingesting external media via public URLs, which represents a potential untrusted data surface.
Ingestion points: Public URLs processed by the download_media tool (SKILL.md).
Boundary markers: The instructions mandate waiting for track_progress confirmations and separate internal metadata registration from the processing of media bytes.
Capability inventory: Limited to media asset management tools (import_media, download_media, edit_item); no arbitrary code execution or filesystem write capabilities are exposed.
Sanitization: All ingested media is processed through a server-side transcoding and ASR pipeline that enforces browser-friendly codecs and resolution limits, mitigating risks associated with malicious media file headers.

asset-import