The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill orchestrates workflows that process third-party untrusted data, exposing an indirect prompt injection surface.
Ingestion points: SKILL.md indicates that workflows like /product-launch-video and /website-to-video crawl external product and website URLs using headless Chrome, while /pr-to-video reads external GitHub pull request data.
Boundary markers: Absent; no specific boundaries or instructions to isolate or disregard untrusted instructions within the crawled text are defined in SKILL.md.
Capability inventory: SKILL.md notes capabilities including headless Chrome execution, GitHub CLI (gh) commands, and local Whisper transcription.
Sanitization: Absent; no sanitization, escaping, or validation logic for the ingested content is specified in this routing configuration.
[EXTERNAL_DOWNLOADS]: SKILL.md references mechanisms for checking, downloading, and updating framework components from external repositories.
Evidence: The documentation describes using npx skills add heygen-com/hyperframes and npx hyperframes skills update to retrieve or refresh components from GitHub.

hyperframes