hyperframes
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from websites, GitHub Pull Requests, and audio transcripts to generate video content, creating a surface for indirect prompt injection.
- Ingestion points: Website content is extracted into
extracted/visible-text.txt(film.md), Pull Request data is read viaghAPI (film.md), and audio is converted to text viatranscribe(captions.md). - Boundary markers: The instructions lack explicit delimiters or "ignore instructions" warnings for the ingested text, though they include guidance for the agent to review content before use.
- Capability inventory: The skill possesses significant local capabilities, including shell execution through the
plugin-cli.mjslauncher and headless browser control via Puppeteer inbake-basemap.mjs. - Sanitization: Scripts like
bake-basemap.mjsuseJSON.stringifyto interpolate parameters into JavaScript, but raw external text is generally passed through for processing. - [COMMAND_EXECUTION]: The skill relies on shell command execution for core functionality. It uses a launcher script (
plugin-cli.mjs) to invoke thehyperframesCLI vianpx. Additionally,bake-basemap.mjsandlocate.mjsexecuteffmpegandffprobefor media processing. - [EXTERNAL_DOWNLOADS]: The skill fetches resources from trusted and well-known services. It downloads map tiles from Esri (
arcgisonline.com) and Carto (cartocdn.com) inbake-basemap.mjs, and font assets from Google Fonts and library components from JSDelivr. - [DATA_EXFILTRATION]: The skill transmits usage telemetry via its
eventscommand, sending the skill name and account-linked metadata to its backend. This behavior is disclosed in the instructions and respects user opt-out settings.
Audit Metadata