remarkable
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The PowerShell scripts Add-RemarkableCustomTemplate.ps1 and Install-RemarkableTemplate.ps1 interact with the reMarkable tablet via SSH and SCP to manage custom templates. This interaction is central to the skill's purpose and follows standard device management practices.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted handwritten notes. * Ingestion points: Handwritten notes from the tablet are converted to PNG images via pdf_to_pages.py and subsequently read by the vision model. * Boundary markers: Transcripts are organized with explicit page-based headings as defined in the reading guidelines (reading-pages.md). * Capability inventory: The skill possesses capabilities to write files and execute shell commands on the tablet via SSH in Add-RemarkableCustomTemplate.ps1 and Install-RemarkableTemplate.ps1. * Sanitization: Detailed guidelines in reading-pages.md mandate literal transcription, use of [illegible] tags, and forbid interpreting or acting on the content, serving as a defense against embedded instructions.
Audit Metadata