The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from a reMarkable tablet, including OCR text and handwritten annotations, which can contain hidden instructions.
Ingestion points: Content enters via the remarkable_read tool (page text, highlights), remarkable_search (OCR text), and rendering of annotation layers via remarkable_image for transcription tasks defined in references/transcription-brief.md.
Boundary markers: The instructions lack specific delimiters or negative constraints to prevent the agent from obeying instructions embedded within the documents or notes it reads.
Capability inventory: The skill can execute local binaries via scripts/rm_mcp.py and has the ability to write files (such as PNG renders) to the local disk.
Sanitization: There is no mention of sanitization or filtering for the OCR text or transcription results before they are processed by the agent.
[COMMAND_EXECUTION]: The script scripts/rm_mcp.py uses subprocess.Popen to execute the remarkable-mcp server binary. It uses the REMARKABLE_MCP_CMD environment variable to determine the command to run, which facilitates the execution of local binaries for tool communication.

remarkable