The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation references the official llmwhisperer-client library for Python and JavaScript (npm), as well as the n8n-nodes-unstract package for workflow integration.
[COMMAND_EXECUTION]: Provides a Docker command example for running an MCP server (unstract/mcp-server), which is a legitimate configuration step for developers using the Unstract platform.
[DATA_EXFILTRATION]: The skill facilitates the transmission of document data (PDFs, images, Office files) to external processing endpoints at llmwhisperer-api.us-central.unstract.com and llmwhisperer-api.eu-west.unstract.com. This data transfer is the intended functional purpose of the OCR service and is performed via authenticated API requests.
[PROMPT_INJECTION]: The skill processes external document content, which presents a surface for indirect prompt injection.
Ingestion points: Document uploads and URL-based extraction specified in SKILL.md and references/api_reference.md.
Boundary markers: None described in the provided documentation.
Capability inventory: Includes network operations to Unstract APIs and file reading via client libraries.
Sanitization: Not explicitly mentioned in the documentation for extracted text.

llmwhisperer