The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's recovery procedures instruct the agent to use pip install to fetch and install Python packages such as pdfplumber, pypdf, pymupdf, pypdfium2, and marker-pdf if they are missing from the environment. These dependencies are not pinned to specific versions, which introduces a potential supply chain risk during runtime acquisition.
[COMMAND_EXECUTION]: The workflow relies heavily on the terminal tool to execute Python scripts (pdf_read.py, pdf_split.py, pdf_page_image.py, extract_pymupdf.py, extract_marker.py) located in productivity skill directories. This provides a mechanism for system-level interaction triggered by document processing.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize large, untrusted documents, which creates an attack surface for instructions embedded within the PDF content.
Ingestion points: Untrusted PDF data enters the agent context through tools like pdf_read.py, extract_pymupdf.py, and the platform's native read_file functionality as described in SKILL.md and references/hermes-pdf-limits.md.
Boundary markers: The instructions do not include explicit delimiters or warnings to ignore potentially malicious instructions embedded within the processed text.
Capability inventory: The skill utilizes high-impact capabilities including terminal for shell command execution, pip for software installation, and delegate_task for sub-agent management.
Sanitization: There is no evidence of sanitization, filtering, or validation of the extracted document content before it is processed by the agent or sub-agents.

omh-long-document-reading