The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automatically installs the docling and docling[ocr] Python packages from PyPI if they are not already installed. These libraries are provided by IBM Research, a well-known technology organization, and are required for the skill's document processing features.
[COMMAND_EXECUTION]: The skill executes shell commands to manage its environment and perform conversions. This includes environment checks, package installation using the --break-system-packages flag, and the execution of a dynamically generated Python script located at .gemini/tmp/docling_convert.py.
[PROMPT_INJECTION]: This skill introduces a surface for indirect prompt injection because it processes untrusted files (PDF, Word, Excel, etc.) and returns the extracted text to the agent's context.
Ingestion points: External files are parsed into Markdown or JSON via the Docling engine as described in SKILL.md.
Boundary markers: Extracted document content is provided to the agent without specialized boundary markers or instructions to ignore embedded commands.
Capability inventory: The skill is capable of executing shell commands, installing external software, and performing file system writes.
Sanitization: No sanitization or safety filtering is performed on the text extracted from documents before it is ingested by the agent.

docling-converter