The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process untrusted external data from local documents and web URLs, which is converted to text and provided to the agent for analysis.
Ingestion points: The extract-text command in tools/eyeball.py is used to read external content (PDFs, Word docs, web pages) into the agent's context.
Boundary markers: The skill does not provide specific delimiters or instructions to ignore embedded commands within the processed text, increasing the risk of the agent following instructions found inside the analyzed material.
Capability inventory: The agent has the ability to execute shell commands (via the Python helper), write files to the user's Desktop, and perform network requests to fetch external URLs.
Sanitization: No sanitization or filtering of the extracted text is performed before it is presented to the agent.
[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to install several third-party Python libraries (pymupdf, pillow, python-docx, playwright, pywin32) and download browser binaries (playwright install chromium) to enable its functionality. These are well-known packages from established registries.
[COMMAND_EXECUTION]: The skill workflow relies on executing a Python utility (eyeball.py) from the command line. This utility further spawns subprocesses to automate Microsoft Word (on macOS and Windows), call LibreOffice for document conversion, and manage browser processes for web rendering.

eyeball