The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes citation data (BibTeX) and cite dialog HTML directly from Google Scholar search results, which are untrusted external sources.
Ingestion points: Browser fetch results from scholar.google.com and content read from scholar.googleusercontent.com (Step 1a and 1c in SKILL.md).
Boundary markers: The instructions do not provide specific delimiters or instructions to ignore embedded commands within the fetched citation text.
Capability inventory: The skill can write files to /tmp, execute a local Python script via bash, perform network downloads via Python's urllib, and interact with local services via HTTP.
Sanitization: The agent is instructed to parse the BibTeX into a specific JSON schema before passing it to the Python script, which provides a layer of structured data validation.
[COMMAND_EXECUTION]: The skill uses the bash tool to execute a local Python script (push_to_zotero.py) located at a hardcoded absolute path (E:/gscholar-skills/...). This script processes citation data and interacts with the local environment.
[EXTERNAL_DOWNLOADS]: The push_to_zotero.py script uses urllib.request to download PDF files from external publisher URLs identified during the Google Scholar search. This is a core feature of the skill but involves outbound network requests to arbitrary domains.
[DATA_EXPOSURE]: The skill communicates with the Zotero Connector API hosted on the local loopback address (127.0.0.1:23119). This is a legitimate interaction with a local service required for the skill's functionality.

gs-export