The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches data from external academic sources, primary source databases (Sefaria, CDLI, ORACC), and web discovery tools (JSTOR, Google Scholar). This content is processed and presented to the agent, creating a surface where third-party content could theoretically contain instructions intended to influence the agent.
Ingestion points: scripts/fetch_sefaria.py, scripts/fetch_cuneiform.py, and scripts/ane_stealth_fetch.sh (via the obscura browser tool).
Boundary markers: No explicit boundary markers are added by the scripts when formatting retrieved content.
Capability inventory: The skill uses urllib.request for API calls, the mistralai Python package for OCR, and calls the external obscura binary for web scraping.
Sanitization: The scraping script ane_stealth_fetch.sh truncates extracted text to specific character limits (e.g., 2000-3000 characters).
[DYNAMIC_EXECUTION]: The shell script scripts/ane_stealth_fetch.sh utilizes the --eval functionality of the obscura browser tool to execute hardcoded Javascript snippets for data extraction from various academic websites. This is a functional requirement for its scraping capabilities and is limited to specific, pre-defined DOM manipulation logic.

Ancient Near East Research