The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process paper abstracts, titles, and full text from external academic sources such as arXiv, Semantic Scholar, and PubMed.
Ingestion points: The search.py, paper.py, and refTree.py scripts ingest data from various web APIs and web pages.
Boundary markers: The current implementation does not appear to wrap external content in delimiters or include instructions for the agent to ignore embedded natural language commands.
Capability inventory: The skill can perform network requests, execute shell commands for helper tools, and write output to local files.
Sanitization: Content is cleaned of HTML tags and normalized for whitespace, but no semantic filtering is applied to detect or strip potential injection attempts within the scholarly text.
[COMMAND_EXECUTION]: The skill executes external binaries to support its functionality.
scripts/arxiv_crawler_search.py and scripts/semantic_scholar_crawler_refTree.py invoke node to execute camoufox-js for generating browser launch options.
scripts/arxiv_pdf_paper.py invokes the pdftotext system utility (if found) to perform text extraction from PDF files.
[DYNAMIC_EXECUTION]: The unified entry points (search.py, paper.py, refTree.py) use importlib.import_module to load specific provider implementations at runtime.
The modules are selected from a static map (PROVIDER_GROUPS) defined within the source code, which mitigates the risk of loading arbitrary external modules.
[EXTERNAL_DOWNLOADS]: The skill relies on external packages and environment setup.
requirements.txt specifies several standard libraries along with deepxiv-sdk, a vendor-specific package from the skill author.
The crawler functionality requires a one-time setup of Playwright browsers (python3 -m playwright install firefox) and camoufox-js via Node.js.

sn-search-academic