The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from web search results and academic PDFs to generate research summaries and proposals.
Ingestion points: The skill ingests web search snippets and paper metadata via WebSearch, as well as PDF full-text extracted via the zotero_get_item_fulltext tool as described in references/zotero-integration-guide.md.
Boundary markers: The skill utilizes logical verification steps such as the 'Claim Promotion Gate' in references/research-contract.md to evaluate evidence strength, but lacks explicit structural delimiters to isolate raw external text from agent instructions during the ingestion process.
Capability inventory: The skill has the ability to write local markdown files (e.g., research-proposal.md) and modify the user's Zotero library through the MCP server tools.
Sanitization: No explicit sanitization or escaping of external content is specified before interpolation into research artifacts.
[EXTERNAL_DOWNLOADS]: The skill automates the retrieval of academic papers and PDF attachments from well-known academic repositories and services.
Evidence: The zotero_add_items_by_identifier and zotero_find_and_attach_pdfs tools fetch literature from academic registries and services (such as arXiv and Unpaywall) as documented in references/zotero-integration-guide.md.

research-ideation