The Agent Skills Directory

[SAFE]: The skill provides structured guidance and templates for research workflows such as brainstorming and gap analysis. All instructions are focused on academic best practices and no malicious override or safety bypass attempts were detected.
[INDIRECT_PROMPT_INJECTION]: The skill interacts with external data by searching the web and reading PDF content via Zotero integration, creating an inherent surface for indirect prompt injection from academic sources. However, it explicitly defines defensive workflows to mitigate this risk. 1. Ingestion points: Web search results and PDF full-text extraction (referenced in references/zotero-integration-guide.md). 2. Boundary markers: Employs structured templates like Evidence Record and Research Question Card (defined in references/research-contract.md) to isolate and categorize external claims. 3. Capability inventory: Limited to managing the user's Zotero library and performing academic searches (SKILL.md). 4. Sanitization: Implements a Claim Promotion Gate and Source Trust Levels logic to ensure evidence is verified and prevent speculative content from being treated as factual (references/research-contract.md).
[SAFE]: Tool usage and external references are limited to well-known academic databases (arXiv, Google Scholar) and the Zotero MCP server. No unauthorized network activity or credential harvesting was detected.

research-ideation