nature-academic-search
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes paper titles, abstracts, and metadata fetched from external academic databases. This content could potentially contain malicious instructions intended to influence the agent's behavior when summarized or analyzed.
- Ingestion points: External search results are retrieved from PubMed, CrossRef, arXiv, and OpenAlex in
mcp-server/sources/andscripts/academic_search.py. - Boundary markers: The skill does not explicitly implement boundary markers or instructions for the agent to ignore embedded commands in the retrieved academic content.
- Capability inventory: The skill has the ability to perform network operations and write files to the local file system (e.g., citation exports via
scripts/format-converter.py). - Sanitization: The skill implements basic HTML tag stripping in
scripts/converters.py(ris_escapefunction) to sanitize metadata for citation formats. - [DYNAMIC_EXECUTION]: The skill uses dynamic module loading within its test suite for mocking and patching purposes.
- Evidence:
mcp-server/tests/test_sources.pyutilizes the__import__function to load source modules dynamically during unit testing. This is a standard testing practice and does not affect the runtime search server.
Audit Metadata