The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it is designed to ingest and process untrusted external data.
Ingestion points: The skill uses create_workbook to process user-provided attachments and automatically fetches content from external websites, including SEC filings, PubMed, and ClinicalTrials.gov.
Boundary markers: There are no specific instructions or delimiters defined to prevent the agent from following instructions potentially embedded in the fetched research materials.
Capability inventory: The skill utilizes a set of MCP tools (create_workbook, ask, read_workbook, workbook_sources, recheck_sources) for data management and retrieval.
Sanitization: The instructions do not specify any validation or sanitization steps for the external data before it is processed by the agent.

biotech-diligence