target-drug-asset-bp
Pass
Audited by Gen Agent Trust Hub on Aug 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user-provided research descriptions and external publications fetched via tools like ls_paper_search. This ingestion creates a vulnerability where malicious instructions embedded in research documents could potentially influence the agent's summary or evaluation. * Ingestion points: Research project descriptions, publication titles/DOIs, and target/mechanism descriptions in SKILL.md and workflow.md; external data fetched via ls_translational_medicine_search and ls_paper_vector_search. * Boundary markers: The skill uses structured Markdown templates for output as defined in output-templates.md, but does not implement explicit delimiters to isolate or ignore instructions that might be contained within the ingested research data. * Capability inventory: The skill calls specialized MCP tools for information retrieval and analysis but does not possess capabilities for arbitrary file writing, shell command execution, or system modification. * Sanitization: The workflow employs ls_ner_nor_normalize to extract and normalize scientific entities, which provides data structure but is not designed to filter out adversarial prompt instructions.
- [COMMAND_EXECUTION]: The skill orchestrates a variety of Model Context Protocol (MCP) tools to gather pharmaceutical and legal data. These tools, such as ls_target_fetch, ls_drug_search, and ls_patent_fetch, are domain-specific resources used correctly within the skill's intended purpose of research assessment.
Audit Metadata