parallel-literature-search
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill allows for local command execution via
scripts/parallel_search.py, which is the intended primary functionality for users wishing to run the search locally. The script takes user-supplied queries through--queryand interacts with external APIs. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from PubMed, Perplexity, and RAG knowledge bases.
- Ingestion points: Data is fetched from the NCBI E-utilities API, Perplexity web search results, and RAG chunks from AstraDB within
scripts/parallel_search.py. - Boundary markers: None identified. The search results are directly interpolated into a large prompt sent to the LLM for synthesis in
synthesize_results. - Capability inventory: The skill includes file-writing capabilities (
output_path.write) and network operations to PubMed/Anthropic APIs. - Sanitization: There is no evidence of sanitization, filtering, or escaping of the content retrieved from external sources before it is passed to the LLM for analysis.
- [DYNAMIC_EXECUTION]: The script uses
json.loadson content generated by the LLM (simulated results). While not directly executing code, this pattern of parsing AI-generated JSON into script logic is a lower-risk form of dynamic handling.
Audit Metadata