biorxiv-search
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local shell and Node.js scripts (
scripts/searchandscripts/search.mjs) to interact with the Valyu API and process search results. It provides a bash-based path resolution mechanism to locate these scripts within the agent's plugin environment. - [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes full-text content from external biological preprints hosted on bioRxiv and retrieved via the Valyu API. This introduces a surface for indirect prompt injection if a preprint were to contain adversarial instructions designed to influence the agent's behavior during analysis.
- Ingestion points: Untrusted research content is retrieved from the
https://api.valyu.ai/v1endpoint. - Boundary markers: The skill processes data in a structured JSON format, which provides some implicit separation, but no explicit instructions are provided to the agent to disregard embedded commands in the paper content.
- Capability inventory: The agent has the capability to execute the skill's local scripts and potentially use other available tools on the processed data.
- Sanitization: There is no evidence of content sanitization or filtering for the retrieved article text.
Audit Metadata