drugbank-search

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
  • COMMAND_EXECUTION (MEDIUM): The skill relies on shell scripts (scripts/search) that take natural language queries as arguments. If the underlying bash script or the AI agent does not strictly escape these inputs, it creates a vector for command injection (e.g., a query containing ; rm -rf /).
  • CREDENTIALS_UNSAFE (MEDIUM): The instructions explicitly direct the agent to pass the user's Valyu API key as a command-line argument: scripts/search setup <api-key>. Secrets passed via CLI arguments are visible in process monitors (like ps) and are often recorded in shell history files, making them accessible to other users or processes on the system.
  • INDIRECT_PROMPT_INJECTION (LOW): The skill ingests untrusted data from an external API (api.valyu.ai) and provides it to the agent.
  • Ingestion points: scripts/search output containing drug descriptions and interaction data from Valyu.
  • Boundary markers: None specified in the prompt templates.
  • Capability inventory: File discovery (find), shell execution (scripts/search), and network access (fetch).
  • Sanitization: No mention of sanitizing or escaping the retrieved content before it is processed by the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 05:30 PM