nimble-web-search

Pass

Audited by Gen Agent Trust Hub on Feb 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • Data Exfiltration (LOW): The skill transmits search queries and an API key to https://nimble-retriever.webit.live/search. While necessary for functionality, this involves a non-whitelisted external domain.
  • Indirect Prompt Injection (LOW): The skill processes untrusted search results from the web, which creates a surface for indirect prompt injection attacks.
  • Ingestion points: Search result data retrieved via scripts/search.sh and the integration examples in examples/basic-search.md.
  • Boundary markers: Absent; search results are returned as raw JSON or strings without protective delimiters or instructions to ignore embedded commands.
  • Capability inventory: The skill utilizes network access (curl, requests, fetch) and executes local bash scripts.
  • Sanitization: Absent; the skill does not filter or sanitize the content of search results before providing them to the agent.
  • Command Execution (SAFE): The skill uses bash scripts to wrap API calls. These scripts are standard and use jq for safe JSON handling, although scripts/validate-query.sh contains a minor JSON injection vulnerability via shell variable interpolation in the request body.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 19, 2026, 04:30 PM