use-api-gateway

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill dynamically ingests a live service catalog from an external API, which defines the available capabilities and their associated schemas, creating a surface for indirect instructions to influence agent behavior.
  • Ingestion points: Service catalog fetched from https://gateway.usebuy.ai/v1/catalog.
  • Boundary markers: Includes a 'Non-negotiable boundaries' section with explicit rules for handling external data and obtaining user consent.
  • Capability inventory: File reads, shell execution via curl and npx, and specialized MCP tools (buy_curl, buy_discover).
  • Sanitization: The instructions mandate strict adherence to the inputSchema provided by the gateway and the rejection of extra provider-specific inputs.
  • [EXTERNAL_DOWNLOADS]: Fetches runtime configuration and metadata from the gateway's catalog endpoint and uses npx to execute the @celo/buy package directly from the registry.
  • [COMMAND_EXECUTION]: Instructs the agent to execute shell commands including curl for interacting with the API gateway and the @celo/buy CLI tool for quoting and purchasing services.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 03:40 PM