b2a-agents

Pass

Audited by Gen Agent Trust Hub on Mar 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The code template references external Node.js packages including @lucid-agents/core, @lucid-agents/http, @lucid-agents/hono, @lucid-agents/payments, and zod.
  • [DATA_EXFILTRATION]: The template includes network operations via the fetch API to external domains (api.example.com) to retrieve data.
  • [PROMPT_INJECTION]: The skill architecture designs agents that ingest and process data from external sources, presenting a surface for indirect prompt injection.
  • Ingestion points: External data retrieved via fetch in the 'lookup' and 'aggregate' handlers in SKILL.md.
  • Boundary markers: None present; the template does not use delimiters to isolate external API content from agent instructions.
  • Capability inventory: Network access (fetch), potential payment processing via @lucid-agents/payments.
  • Sanitization: Input parameters are validated with Zod, but the payload received from external APIs is returned to the agent context without further sanitization or escaping.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 4, 2026, 02:15 PM