edge-iot

Pass

Audited by Gen Agent Trust Hub on Oct 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill implements communication patterns that ingest untrusted data from external sources, such as MQTT telemetry streams and Cloudflare Worker HTTP endpoints. This creates an attack surface for indirect prompt injection if the resulting data is subsequently interpreted by an AI agent without adequate filtering.
  • Ingestion points: SKILL.md defines multiple data entry points via MQTTClient.subscribe (e.g., telemetry and command topics) and the Cloudflare Worker's fetch handler for /ingest POST requests.
  • Boundary markers: The code snippets do not include explicit boundary markers or instructions designed to prevent the AI from following commands embedded within the sensor data or telemetry payloads.
  • Capability inventory: The skill possesses the capability to perform network requests via fetch, store data in Cloudflare KV/Durable Objects, and issue hardware commands via MQTT publishing.
  • Sanitization: The logic focuses on JSON parsing and functional processing, lacking specific sanitization routines to detect or neutralize malicious natural language instructions hidden in technical data fields.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 8, 2026, 09:40 PM