mijia-automation

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-defined requirements and reads existing automation graphs to generate complex JSON structures for smart home controllers. * Ingestion points: The agent reads existing configuration data from the Mijia gateway using tools like mijia_get_graph and processes natural language user input. * Boundary markers: The instructions in references/patterns/adapters-and-templates.md explicitly mandate the desensitization of sensitive information related to door locks and personnel before data is used in templates. * Capability inventory: The agent has the ability to modify physical smart home logic via mijia_create_graph and mijia_patch_graph. * Sanitization: The workflow enforces the use of dedicated validation tools (mijia_validate_graph_capabilities and mijia_validate_graph) to verify permissions and structural integrity before any changes are committed.
  • [COMMAND_EXECUTION]: The skill describes a developer workflow involving local synchronization scripts, specifically scripts/rules.mjs. These are used to pull or push rule definitions between the local environment and the gateway. This represents a standard development capability within the local project scope and does not involve remote code execution or untrusted source ingestion.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 01:52 PM