yeelight-interactive-light-experiences

Pass

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted visitor input that is interpolated into prompts for an external AI provider. * Ingestion points: User data is received via the /api/experience/run endpoint in scripts/server.mjs. * Boundary markers: Visitor input is serialized into JSON and appended to the prompt in scripts/lib/provider.mjs. * Capability inventory: The executor triggers light control commands through a subprocess call to the yeelight-home binary in scripts/lib/command-adapter.mjs. * Sanitization: Input values are truncated and filtered by the safeInput utility in scripts/lib/plans.mjs to limit the injection surface.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 2, 2026, 01:29 PM
Security Audit — agent-trust-hub — yeelight-interactive-light-experiences