Homey

**Quick judgement:** A capable and fairly mature smart-home control skill for Athom Homey, well-suited to technically proficient users who want an AI agent to interact with their Homey ecosystem via CLI and JSON. It offers broad device/flow control and flexible local/cloud connectivity, but it operates in a high-risk physical domain (real-world devices) and requires careful security and access scoping. **Strengths** - Wide coverage of Homey features: list and control devices, query capabilities and zones, trigger flows, and obtain snapshots for AI planning. - Designed for agents: consistent `--json` output, snapshot mode, and fuzzy matching make it easier for LLMs to discover and target devices. - Local + cloud support using official APIs and tokens, appropriate for both home-network and remote/VPS setups. **Key risks / limitations** - **Physical impact risk:** The skill can directly control lights, thermostats, locks, and other actuators. Misuse, misinterpretation of user intent, or prompt injection could cause unwanted or unsafe changes in the home environment. - **Fuzzy matching hazards:** Typo-tolerant and substring matching can select the wrong device/flow if names are similar, particularly dangerous for security-relevant devices (e.g., locks, garage doors). Agents should prefer IDs or confirm targets before critical actions. - **Security & privacy:** Requires storing and using Homey local API keys or cloud tokens. If these are compromised or the agent is overly permissive, an attacker could gain remote control of the home. - **Operational complexity:** Requires Node.js ≥ 18, `homeycli` installation, and manual auth setup. This is fine for power users but may be challenging for less technical operators. **Recommended scenarios** - Self-hosted or tightly controlled agents that act as smart-home assistants for a trusted user or household. - Technical users who can manage tokens, networking (LAN/VPN vs cloud), and OS-level safeguards (e.g., restricting which commands the agent may call). - Environments where the operator can define clear policies, like: read-only for sensors by default, explicit confirmation for locks or alarming actions, and preference for device IDs over fuzzy name matching. Given its capabilities and domain, this skill is powerful but should be deployed with **strict access controls, careful prompt design, and human oversight** for safety-critical actions.