Japanese Smart Home Command Normalizer

Normalize short Japanese smart-home voice transcripts after STT into safe structured intents and slots. Use when handling Japanese commands for lights and air conditioners, especially when transcripts contain STT drift such as れーぼー, だんぼー, そうふう, エアコントメテ, or 電気化して.

Use this skill when a short Japanese STT transcript needs to be normalized before smart-home execution.

Workflow

1. Read references/design.md for the normalization pipeline and result shape.
2. Read references/domains.md for the supported domains and vocabulary.
3. Reuse lib/normalize.js as the core pure module.
4. Use scripts/demo.js to try sample transcripts from the terminal.
5. Integrate the normalized result into a device-control skill such as switchbot-light or a hook such as audio-router.

Current domains

• light
  • device aliases: 電気, ライト, 照明
  • actions: on, off
• aircon
  • device aliases: エアコン
  • actions: on, off, set_mode
  • modes: cool, heat, dry, fan

Notes

• This skill only normalizes and classifies text. It does not call device APIs.
• Prefer fixed vocabulary plus lightweight fuzzy matching over open-ended LLM interpretation for safety-critical home actions.
• When confidence is low or required slots are missing, return needsConfirmation: true instead of auto-executing.
• Add future devices by extending the domain vocabulary, not by piling more ad-hoc regex into callers.

Resources

• lib/normalize.js: core normalization and classification module.
• scripts/demo.js: print normalized results for sample inputs.
• fixtures/samples.json: sample transcripts and expected outcomes.
• references/design.md: pipeline, API shape, and confidence rules.
• references/domains.md: supported vocabulary and extension guidance.
• references/openclaw-integration.md: thin-hook integration guidance.