Home Assistant Dev Ecosystem

A development ecosystem for Home Assistant: eleven reference "skills" (firmware, integrations, add-ons, dashboards, theming, data, hardware, CI/CD) plus small Python tools that route between them, scaffold new projects, and search across them — built to be used from Claude Code in VS Code, but the skill files and templates are plain markdown/YAML/TS you can read on their own.

The goal: take a Home Assistant project from idea to working device/dashboard — ESPHome firmware → HA integration/automation → Lovelace card, optionally starting from custom PCB hardware — without re-explaining conventions and constraints in every session.

What's here

skills/        8 skill .md files + orchestrator.py (routing engine)
tools/         knowledge_search, router, project_generator, workflow_planner
config/        skills.json (skill/workflow/routing metadata)
templates/     5 scaffolds: ha_sensor, esphome_component, ha_dashboard,
               ci_cd_pipeline, pcb_design
examples/      worked, end-to-end walkthroughs
docs/          ARCHITECTURE, INTEGRATION, API_REFERENCE, TROUBLESHOOTING
.vscode/       tasks.json (task runner) + skills.code-snippets
CLAUDE.md      auto-read by Claude Code — skill map + routing rules

The 11 skills

Skill	What it's for
master_orchestrator	Routes a request to the right skill(s), plans multi-skill workflows
ha_developer	Home Assistant integrations, automations, entities, services (Python/YAML)
ha_addon	HAOS Supervisor add-ons: config.yaml, Dockerfile, bashio, ingress, /data storage
ha_data	Recorder, long-term statistics, InfluxDB/Grafana, data export for analysis/ML
esphome	Custom ESPHome firmware, sensors, components (C++/YAML)
lovelace_cards	Custom, responsive Home Assistant dashboard cards (TypeScript)
ha_design_system	HA design tokens, accessibility, responsive design rules
ha_themes	YAML frontend themes, card-mod, light/dark UI restyling
electrical_engineering	Circuits, schematics, power/thermal budgets, KiCad — for custom sensor hardware
vector_design	Icons, gauges, diagrams, SVG optimized for HA dashboards
github_actions	CI/CD for firmware builds, testing, releases

Eight of the eleven are Home Assistant–specific (ha_developer, ha_addon, ha_data, esphome, lovelace_cards, ha_design_system, ha_themes, plus master_orchestrator tying them together); the rest (electrical_engineering, vector_design, github_actions) support building the hardware, graphics, and pipelines a real HA project needs around that core.

Quick start

cd dev-ecosystem
python3 -m skills.orchestrator --list-skills
python3 -m skills.orchestrator "I want to build a temperature sensor for Home Assistant"
python3 -m tools.knowledge_search "esphome ble sensor"
python3 -m tools.project_generator ha_sensor my_first_sensor --dest ..

Requires only the Python 3.9+ standard library — see requirements.txt.

Using it with Claude Code in VS Code

• Keep CLAUDE.md at the workspace root — Claude Code reads it automatically and learns the skill map and routing rules.
• Keep .vscode/tasks.json and .vscode/skills.code-snippets at the root for the task runner and code snippets.
• For deep work, mention a skill file in chat: @skills/esphome/esphome_expert_agent.md

See docs/INTEGRATION.md for the full command list and Python API.

Example workflow

custom_ha_sensor (python3 -m skills.orchestrator --workflow custom_ha_sensor): ESPHome firmware → HA entity/automation → icon/gauge graphic → Lovelace card. Walked through end-to-end in examples/example_1_custom_sensor.md.

A note on secrets

Templates that need credentials (WiFi, API keys, HA tokens) ship as *.example.* files with placeholder values only — see templates/ha_sensor/firmware/secrets.example.yaml. Copy to the real filename and fill in your own values; .gitignore keeps secrets.yaml out of version control.

Verified

Every tool in this package was run against this exact tree before shipping: list-skills, list-workflows, skill/workflow detail, search (keyword + concept), router, project generation (with placeholder substitution), and workflow planning all produce correct output. See docs/API_REFERENCE.md for known limitations (keyword routing, raw relevance score).