Scaffold Skill

Process steps in order. Do not skip ahead.

The always-on Hubitat rules (sandbox-constraints, app-lifecycle, driver-lifecycle, logging-conventions, state-vs-attributes, groovy-gotchas) are already in context — apply them while generating. The capability contract lives in reference/capabilities.json; the import allow-list in reference/allowed-imports.txt.

Step 1 — Determine what to build

Establish, asking the user only for what is not already clear:

• app or driver
• name, namespace (unique — typically the user's handle), author
• for a driver: the device protocol (Zigbee, Z-Wave, LAN/cloud, virtual) and the capabilities it exposes
• for an app: what it watches and what it controls

Proceed to Step 2.

Step 2 — Resolve the capability contract (driver) or UI shape (app)

For a driver, for each declared capability read its required commands and attributes from reference/capabilities.json (capabilities.<Name>.commands[].name, .attributes[].name). Every required command becomes a Groovy method. Marker capabilities (Actuator, Sensor) add no methods.

For an app, decide the preference inputs (see reference/input-types.md) and the events to subscribe to.

Proceed to Step 3.

Step 3 — Generate the skeleton

Emit one Groovy file wiring in, per the rules:

• Driver: metadata { definition(...) { capability lines, custom command/attribute } preferences { logEnable/txtEnable } }; installed(), updated() (with runIn(1800, logsOff) when logging), logsOff(); a method for every required command; configure()/refresh() if those capabilities are declared; a parse(String description) with the protocol-appropriate decode and a log.debug of raw input when a protocol is involved.
• App: definition(...), preferences { page { section { inputs } } }, installed() { updated() }, updated() { unsubscribe(); initialize() }, initialize() with the subscriptions, and a handler method for each subscription. Guard chatty logs on logEnable/txtEnable.

Write the file to the path the user wants (default <name>.groovy). Proceed to Step 4.

Step 4 — Self-check with the linter

Run the sandbox linter on the generated file and resolve anything it flags:

python3 .tessl/plugins/jbaruch/hubitat-dev/scripts/hub_lint.py <file.groovy>

Contract and finding shape: scripts/hub_lint.py module docstring. A correct skeleton should lint clean; if a missing-command or unresolved-handler appears, a required method or handler is absent — add it. If no findings, say so and proceed.

Step 5 — Offer to deploy

The skeleton exists locally but is not on any hub. Offer to deploy it with Skill(skill: "deploy"). Finish here.