Device-Sequence Skill

Process steps in order. Do not skip ahead.

Answers "which physical thing does this device control?" by activating devices in a known order with a hold on each, so the person in the field binds every photo or observation to a device id instead of guessing. It builds on device-command — each command is validated against that device's real surface and dispatched over /device/runmethod — and adds the ordered walk, the timed hold, and live per-step narration.

A command is dispatched, not proven executed (rules/state-vs-attributes.md). The confirmation here is the operator watching the physical device during its hold — that is the whole point of the timed walk, not a return code.

Step 1 — Frame the walk

Establish the hub (--ip <addr> or --hub <name>), the ordered device list (--devices 1639,1640,1641 by id, or --names "Zone 1,Zone 2" by exact display name), the command to run on each (--command, default on), the per-device hold (--duration seconds), and an optional --off-command (e.g. off) run on each device after its hold before the next. Set the hold long enough for the person to reach and observe each device. Proceed to Step 2.

Step 2 — Confirm the physical run

This physically activates each device in turn — valves open, lights turn on, shades move. Confirm with the user which devices will fire, in what order, and that they are positioned to observe and record as the walk runs. For irrigation or anything with a real-world cost, confirm the command and duration before starting. Proceed to Step 3.

Step 3 — Run the sequence

python3 .tessl/plugins/jbaruch/hubitat-dev/skills/_scripts/hub_device_sequence.py --ip <addr> --devices <id,id,...> --command <name> --duration <seconds> [--off-command <name>]

Argument contract, exit codes, and output shape: skills/_scripts/hub_device_sequence.py module docstring. Live per-step narration ([i/N] <device>: <command>, then the hold) goes to stderr as the walk runs; the JSON summary lands on stdout at the end. A device that fails to dispatch is skipped without a hold and the walk continues. Proceed to Step 4.

Step 4 — Bind observations to device ids

Read the steps[] log: each entry carries the device_id, device_name, and whether the command dispatched. Bind each photo or field note to the device_id/device_name of the step that was firing when it was taken — that mapping is the deliverable. Re-run any dispatched:false step individually with Skill(skill: "device-command") to see the command's own error and verification. Finish here.