manage-ac

name	manage-ac
description	Operate the home's Sensibo ACs from the terminal: read current state and sensor data, change power/mode/target temperature/fan/swing, and automate with local rules, cloud schedules/timers, or Climate React. Use when the user asks to turn the AC on or off, set the temperature or mode, check the room temperature/humidity, or automate the AC.

Manage AC (sensibo)

The installed sensibo console command (PyPI dist sensibo-cli) is the sole interface to the fleet. This skill is a quick reference; the CLI's own docs are authoritative and always current:

• sensibo learn — self-teaching prompt: every noun, the safety model, the exit-code policy.
• sensibo explain <noun> [verb] — detailed markdown for any path (e.g. sensibo explain set, sensibo explain rule arm). When unsure of a flag, read explain instead of guessing.

One-shot wrapper (scripts/ac.sh)

For explicit control actions, prefer the wrapper — one command resolves the target pod (single-pod fleets need no id) and commits in one shot:

bash .claude/skills/manage-ac/scripts/ac.sh status              # fleet (live) + latest stored readings
bash .claude/skills/manage-ac/scripts/ac.sh on                  # power on the (single) pod
bash .claude/skills/manage-ac/scripts/ac.sh off
bash .claude/skills/manage-ac/scripts/ac.sh set 22              # target 22°
bash .claude/skills/manage-ac/scripts/ac.sh mode cool
bash .claude/skills/manage-ac/scripts/ac.sh fan auto
bash .claude/skills/manage-ac/scripts/ac.sh read                # live snapshot
bash .claude/skills/manage-ac/scripts/ac.sh on --dry-run        # preview only
bash .claude/skills/manage-ac/scripts/ac.sh status --json       # single JSON document
bash .claude/skills/manage-ac/scripts/ac.sh set 22 --json      # structured result on stdout

The pod argument is optional (resolved from sensibo devices --json when the fleet has exactly one pod) or all for the whole fleet; on <pod-id> and on all both work. --json is accepted by every verb: results go to stdout, diagnostics (including the "Running:" line) to stderr, and status --json emits one merged JSON document. status makes exactly one fleet API call — readings come from the offline store, so an empty store prints a hint instead of polling; read stays the live single-location snapshot.

Safety contract

• Every write verb is dry-run by default; --apply commits. A set without --apply prints what it would do and changes nothing. Show the user the dry-run output and commit with --apply only when the user asked for the change (an explicit "turn the AC on" is that request).
• The wrapper is one-shot by design. scripts/ac.sh on|off|set|mode|fan runs sensibo set ... --apply, so a single call changes the AC — that is the point of the wrapper. Use it for explicit control requests; preview with --dry-run. Do not make it preview-by-default or add a --no-apply flag.
• rule arm requires a fresh rule dry-run of the rule's current definition, and the engine enforces a minimum off-time (10-minute floor) so a rule cannot short-cycle a compressor.
• Every command takes --json; in JSON mode errors emit {"code","message","remediation"} on stderr, and stdout and stderr never mix. Exit codes: 0 success, 1 user error, 2 environment error, 3+ reserved.

Read (always safe)

sensibo devices                    # whole fleet: pods + nested Room Sensors, one API call
sensibo read <location-id>         # one live snapshot (pod id or ms_* Room Sensor id)
sensibo query locations            # every known sensing location in the local store
sensibo query latest               # latest reading per location (offline)
sensibo query latest <loc> --field temperature
sensibo query range <loc> --field temperature --since 2026-08-01   # offline time series
sensibo room list                  # locations, flagging stale sensors

query never touches the network (local store); read and devices poll the Sensibo cloud.

Write (dry-run by default)

For one-shot control use the wrapper above; the raw CLI form is:

sensibo set <pod-id> --power on                              # dry-run: prints intent
sensibo set <pod-id> --mode cool --target 24 --fan auto      # dry-run
sensibo set <pod-id> --target 22 --apply                      # commits
sensibo set --all --power off                                # fleet-wide dry-run

--mode is cool|heat|fan|dry|auto; --fan and --swing values are device-specific (see sensibo explain set). --all applies the same change to every pod instead of one.

Naming

sensibo room name <location> <alias>            # preview only
sensibo room name <location> <alias> --apply    # persist

A location's stable id, current alias, or Sensibo room name all identify it; ambiguous matches fail listing every candidate. Aliases are accepted in rule thresholds (--when-location) and other location arguments that accept a name, and a rename never orphans history (readings stay keyed on the stable id).

Automation

• Local rules — this machine evaluates the local store and drives the AC:

  sensibo rule add --name hotday --pod <pod-id> --power on --mode cool --target 23 \
      --when-location <loc> --when-field temperature --when-op '>' --when-value 28
  sensibo rule dry-run hotday     # evaluate NOW, read-only (required before arm)
  sensibo rule arm hotday
  sensibo rule run --once         # or --daemon for continuous evaluation
  sensibo rule list               # every rule with its armed / dry-run state
  sensibo rule disarm hotday
  

• Cloud — executes in Sensibo's cloud, so it holds when this machine is off: sensibo schedule create ..., sensibo timer set ..., sensibo smartmode enable (Climate React). All dry-run by default; --apply commits. See sensibo explain schedule|timer|smartmode.

Prefer rule for conditions that react to sensor readings on this machine; prefer schedule/timer for time-based behaviour that must survive this machine being off.

Scripts

Script	Purpose
scripts/ac.sh	One-shot control wrapper: status, read, on, off, set <temp>, mode <m>, fan <level> (all taking [pod|all] and --json; control verbs also take --dry-run). Resolves the single-pod target automatically; control verbs commit via sensibo set ... --apply. status costs one fleet API call (readings come from the offline store).

Red flags

• Never commit a write (--apply) without the user asking for that change; when in doubt, show the dry-run and ask.
• Never toggle power in a loop or on a timer with the one-shot verbs — ac.sh on/off is for deliberate single actions. Automated compressor start/stop must go through sensibo rule, which enforces the minimum off-time (10-minute floor) that the direct set path intentionally does not, so a rule cannot short-cycle a unit.
• Never bypass the arm contract (fresh dry-run before rule arm) or lower the minimum off-time floor.
• Never poll faster than ~60 s, and never loop one call per device — one devices call covers the whole fleet (unpublished rate limit).
• Don't hard-code pod ids in scripts or docs — resolve them from sensibo devices --json at run time (ids change if a device is re-paired).