Homey SDK / upstream watch — zigbee-clusters 3.9.0 and tuya-local (24/10/2026)
Written in our own words from the upstream commits; nothing copied. Every point cites its source commit.
Athom zigbee-clusters 3.9.0 (released 23/10/2026)
Source: athombv/node-zigbee-clusters e6846c7…b859be7 (PRs #235–#254, tag 3.9.0 at 5ca024f).
Our apps pin zigbee-clusters ^2.6.0, so none of this reaches us automatically (major bump). These are leads for a future, tested upgrade.
- Transaction sequence numbers. Each node now keeps one counter that starts at a random value and is shared by all its endpoints (PR #235). Responses are matched to the pending request by command as well as by sequence number (PR #237), and a reply with no pending request is ignored (PR #254). Impact for us: fewer false "response timeout" cases on chatty Tuya devices that reuse sequence numbers.
- Attribute reports carry their sequence number. An
attr.<name>listener now gets a second argument withtransactionSequenceNumber(PR #240, commit 8797f22). Lead: our duplicate-report filter (dynamic per-device windows) could use the sequence number to tell a real repeat from a firmware retransmit, as an optional extra signal only. - Default Response behaviour. Only the global Default Response (0x0B, not cluster-specific) is skipped (PR #236). No Default Response is sent to broadcast frames (PR #238). Unknown commands or clusters are answered with the ZCL statuses UNSUP_COMMAND / UNSUPPORTED_CLUSTER (PR #239).
- Error text change. The internal errors
cluster_unavailableandbinding_unavailablebecome aZCLError('UNSUPPORTED_CLUSTER')(lib/Endpoint.js, PR #239). Pitfall for the upgrade: any code that matches those strings must also accept the ZCL error. Today our tree only mentionsbinding_unavailablein comments (remote_button_wireless_wall, remote_button_wireless_smart). - Alarms cluster.
getAlarmnow declares its response (status, alarmCode, clusterId, timeStamp) (commit 709c921). - Tooling. Tests moved to
node:test, development and CI pinned to Node 22, ESLint 10 flat config, Prettier (PRs #244–#247, #250). Matches our own Node 22 /node --testdirection.
tuya-local (21–24/10/2026)
Source: make-all/tuya-local d51e586…7169307.
- Long phase_a payload on a DIN breaker. The Powbay JGQW01 breaker sends DP 6 as an 18-byte block instead of the usual 8 bytes: voltage in bytes 2–3, current in bytes 4–6, active power in bytes 7–9, reactive power in bytes 10–12, apparent power in bytes 13–15, power factor (0.01) in byte 16, and sign flags in byte 17 (0x01 current, 0x02 active power, 0x04 reactive power, 0x08 power factor; set means negative; corrected 24/10 after re-reading the tuya-local template) (commit 8b12d78). Lead: our energy parser should pick the layout from the payload length and keep the 8-byte path as the default, additive only. Done in P3002b:
lib/tuya/TuyaPhaseLongV2.js, wired intoTuyaPhaseStruct.decodePhaseVariant2andSpmPhaseDecoder.decodePhaseonly for 18 bytes starting02 0F; 8-byte path unchanged (testtest/critical/p3002b-phase-long-v2.test.js). - zigbee-clusters 3.9.0 lesson (P3002c). 3.9.0 throws
ZCLError('UNSUPPORTED_CLUSTER')with.zclStatus; our pinned 2.6.0 throwsnew Error(<status name>).UnsupportedRegistry.isUnsupportedErroralready matched the 3.9.0 shape but missed the 2.6.0 namesUNSUP_CLUSTER_COMMAND,UNSUP_MANUF_CLUSTER_COMMAND,UNSUP_MANUF_GENERAL_COMMAND(0x81/0x83/0x84 in zclTypes). Added, no upgrade (testtest/critical/p3002c-zcl-unsup-status-names.test.js). - New Wi-Fi configs (not Zigbee): Lucci Aria DC ceiling fan, product id
qzr8t7c8qy78bhji, DPs 1, 2, 3, 8, 15, 16, 22, 101 (PR #6289); Mowe MW278I induction hob, product idi6qpkzm123tvw3a2, DPs 101–107, 110, 111, 120 (PR #6279). Kept as Wi-Fi leads only.
Unchanged sources this run
zigbee-herdsman-converters, Zigbee2MQTT docs, ZHA quirks, deCONZ DDF, Blakadder, tinytuya, localtuya, Johan's app and the Koenkk OTA index had no new commit since the 21/10 refresh (checked against the GitHub API on 24/10/2026). No new exact Zigbee couple entered the enrichment flow.