…un 2)

…un 4)

…un 5)

…un 6)

…un 7)

…un 8)

…un 9)

…un 10)

… access)

… access)

…un 13)

…un 14)

…un 15)

…un 16)

…un 17)

…un 18)

…un 19)

…un 20)

…un 21)

…un 22)

…un 23)

…un 24)

…un 25)

…un 26)

https://claude.ai/code/session_0192kFJRhf69MzmfSb5p9o4B

https://claude.ai/code/session_011BUf4sxSmhjTJV1LrzF5hv

https://claude.ai/code/session_014Zbf87BbccCFYJHmYAZgQw

https://claude.ai/code/session_014A46VG6TVew4o3ZWmiJc2P

Before this fix, if onTimer rejected unexpectedly (e.g. a Node.js
internal error or GC pressure causing an exception in the finally
block's armTimer call), the .catch() handler only logged the error.
The scheduler chain was then permanently broken with no timer set,
silently halting all cron jobs until the next gateway restart.

Fix: call armTimer(state) inside the .catch() handler so a rare
unexpected rejection does not permanently stop the scheduler.

Regression test exercises the path by making nowMs() throw on the
4th call (inside the finally block's armTimer), which causes onTimer
to reject; the .catch() re-arm is then verified via state.timer.

Closes openclaw#73166.

https://claude.ai/code/session_01NHHoPHTrH4F9qFJBJHqjTk

web_fetch decoded all HTTP response bodies as UTF-8 unconditionally.
readResponseText() used `new TextDecoder()` (UTF-8) in the streaming
path and `res.text()` (also UTF-8 per WHATWG Fetch spec) in the
fallback path, causing mojibake for legacy-charset pages such as
Shift_JIS, Big5, and ISO-8859-1.

Fix: in the streaming path, collect raw bytes before decoding and
resolve the charset from the Content-Type `charset=` parameter; if
absent and the response is HTML, scan the first 4 KB for a
`<meta charset>` or http-equiv declaration, then decode with
`TextDecoder(detectedCharset)`. The non-streaming fallback uses
`arrayBuffer()` + the same charset resolution for environments that
expose it, and retains the old `text()` path as a last resort.

Closes openclaw#72916.

https://claude.ai/code/session_01NHHoPHTrH4F9qFJBJHqjTk