verbosity level too high

[gloomy willow]

Continually getting messages like:
-Advancing the checkpoint to the newest seen message so the watcher stays quiet on the next poll.
-I confirmed the latest thread entries are all outbound watcher messages, so I’m only advancing the stored cursor to the newest seen id and leaving the inbound marker untouched.
-I’ve got the stored cursor. Now I’m reading the latest 5 messages to see whether anything inbound is newer than that marker.
-The latest thread slice is only prior watcher messages, so I’m checking the persisted cursor before deciding whether anything new actually arrived.

how do I turn down the verbosity but keep Claw doing these things ?

[icy abyss]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[glacial agate]

Those lines are internal/tool progress leaking into the chat. The work can keep running; turn off the session’s visibility knobs in the same thread/session where you’re seeing it:

/verbose off
/trace off
/reasoning off

Send them as standalone messages. For future sessions, make sure your config default is off:

openclaw config set agents.defaults.verboseDefault off
openclaw config validate

If you use the Control UI, also check that the session’s verbose setting is set to inherit/off, not on/full.

[glacial agate]

Docs I checked:
https://docs.openclaw.ai/help/faq#how-do-i-stop-internal-system-messages-from-showing-in-chat
https://docs.openclaw.ai/gateway/configuration-reference

If those commands don’t quiet it down, please share openclaw --version plus whether this is Discord, TUI, or another channel.

[gloomy willow]

This is being communicated via iMessage.

[glacial agate]

Got it, iMessage uses the same session-level knobs. Send these in the iMessage conversation with Claw, as standalone messages:

/verbose off
/trace off
/reasoning off

Then set the default for future runs too:

openclaw config set agents.defaults.verboseDefault off
openclaw config validate

If you’re using BlueBubbles and you also enabled block/progress-style streaming, that can add extra visible progress chunks. Check this too:

[glacial agate]

openclaw config get channels.bluebubbles.blockStreaming

If it prints true and you don’t want interim chunks in iMessage:

openclaw config set channels.bluebubbles.blockStreaming false
openclaw config validate

Docs:
https://docs.openclaw.ai/help/faq#how-do-i-stop-internal-system-messages-from-showing-in-chat
https://docs.openclaw.ai/channels/bluebubbles
https://docs.openclaw.ai/channels/imessage

If it still happens after that, please share openclaw --version and whether your iMessage setup is BlueBubbles or legacy imsg.

[gloomy willow]

its still sending these types of messages. Do I need to reboot / restart my Claw ?

[glacial agate]

A restart is reasonable here, but first stop the noisy run/session if it is still active.

In the iMessage conversation, send this as a standalone message:

stop current run

Then send, also standalone:

/verbose off
/trace off
/reasoning off

After that, restart the background Gateway service so any config/default changes are definitely reloaded:

[glacial agate]

openclaw gateway restart

Then verify:

openclaw --version
openclaw config get agents.defaults.verboseDefault
openclaw config validate

The Gateway does support config hot-reload, but restarting is a clean next step when a channel keeps emitting old behavior. If it still posts those watcher/checkpoint lines after restart, please paste the openclaw --version output and whether this is BlueBubbles or legacy imsg.

Docs:
https://docs.openclaw.ai/help/faq#how-do-i-stop-internal-system-messages-from-showing-in-chat
https://docs.openclaw.ai/help/faq#do-i-have-to-restart-after-changing-config

[glacial agate]

https://docs.openclaw.ai/cli/gateway

[gloomy willow]

Still seeing this "The thread query came back empty, so I’m checking whether the watcher state already has a watermark before deciding there’s anything new to handle."

how do I quite it down. ?

[glacial agate]

That exact sentence does not appear in the OpenClaw source/docs I checked, so this is probably not a built-in verbosity toggle anymore. It looks like a watcher/recurring agent task is still choosing to send its own narration.

I’d debug it as a running task/cron now:

openclaw tasks list --status running
openclaw cron list

If you see a watcher-like task:

openclaw tasks cancel <taskId-or-runId>

If you see a watcher-like cron job, inspect it:

[glacial agate]

openclaw cron show <jobId>
openclaw cron runs --id <jobId> --limit 10

Then either stop delivery:

openclaw cron edit <jobId> --no-deliver

or remove the job if you don’t need it:

openclaw cron remove <jobId>

[glacial agate]

Important nuance: --no-deliver disables the cron runner’s final fallback delivery, but if the watcher prompt tells the agent to use the message/iMessage tool, it can still send messages. In that case the real fix is to edit the watcher prompt so no-op polls return exactly NO_REPLY and only send when there is something actionable.

To locate the session that’s generating it, you can search locally:

rg "thread query came back empty|watcher state already has a watermark" ~/.openclaw/agents/*/sessions

No need to paste private transcript contents here. Just paste openclaw --version, whether tasks list / cron list shows a watcher, and the job/task id shape if you want help deciding which one to change.

[glacial agate]

To quiet it down while keeping the watcher running, change the watcher job so no-op polls produce no outbound message.

Do this first:

openclaw cron list
openclaw tasks list --status running

Find the watcher job. Then use the least disruptive option:

openclaw cron edit <jobId> --no-deliver

If it still sends iMessages after that, the job prompt itself is telling the agent to send messages. Edit the cron prompt so it says something like:

[glacial agate]

If there is nothing new/actionable, reply exactly NO_REPLY.
Do not narrate polling, cursor, watermark, checkpoint, or watcher-state steps.
Only send a message when there is a real new inbound item that needs attention.

If you just want silence immediately:

openclaw tasks cancel <taskId-or-runId>

or remove the recurring watcher:

openclaw cron remove <jobId>

[glacial agate]

The key thing: /verbose off hides OpenClaw’s tool/status output, but it won’t stop a custom watcher prompt from sending its own “I checked the cursor…” prose. That prompt needs NO_REPLY for quiet no-op runs.

[gloomy willow]

Still getting things like this "The acknowledgement is out. I’m updating the watcher state with the inbound message id and timestamp that triggered it."

[glacial agate]

That one confirms it: the watcher is sending its own narration after it sends an acknowledgement. This is not something /verbose off can suppress.

Fastest way to quiet it immediately:

openclaw tasks list --status running
openclaw tasks cancel <taskId-or-runId>

Then find the recurring job and either pause/remove it:

openclaw cron list
openclaw cron show <jobId>
openclaw cron remove <jobId>

[gloomy willow]

how do I quiet it down ?

[glacial agate]

If you want to keep the watcher, edit its prompt. It needs a hard rule like:

Never send progress/status narration.
Never mention cursor, watermark, checkpoint, watcher state, ids, timestamps, or polling.
If an acknowledgement was already sent, finish silently with exactly NO_REPLY.
If there is nothing actionable, reply exactly NO_REPLY.
Only send user-facing messages that should actually appear in iMessage.

Also check whether the job prompt explicitly says to use iMessage/message after every step. If it does, --no-deliver will not help, because the agent is directly sending via the messaging tool. The fix is the prompt rule above, or removing/canceling that watcher.

[glacial agate]

Do this in order:

1. Quiet it immediately:

openclaw tasks list --status running
openclaw tasks cancel <taskId-or-runId>

2. Find the recurring watcher:

openclaw cron list
openclaw cron show <jobId>

3. Either remove it:

[glacial agate]

openclaw cron remove <jobId>

or keep it but replace/edit the watcher prompt so no-op/status work is silent:

openclaw cron edit <jobId> --message "Watch for new inbound messages. Only send a user-facing acknowledgement when there is a real new inbound item that requires one. Never narrate polling, cursor, watermark, checkpoint, watcher state, message ids, timestamps, or tool steps. If there is nothing actionable, or after any acknowledgement has already been sent, reply exactly NO_REPLY."

Then restart once:

openclaw gateway restart

The important bit is NO_REPLY. Your watcher is currently treating its internal state update as something to say out loud.

[gloomy willow]

how do I get more inforamtion about a cron other than openclaw cron show <jobID> ?

[glacial agate]

Useful ways to get more than cron show:

openclaw cron list --json
openclaw cron show <jobId> --json
openclaw cron runs --id <jobId> --limit 50

For a specific run from that history, take the runId / session key and inspect the task record:

openclaw tasks show <runId-or-sessionKey>
openclaw tasks show <runId-or-sessionKey> --json

For live failures/noisy behavior, logs are usually the next best source:

[glacial agate]

openclaw logs --follow

And for finding the exact transcript that contains the watcher prose:

rg "watcher state|watermark|checkpoint|acknowledgement is out" ~/.openclaw/agents/*/sessions

Don’t paste private transcript contents here; just use that to identify the session/job locally.

Also please include openclaw --version when sharing outputs, because cron/task behavior has changed across versions.

Docs:
https://docs.openclaw.ai/cli/cron
https://docs.openclaw.ai/cli/tasks

[glacial agate]

https://docs.openclaw.ai/automation/cron-jobs

[gloomy willow]

Can I alter a cron to be quiet ?

[glacial agate]

Yes.

For the cron runner’s normal final delivery, make it quiet with:

openclaw cron edit <jobId> --no-deliver

But in your case, the noisy text looks like the cron’s agent prompt is directly sending iMessages. --no-deliver may not stop that. You also need to edit the cron message/prompt so no-op work returns NO_REPLY:

openclaw cron edit <jobId> --message "Watch for new inbound messages. Only send a user-facing message when there is a real new inbound item that requires one. Never narrate polling, cursor, watermark, checkpoint, watcher state, message ids, timestamps, or tool steps. If there is nothing actionable, or after any acknowledgement has already been sent, reply exactly NO_REPLY."

Then restart once:

[glacial agate]

openclaw gateway restart

Quick check after editing:

openclaw cron show <jobId>
openclaw cron runs --id <jobId> --limit 10

So: --no-deliver quiets the cron fallback delivery; NO_REPLY in the prompt quiets the agent itself.

[gloomy willow]

I think we have it!

Well done!

[glacial agate]

Nice, glad we narrowed it down.

If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution. That helps Answer Overflow index the fix for the next person who gets a chatty watcher.

[glacial agate]

Happy to help! 🦞

[gloomy willow]

How do I mark it as solved using discord app ?

[glacial agate]

In the Discord app:

1. Long-press the message that solved it.
2. Tap Apps.
3. Tap ✅ Mark Solution.

On desktop it’s similar: right-click the solution message → Apps → ✅ Mark Solution.

If you don’t see Apps or Mark Solution, try the desktop app/browser; Discord mobile sometimes hides app actions depending on the client/version.