#Since I tried upgrading to 2026.4.24, agents strange.

1 messages · Page 1 of 1 (latest)

serene jetty
#

I upgraded to 2026.4.24, but that broke my openclaw setup, so I had to revert back to version 2026.4.23... But my agents acting weird. they appear to stop mid work. they try something they cannot get it to work and just restart the gateway... they have restarted the gateway 16 times in the last hour and sometimes it does not even start properly and I need to start it manually.

ivory moatBOT
#

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

toxic dragon
#

go back to 2026.4.23

#

2026.4.24 is broken

#

thats what i did

#

npm install -g openclaw@2026.4.23

serene jetty
torn terraceBOT
#

Yeah, reverting package version may not undo state/config touched by 2026.4.24. First I’d stop the restart loop, then gather diagnostics.

Quick tourniquet: disable agent-triggered gateway restarts:

openclaw config set commands.restart false --dry-run
openclaw config set commands.restart false
openclaw config validate

commands.restart=false blocks /restart and gateway restart tool actions. Docs: https://docs.openclaw.ai/tools/slash-commands

Then please paste this output:

openclaw --version
# 
which openclaw
openclaw config get meta.lastTouchedVersion
openclaw gateway status --deep
openclaw status --all
openclaw doctor

The important thing I’m checking: if meta.lastTouchedVersion says 2026.4.24 while the binary is 2026.4.23, that’s downgrade/split-brain territory and can make service mutations weird.

Also check whether agents are repeatedly “bootstrapping” or recovering after restart. If logs mention BOOTSTRAP.md, stale session recovery, or plugin runtime deps, include that log snippet too:

openclaw logs --tail 200

This does sound like real 2026.4.24 fallout, not normal behavior. If the status/logs show dependency/plugin-runtime errors or downgrade corruption, it’s worth filing a GitHub issue with the exact repro: https://github.com/openclaw/openclaw/issues

serene jetty
#

I have reverted everything I can think of myself, so now it seems to work ok..

torn terraceBOT
#

Glad it’s stable again 🎉

If it comes back, try to grab the evidence before reverting more:

openclaw --version
openclaw gateway status --deep
openclaw status --all
openclaw logs --tail 200

Also, if you disabled restarts with commands.restart=false while stabilizing, remember that blocks /restart and gateway restart tool actions until you turn it back on.

If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution so Answer Overflow can index it for future users. 🦐

serene jetty
#

with commands.restart=false agents can still manually restart gateway right? it just does not auto restart when a config change is detected

torn terraceBOT
#

No — commands.restart=false blocks manual restart actions from chat/agent surfaces:

  • /restart disabled
  • gateway tool action: "restart" disabled
  • attempts should error with Gateway restart is disabled

It’s not the config-change auto-restart switch. That’s controlled separately by:

openclaw config get gateway.reload.mode

Default is usually hybrid: hot-reload safe changes, restart for critical ones.

So if your goal is “agents may not restart the gateway,” commands.restart=false is right. If your goal is “agents may not touch gateway config/update at all,” also deny the gateway tool for that agent/surface.

serene jetty
#

I do want my agents to be able to restart gateway, I ask them to perform changes somethimes, but felt like the gatewat way was more restarting then running.

#

ps what llm model are you running on krill?

torn terraceBOT
#

Yeah, that makes sense. In that case I’d leave:

openclaw config set commands.restart true

…but add guardrails so restarts are intentional, not a panic button.

Two good options:

  1. Instruction-level guardrail in the agent’s AGENTS.md:
Gateway restart policy:
- Do not restart the gateway as a troubleshooting guess.
- Restart only after an explicit user request, or after a config change that requires it.
# 
- Max one gateway restart per task. If it still fails, stop and report logs/status instead of restarting again.
  1. Block subagents from restarting, while your main agent can still do it:
{
  tools: {
    subagents: {
      tools: {
        deny: ["gateway"]
      }
    }
  }
}