Reproduction

On a fresh OpenClaw 2026.5.4 install where the active tool policy does not enable the message tool (e.g. tools.profile = full with no explicit message tool wired), run:

openclaw doctor --fix --yes
openclaw doctor

The first run applies a recommended config change; the second run warns about that same change.

Observed

After --fix, the next doctor pass emits:

messages.groupChat.visibleReplies is set to "message_tool", but the message tool is unavailable for default tool policy; OpenClaw falls back to automatic visible replies, so normal replies may post to the source chat. Enable the message tool or set messages.groupChat.visibleReplies to "automatic".

The value it warns about was set by the previous --fix run. Every subsequent doctor invocation continues to emit the warning until the operator manually patches messages.groupChat.visibleReplies back to automatic.

Expected

doctor --fix should not apply a recommendation that the next doctor run will immediately flag as a warning. One of two things should happen instead:

1. Skip the visibleReplies = message_tool recommendation when the message tool is not present in the active tool policy.
2. Recommend automatic in that case (with a note that message_tool becomes the recommended setting if/when the message tool is enabled).

Suggested fix

Gate the visibleReplies = message_tool recommendation on whether the message tool is actually present in the resolved tool policy at fix time:

if active_tool_policy_has_message_tool:
    recommend(visibleReplies = "message_tool")
else:
    ensure(visibleReplies in ["automatic", undefined])

This affects installs that use a default tool profile without explicitly enabling the message tool, which is common for single-channel deployments.

Workaround

Re-patch after every --fix:

openclaw config set messages.groupChat.visibleReplies automatic

Environment

• OpenClaw 2026.5.4
• Ubuntu 26.04, kernel 7.0.0-15
• Single Telegram channel
• tools.profile = full, tools.exec.security = full
• Local Ollama provider (gpt-oss:120b)

Related

#73471 frames a more general concern about doctor --fix applying recommendations automatically. This is another concrete instance of that shape.