openclaw - 💡(How to fix) Fix [ClawX] Session freezes after compaction — Thinking... stuck, stop button not working [1 participants]

Bug Description

After ClawX 3.10 + OpenClaw 2026.4.21 upgrade, sessions freeze in 'Thinking...' state after compaction triggers. The bot reacts with 👀 emoji but the LLM call never completes. Stop button does not work.

This appears to be the same root cause as #60867 (Discord) but in ClawX webchat.

Steps to Reproduce

1. Have an active ClawX session (webchat, not Discord)
2. Session grows over 1-4 hours of work
3. Compaction triggers automatically (context window overflow)
4. Bot acknowledges with 👀 reaction
5. Session enters zombie state — no LLM response ever comes
6. Stop button is unresponsive — user cannot interrupt
7. Closing and reopening ClawX DOES NOT fix — session reconnects as zombie
8. Only solution: start completely new session (losing all context)

Environment

• ClawX: 3.10 (upgraded ~3 days ago)
• OpenClaw: 2026.4.21 (upgraded same day as ClawX 3.10)
• Model: MiniMax-M2.7 (200K context)
• Channel: webchat / ClawX desktop app
• OS: Windows 10

Compaction Settings

{
  "compaction": {
    "maxHistoryShare": 0.4,
    "keepRecentTokens": 16000,
    "reserveTokens": 6000,
    "recentTurnsPreserve": 5
  }
}

Timing Observations

• April 18-20: Sessions up to 20MB+ worked fine
• April 21: First stuck incident — small session (37K tokens) froze after compaction
• April 22: More stuck incidents after OpenClaw 2026.4.21 regression

Key Finding

Stuck state is NOT caused by session size. Small sessions (37K tokens) freeze too. The issue is compaction + 2026.4.21 regression making the LLM call never complete after compaction.

Additional Observations

• Tool calls execute successfully (file writes, git commits, etc.) — model CAN use tools
• After tool call completes, the next LLM response never comes
• Session appears "alive" to ClawX but is actually in zombie state
• Restarting ClawX does NOT fix — session reconnects as zombie
• This happened 3 times in the past 3 days

Workaround Attempted

Changed recentTurnsPreserve from 11 to 5 (more frequent but smaller compactions). Still experiencing freezes.

Related Issues

• #60867 — Discord session freeze after compaction (same root cause)
• #63279 — Compaction timeout cascades
• #67750 — Successful compaction still times out
• #69994 — BOOTSTRAP.md overwrites SOUL.md after 2026.4.21
• #70022 — Telegram slash commands break after 2026.4.21

What Would Help

A compaction.timeoutAction setting as suggested in #44031 would allow users to choose behavior when LLM doesn't respond after compaction. Or a "force restart session" option that actually kills the stuck session.

I'm also affected by this issue. Please prioritize a fix.