openclaw - 💡(How to fix) Fix Discord group session freezes after compaction — reacts with 👀 but never responds [1 participants]

Bug Description

Discord group chat sessions become unresponsive after context compaction when the session reaches ~100K-130K tokens. The bot acknowledges incoming messages (adds 👀 reaction) but never generates a response. The session appears alive (not timed out, not errored) but is permanently stuck.

Steps to Reproduce

1. Have an active Discord group session with heavy tool usage (file reads, edits, web searches, exec commands)
2. Session grows to ~100-130K tokens over the course of a work session
3. Context compaction triggers
4. User sends a new message after compaction
5. Bot reacts with 👀 (message acknowledged) but never responds
6. All subsequent messages get 👀 but no response
7. sessions_send from main session either times out or returns empty messages: []

Observed Behavior

• Session shows totalTokens: 115000-130000 and abortedLastRun: false
• sessions_list returns the session with messages: [] (empty last message)
• The session is not dead — it's mapped correctly, reacts to messages, but the LLM call never completes or never starts
• No error message posted to the channel (unlike timeout/overload errors which do get posted)
• Creating a new channel (e.g. -2, -3 suffix) and starting fresh is the only workaround

Frequency

Extremely high. In our case, this happens on nearly every work session that exceeds ~2 hours of active use. Over 5 days (Mar 29 - Apr 3), we experienced this freeze in:

• #ck-290326-2
• #ck-300326 (twice, moved to -2 then -3)
• #ck-300326-2
• #ck-010426
• #ck-010426-2 (new channel created Apr 3)
• Two additional channels on Apr 3

The pattern is consistent: 1-2 hours of active work → compaction → freeze.

Workaround

The only reliable workaround is to read the channel messages from the main session using message action=read and respond using message action=send. This works but loses all session context and tool capabilities for that channel.

Environment

• OpenClaw version: 2026.2.12
• Model: anthropic/claude-opus-4-6
• Channel: Discord (group sessions in private guild)
• OS: Linux 6.8.0-100-generic (x64), 16GB RAM
• Node: v22.22.0

Possibly Related Issues

• #18194 — Session lost after compaction timeout
• #11224 — Discord channel sessions bypass compaction
• #36605 — Pre-compaction channel notification

Suggestion

1. Add a timeout/watchdog for post-compaction message handling — if the session hasn't responded within N seconds after compaction, log a diagnostic and attempt recovery
2. Surface the frozen state to the user (e.g. post "Session needs restart — type /reset or create a new channel")
3. Add compaction diagnostics to sessions_list output (last compaction time, compaction status, post-compaction health)