Problem

Long-running agent sessions accumulate too many messages, leading to:

• HTTP 400: invalid_request_error: context is too long errors
• Model API failures
• Manual intervention required to clear sessions
• No automatic cleanup mechanism

Current Behavior

• Sessions grow indefinitely
• compaction.mode: "safeguard" is too conservative
• No built-in way to reset sessions without deleting files
• Token usage is not visible to users

Impact

This affects:

• Long-running group chats
• Personal assistants with continuous conversation
• Multi-agent deployments with high message volume

Users experience unexpected failures and must manually delete session files.

Proposed Improvements

1. Enhanced Auto-Compaction

{
  "agents": {
    "defaults": {
      "compaction": {
        "mode": "auto",
        "strategy": "sliding-window",
        "maxTokens": 150000,
        "targetTokens": 100000,
        "preserveRecent": 10,
        "preserveTools": true,
        "summarizeOlder": true
      }
    }
  }
}

Features:

• Automatically compact when approaching token limit
• Preserve recent messages for context
• Keep tool call history for continuity
• Optionally summarize older conversations

2. Session Reset Command

Add built-in /reset command to clear session history:

{
  "commands": {
    "native": "auto",
    "custom": {
      "/reset": {
        "description": "Clear session history",
        "action": "reset-session",
        "confirmRequired": false
      }
    }
  }
}

User can type /reset in conversation to start fresh without deleting files manually.

3. Automatic Session Archival

• Archive sessions older than N days
• Compress old sessions to reduce disk usage (.jsonl.gz)
• Option to auto-delete very old sessions

{
  "agents": {
    "defaults": {
      "sessions": {
        "archival": {
          "enabled": true,
          "archiveAfterDays": 7,
          "deleteAfterDays": 30,
          "compress": true
        }
      }
    }
  }
}

4. Context Length Warning and Visibility

• Warn user when session is approaching token limit (e.g., at 80%)
• Show current token usage in session status
• Suggest compaction or reset proactively

Example output:

⚠️ Session is using 160K/200K tokens (80%). Consider /reset to free up context.

Use Cases

1. Group chat agent: Conversations can span days/weeks
2. Personal assistant: Continuous daily interactions
3. Customer support: Long support threads
4. Development agent: Extended debugging sessions

Benefits

• Prevents "context too long" errors automatically
• Reduces manual maintenance burden
• Improves system reliability
• Better user experience with proactive warnings

Additional Context

Current workaround requires users to:

rm -rf ~/.openclaw/agents/*/sessions/*.jsonl
echo '{}' > ~/.openclaw/agents/*/sessions/sessions.json
openclaw gateway restart

This is not user-friendly and loses all conversation history.