openclaw - 💡(How to fix) Fix [Bug] Null agent workspace field causes crash loop — auto-restore does not repair missing required fields [1 comments, 2 participants]

Bug Summary

If any agent in agents.list has a missing/null workspace field, the gateway crashes with Invalid input: expected string, received null and enters a crash loop. The openclaw doctor --fix auto-restore mechanism does not repair this field, so subsequent restarts also fail until the user manually edits the config.

Environment

• OpenClaw: 2026.5.3-1
• Node.js: 22.22.2 (Linux)
• Config: ~/.openclaw/openclaw.json

Steps to Reproduce

1. Create or modify an agent entry in agents.list without a workspace field
2. Start or restart the gateway
3. Observe crash with startup bundle reason gateway.startup_failed
4. Run openclaw doctor or openclaw gateway restart
5. Gateway crashes again — doctor does not repair the missing field

Observed Error Log

Invalid config at /home/moltbot/.openclaw/openclaw.json.
agents.list.2.workspace: Invalid input: expected string, received null
Run `openclaw doctor` to repair, or fix the keys above manually.

Error: systemctl restart failed: Job for openclaw-gateway.service failed because the control process exited with error code.

Observed Behavior

After the crash, stability bundles are written every 6 seconds:

wrote stability bundle: ...openclaw-stability-...-gateway.startup_failed.json

This creates a tight crash loop until the bad config entry is manually fixed.

Root Cause

The agents.list[N].workspace field is a required string in the config schema. The gateway validates this on startup and refuses to start if it's null. However:

1. The openclaw doctor tool does not repair missing required string fields (only repairs certain known patterns)
2. The last-known-good auto-restore mechanism (doctor-invalid-config) restores the same broken config because it was the last valid config before the field went null
3. No sentinel or schema validation repair is applied during the restore

Impact

• Gateway downtime ~7 minutes (observed in production)
• Requires manual config editing to recover
• Crash loop generates ~60 stability bundles per minute

Suggested Fix

One of:

1. Schema default: Treat missing workspace as inheriting the global default (agents.defaults.workspace)
2. Doctor repair: Make openclaw doctor --fix detect and repair null/missing required string fields in agent entries
3. Startup guard: If an agent entry is missing required fields, warn instead of crashing; start the gateway with the other valid agents

Config Entry That Causes This

{
  "id": "small-model-sandbox",
  "model": "ollama/qwen2.5-coder:7b",
  "sandbox": { "mode": "all" },
  "tools": { "deny": ["group:web", "browser"] }
  // ← no "workspace" field
}

Fix: add "workspace": "/path/to/workspace" to resolve.