Summary

After updating OpenClaw from 2026.5.7 to 2026.5.12, the gateway started successfully but Codex-backed turns failed with:

Requested agent harness "codex" is not registered.

This looks related to the previously closed harness activation issue #66260, but the observed failure mode here is narrower: the update/config repair path rewrote ~/.openclaw/openclaw.json into a state where the Codex plugin was disabled and the old runtime config was gone or invalid, while existing/new sessions still attempted to use the Codex harness.

Environment

• OpenClaw before update: 2026.5.7
• OpenClaw after update: 2026.5.12 (f066dd2)
• Install: global CLI under Homebrew/node path
• OS: macOS
• Gateway: LaunchAgent on 127.0.0.1:18789
• Primary model: gpt-5.5 / repaired to openai/gpt-5.5

What happened

The log shows the update sequence:

2026-05-16T00:03:57.854Z Updating OpenClaw...
2026-05-16T00:04:43.222Z Update Result: OK
Before: 2026.5.7
After: 2026.5.12

Afterward, the current config had:

{
  "plugins": {
    "entries": {
      "codex": {
        "enabled": false
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "gpt-5.5"
      }
    }
  }
}

The adjacent backup still showed the older intended Codex setup:

{
  "plugins": {
    "entries": {
      "codex": {
        "enabled": true,
        "config": {
          "computerUse": {
            "autoInstall": true,
            "enabled": false,
            "marketplaceDiscoveryTimeoutMs": 60000,
            "pluginName": "computer-use",
            "mcpServerName": "computer-use"
          },
          "codexDynamicToolsProfile": "native-first",
          "codexDynamicToolsExclude": []
        }
      }
    }
  },
  "agents": {
    "defaults": {
      "agentRuntime": {
        "id": "codex"
      },
      "model": {
        "primary": "gpt-5.5"
      }
    }
  }
}

Trying to preserve the old Codex plugin config shape failed validation on 2026.5.12:

plugins.entries.codex.config: invalid config: must NOT have additional properties

Trying to keep the old default runtime shape also failed/was ignored:

agents.defaults.agentRuntime is ignored; set models.providers.<provider>.agentRuntime or a model-scoped agentRuntime instead
agents.defaults.agentRuntime: Unrecognized key: "fallback"

Once the gateway was running in the broken state, every Codex-backed turn failed with:

error diagnostic lane task error: lane=main durationMs=36 error="Error: Requested agent harness "codex" is not registered."
error Embedded agent failed before reply: Requested agent harness "codex" is not registered.

Expected behavior

The 2026.5.12 update/config migration should preserve Codex intent in a valid current config shape. If the old config had Codex enabled and agents.defaults.agentRuntime.id = "codex", the migrated config should keep the Codex plugin enabled and move runtime policy to a valid provider/model-scoped location.

Actual behavior

The update left the gateway with:

• plugins.entries.codex.enabled = false
• no effective default/model-scoped agentRuntime.id = "codex"
• stale sessions or route state still requesting the codex harness
• repeated runtime failure: Requested agent harness "codex" is not registered

Workaround that repaired it locally

The working repair was:

{
  "plugins": {
    "entries": {
      "codex": {
        "enabled": true
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "openai/gpt-5.5"
      },
      "models": {
        "openai/gpt-5.5": {
          "agentRuntime": {
            "id": "codex"
          }
        }
      }
    }
  }
}

Then restarting the gateway fixed the registration problem:

gateway http server listening (6 plugins: browser, codex, imessage, memory-core, memory-wiki, telegram; 2.7s)
gateway ready

openclaw gateway status --deep then reported:

Gateway version: 2026.5.12
Runtime: running
Connectivity probe: ok
Capability: admin-capable

Why this may need a migration fix

This appears to be a config migration/update regression rather than just user config drift:

1. The pre-update backup had Codex enabled and selected as the default agent runtime.
2. The post-update config disabled the Codex plugin.
3. The old plugin config shape is no longer accepted by 2026.5.12.
4. The old agents.defaults.agentRuntime location is now legacy/ignored.
5. The gateway still had sessions requesting codex, but no live Codex harness owner plugin was loaded.

A migration path should probably map old Codex intent into the current model-scoped or provider-scoped agentRuntime.id = "codex" shape and leave plugins.entries.codex.enabled = true.