Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

After upgrading a live npm install from OpenClaw 2026.6.1 to 2026.6.5-beta.6, Telegram /status no longer shows the OpenAI/Codex quota usage line even though the active model is still openai/gpt-5.5, runtime is still OpenAI Codex, and model auth/routing remains usable.

Steps to reproduce

Observed on a live VPS install:

1. Run OpenClaw 2026.6.1 with Telegram enabled, openai/gpt-5.5 as the default model, and OpenAI Codex runtime/auth.
2. Send /status in a Telegram topic.
3. Confirm status includes the Usage line for 5h/week OpenAI usage windows.
4. Upgrade to 2026.6.5-beta.6 via npm global install and restart the gateway.
5. Send /status in the same Telegram topic.
6. Observe that the Usage line is gone.
7. Run openclaw status --usage and openclaw models auth list from the VPS shell.

Expected behavior

/status should continue to show the OpenAI/Codex usage window when the active model is OpenAI Codex and the usage/auth data is available, as it did on 2026.6.1.

Known-good Telegram status on 2026.6.1 included:

🦞 OpenClaw 2026.6.1 (0e30120)
🧠 Model: openai/gpt-5.5 · 🔑 oauth (...redacted...)
🔄 Fallbacks: google/gemini-3.1-pro-preview
📊 Usage: 5h 98% left ⏱1h 17m · Week 58% left ⏱3d 5h
⚙️ Execution: direct · Runtime: OpenAI Codex · Think: off · Fast: off · Text: low

Actual behavior

After upgrade, Telegram /status still shows the same model/runtime family but no usage line:

🦞 OpenClaw 2026.6.5-beta.6 (1514dc7)
🧠 Model: openai/gpt-5.5
🔄 Fallbacks: google/gemini-3.1-pro-preview
🧮 Tokens: 852 in / 283 out
🗄️ Cache: 99% hit · 68k cached, 0 new
📚 Context: 69k/272k (25%) · 🧹 Compactions: 0
🧵 Session: agent:main:telegram:group:-1003871627242:topic:23 · updated just now
⚙️ Execution: direct · Runtime: OpenAI Codex · Think: off · Fast: off · Text: low

The missing line is not Telegram-only. Shell status also does not find OpenAI/Codex usage:

$ openclaw status --usage
OpenClaw 2026.6.5-beta.6 (1514dc7)
...
Usage
Usage:
  Claude: HTTP 403: OAuth token does not meet scope requirement user:profile

There is no OpenAI/Codex usage entry in that output.

OpenClaw version

Bad: 2026.6.5-beta.6 (1514dc7)

Last observed good: 2026.6.1 (0e30120)

Operating system

Ubuntu/Linux VPS, arm64 kernel 6.17.0-1011-oracle, Node 22.22.0.

Install method

npm global system install:

CLI version: 2026.6.5-beta.6 (/usr/bin/openclaw)
Gateway version: 2026.6.5-beta.6
Command: /usr/bin/node /usr/lib/node_modules/openclaw/dist/index.js gateway --port 18789
Gateway service: systemd user service

Model

openai/gpt-5.5

Provider / routing chain

OpenClaw Telegram topic -> gateway -> OpenAI Codex runtime / Codex app-server auth.

Additional provider/model setup details

Relevant redacted config shape:

{
  "agents.defaults.model.primary": "openai/gpt-5.5",
  "agents.defaults.model.fallbacks": ["google/gemini-3.1-pro-preview"],
  "agents.defaults.models": {
    "codex/gpt-5.5": { "alias": "codex", "agentRuntime": { "id": "codex" } },
    "openai/gpt-5.5": { "alias": "codex", "agentRuntime": { "id": "codex" } }
  },
  "plugins.entries.codex.enabled": true,
  "plugins.entries.tokenjuice.enabled": true,
  "plugins.entries.active-memory.enabled": false
}

openclaw models status --json --probe --probe-provider openai --probe-timeout 20000 --probe-max-tokens 4 reports runtime auth as usable but synthetic/plugin-owned:

{
  "runtimeAuthRoutes": [
    {
      "provider": "openai",
      "runtime": "codex",
      "authProvider": "openai",
      "status": "usable",
      "effective": {
        "kind": "synthetic",
        "detail": "codex-app-server"
      }
    }
  ],
  "providers": [
    {
      "provider": "openai",
      "effective": {
        "kind": "synthetic",
        "detail": "codex-app-server"
      },
      "profiles": {
        "count": 0,
        "oauth": 0,
        "token": 0,
        "apiKey": 0,
        "labels": []
      },
      "syntheticAuth": {
        "value": "plugin-owned",
        "source": "codex-app-server",
        "credential": "codex-app-server",
        "mode": "token"
      }
    }
  ]
}

openclaw models auth list shows the agent-local profile store has no profiles:

Agent: main
Auth state store: ~/.openclaw/agents/main/agent/openclaw-agent.sqlite
Profiles: (none)

That may be the key distinction: model execution can still use the synthetic Codex app-server auth route, but provider usage/status discovery appears to require an agent-local OAuth/token profile and therefore suppresses the OpenAI usage line.

Logs, screenshots, and evidence

Screenshots were captured for both states. They show:

• 2026.6.1: /status includes 📊 Usage: 5h 98% left ... Week 58% left ...
• 2026.6.5-beta.6: /status has the same model/runtime family but no Usage line.

Gateway/service sanity after upgrade:

$ openclaw gateway status --deep
CLI version: 2026.6.5-beta.6 (/usr/bin/openclaw)
Gateway version: 2026.6.5-beta.6
Runtime: running (pid 2332098, state active, sub running, last exit 0, reason 0)
Connectivity probe: ok

Status usage after upgrade:

$ openclaw status --usage
Usage:
  Claude: HTTP 403: OAuth token does not meet scope requirement user:profile

No exact duplicate found from targeted issue searches for:

• status usage OpenAI Codex oauth missing Usage line
• provider usage codex app-server synthetic auth usage missing
• providersWithOAuth codex-app-server usage

Related but not exact issues found during search include #58498, #78714, #74085, and #86215.

Impact and severity

Affected: OpenAI Codex users relying on /status or openclaw status --usage to monitor 5h/week usage windows.

Severity: Medium. The model still runs, but the operator loses quota/usage visibility. This has been a recurring status regression surface across several recent releases, so a regression test around Codex/OpenAI usage visibility would be useful.

Frequency: Always observed after upgrading this VPS to 2026.6.5-beta.6.

Consequence: Users cannot tell from /status how much OpenAI/Codex usage remains, even though the session is actively using OpenAI Codex.

Additional information

Potentially relevant recent change area: the version diff between 2026.6.1 and 2026.6.5-beta.6 includes auth-profile/session-state work. One likely suspect is commit cee432f0f0 (fix(auth): prefer agent-local auth profiles, fixes #64274), because the failing host has no agent-local auth profiles while runtime auth is reported as synthetic codex-app-server.

Release discussion also mentioned that the SQLite migration for sessions was rolled back for now. I do not have proof that this is causal, but it may be relevant because this failure looks like status/usage discovery depending on a SQLite-backed or agent-local auth/session view that does not contain the Codex app-server OAuth usage source.

The expected behavior is not that the runtime must use agent-local auth profiles; it is that provider usage/status should still discover the active Codex/OpenAI usage source when the effective runtime auth route is usable.