21:08:02 [plugins] plugins.allow is empty; discovered non-bundled plugins may auto-load: codex (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\codex\dist\index.js). Set plugins.allow to explicit trusted ids.
21:08:42 [agents/harness] Codex agent harness failed; not falling back to embedded PI backend
21:08:42 [diagnostic] lane task error: lane=main durationMs=2176 error="Error: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary."
21:08:42 [diagnostic] lane task error: lane=session:agent:main:dashboard:6d2e3508-cd02-476d-91b7-6da30e420b21 durationMs=2180 error="Error: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary."
21:08:42 Embedded agent failed before reply: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary.
#help version 2026.5.7 update to 2026.5.12 codex error
real patio
low cryptBOT
flat foxBOT
21:08:02 [plugins] plugins.allow is empty; discovered non-bundled plugins may au...
That error is the Codex harness failing before the agent can reply. The important bit is:
Managed Codex app-server binary was not found for @openai/codex
So this is probably not an auth/model problem. OpenClaw is selecting the Codex app-server harness, but the Codex app-server executable/package it expects is missing or stale after the update. The plugins.allow is empty line is a hardening warning, not the immediate crash.
Can you run these from the same Windows shell/user that starts OpenClaw and paste the output?
openclaw --version
openclaw status --all
openclaw plugins list --verbose
codex --version
codex app-server --help
Likely fixes to try after that:
flat foxBOT
21:08:02 [plugins] plugins.allow is empty; discovered non-bundled plugins may au...
openclaw plugins update codex
openclaw plugins doctor
If you run OpenClaw from a source checkout, also run this in the checkout after updating:
pnpm install
pnpm build
pnpm ui:build
If codex --version or codex app-server --help fails, reinstall/update the Codex CLI too. The Codex harness docs say OpenClaw expects Codex app-server 0.118.0 or newer and by default launches codex app-server --listen stdio://.
Docs: https://docs.openclaw.ai/plugins/codex-harness and https://docs.openclaw.ai/help/faq
real patio
21:25:06 [plugins] plugins.allow is empty; discovered non-bundled plugins may auto-load: codex (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\codex\dist\index.js). Set plugins.allow to explicit trusted ids.
this is what i get
flat foxBOT
21:25:06 [plugins] plugins.allow is empty; discovered non-bundled plugins may au...
That line by itself is only a warning: OpenClaw found a non-bundled codex plugin under:
C:\Users\ibrah\.openclaw pm ode_modules\@openclaw\codex\dist\index.js
It is telling you to make plugin loading explicit. It does not explain the app-server crash by itself.
Please paste the output of these exact commands too, especially the Codex ones:
openclaw --version
openclaw plugins list --verbose
codex --version
codex app-server --help
If codex app-server --help fails, that’s the root problem: install/update Codex CLI for that same Windows user, then restart OpenClaw.
Also, because your plugin is installed in .openclaw pm, try updating the plugin package:
openclaw plugins update codex
openclaw plugins doctor
After it works, you can quiet the warning by explicitly allowing the trusted plugin in config:
{
plugins: {
allow: ["codex"],
entries: {
codex: {
enabled: true
}
}
}
}
PS C:\Users\ibrah> codex app-server --help
[experimental] Run the app server or related tooling
Usage: codex app-server [OPTIONS] [COMMAND]
Commands:
proxy Proxy stdio bytes to the running app-server control socket
generate-ts [experimental] Generate TypeScript bindings for the app server protocol
generate-json-schema [experimental] Generate JSON Schema for the app server protocol
help Print this message or the help of the given subcommand(s)
Options:
-c, --config <key=value>
Override a configuration value that would otherwise be loaded from ~/.codex/config.toml. Use a dotted path
(foo.bar.baz) to override nested values. The value portion is parsed as TOML. If it fails to parse as
TOML, the raw string is used as a literal.
Examples: - `-c model="o3"` - `-c 'sandbox_permissions=["disk-full-read-access"]'` - `-c
shell_environment_policy.inherit=all`
--enable <FEATURE>
Enable a feature (repeatable). Equivalent to `-c features.<name>=true`
--disable <FEATURE>
Disable a feature (repeatable). Equivalent to `-c features.<name>=false`
--listen <URL>
Transport endpoint URL. Supported values: `stdio://` (default), `unix://`, `unix://PATH`, `ws://IP:PORT`,
`off`
[default: stdio://]
--analytics-default-enabled
Controls whether analytics are enabled by default.
Analytics are disabled by default for app-server. Users have to explicitly opt in via the `analytics` section
in the config.toml file.
However, for first-party use cases like the VSCode IDE extension, we default analytics to be enabled by
default by setting this flag. Users can still opt out by setting this in their config.toml:
```toml [analytics] enabled = false ```
See https://developers.openai.com/codex/config-advanced/#metrics for more details.
--ws-auth <MODE>
Websocket auth mode for non-loopback listeners
[possible values: capability-token, signed-bearer-token]
--ws-token-file <PATH>
Absolute path to the capability-token file
--ws-token-sha256 <HEX>
Hex-encoded SHA-256 digest of the capability token
--ws-shared-secret-file <PATH>
Absolute path to the shared secret file for signed JWT bearer tokens
--ws-issuer <ISSUER>
Expected issuer for signed JWT bearer tokens
--ws-audience <AUDIENCE>
Expected audience for signed JWT bearer tokens
--ws-max-clock-skew-seconds <SECONDS>
Maximum clock skew when validating signed JWT bearer tokens
-h, --help
Print help (see a summary with '-h')
flat foxBOT
PS C:\Users\ibrah> codex app-server --help
[experimental] Run the app server or ...
Good, that narrows it down: your Codex CLI does have app-server. So the next likely issue is that the OpenClaw gateway process cannot find the same codex binary that your PowerShell can see, or the installed Codex plugin is still stale.
Please run:
codex --version
where.exe codex
openclaw --version
openclaw plugins list --verbose
If where.exe codex prints a path, add that exact path to the Codex plugin config so OpenClaw does not depend on PATH lookup. Example shape:
{
plugins: {
allow: ["codex"],
entries: {
codex: {
enabled: true,
config: {
appServer: {
command: "C:\\Users\\ibrah\\AppData\\Roaming\
pm\\codex.cmd"
}
}
}
}
}
}
Use your actual where.exe codex path, not necessarily the example above, then restart the gateway/OpenClaw app.
The same workaround can also be done with an env var:
$env:OPENCLAW_CODEX_APP_SERVER_BIN="C:\path\to\codex.cmd"
openclaw gateway
If you run OpenClaw as a background/service/app process, config is usually better than a temporary PowerShell env var because services often do not inherit your shell PATH.
real patio
PS C:\Users\ibrah> codex --version
codex-cli 0.130.0
PS C:\Users\ibrah> where.exe codex
C:\Users\ibrah\AppData\Roaming\npm\codex
C:\Users\ibrah\AppData\Roaming\npm\codex.cmd
PS C:\Users\ibrah> openclaw --version
OpenClaw 2026.5.12 (f066dd2)
flat foxBOT
PS C:\Users\ibrah> codex --version
codex-cli 0.130.0
PS C:\Users\ibrah> where.ex...
Those versions look fine: OpenClaw 2026.5.12, Codex CLI 0.130.0, and app-server exists.
Next I’d do the explicit binary path workaround, because Windows services/app launches often don’t see the same PATH as PowerShell. In C:\Users\ibrah\.openclaw\openclaw.json, set the Codex app-server command to the .cmd path:
{
plugins: {
allow: ["codex"],
entries: {
codex: {
enabled: true,
config: {
appServer: {
command: "C:\\Users\\ibrah\\AppData\\Roaming\
pm\\codex.cmd"
}
}
}
}
}
}
Then fully restart OpenClaw / the gateway.
Also run this once, because your log shows the Codex plugin is installed separately under .openclaw pm, so it may not have updated with OpenClaw itself:
openclaw plugins update codex
openclaw plugins doctor
If it still throws the same “Managed Codex app-server binary was not found” after those two steps, paste:
openclaw plugins list --verbose
openclaw plugins inspect codex --json
That should tell us whether this is plugin package resolution versus gateway process PATH.
real patio
openclaw.json
{
"agents": {
"defaults": {
"model": {
"primary": "openai/gpt-5.5-pro"
},
"models": {
"openai/gpt-5.5": {
"alias": "GPT"
},
"openai/gpt-5.5-pro": {}
},
"workspace": "C:\Users\ibrah\.openclaw\workspace",
"contextInjection": "always",
"bootstrapMaxChars": 50000,
"bootstrapTotalMaxChars": 300000
}
},
"tools": {
"profile": "full",
"web": {
"search": {
"provider": "gemini",
"enabled": true
}
}
},
"session": {
"dmScope": "per-channel-peer"
},
"gateway": {
"port": 18789,
"mode": "local",
"bind": "loopback",
"auth": {
"mode": "token",
"token": "d3212ad35ecf9c738903e767b98c23230975c04d1ca279d1"
},
"tailscale": {
"mode": "off",
"resetOnExit": false
},
"controlUi": {
"allowInsecureAuth": true
},
"nodes": {
"denyCommands": [
"camera.snap",
"camera.clip",
"screen.record",
"contacts.add",
"calendar.add",
"reminders.add",
"sms.send",
"sms.search"
]
}
},
"plugins": {
"entries": {
"openai": {
"enabled": true,
"config": {
appServer: {command: "C:\Users\ibrah\AppData\Roaming\pm\codex.cmd"},
"codex": {
"enabled": true
},
"google": {
"enabled": true,
"config": {
"webSearch": {
"apiKey": "AIzaSyCAOB9FEFuglX6YTYgRVErJaQ0aw0TqmWw"
}
}
}
}
},
"auth": {
"profiles": {
"openai:default": {
"provider": "openai",
"mode": "api_key"
}
}
},
"skills": {
"entries": {
"sag": {
"apiKey": "c875d66dd021cb282364287a24f8673bd44ab1cd15e6c64dea70159a998affd4"
}
}
},
"wizard": {
"lastRunAt": "2026-05-15T19:24:54.749Z",
"lastRunVersion": "2026.5.12",
"lastRunCommand": "onboard",
"lastRunMode": "local"
},
"meta": {
"lastTouchedVersion": "2026.5.12",
"lastTouchedAt": "2026-05-15T19:27:30.049Z"
}
}
flat foxBOT
openclaw.json
{
"agents": {
"defaults": {
"model": {
"prima...
First: you pasted live secrets in that config. Please rotate anything exposed there: the gateway token, Google API key, and the skill/API key. Also edit/delete the message if Discord lets you.
Config issue: the Codex settings are in the wrong place and the path has a typo.
You currently put appServer inside plugins.entries.openai.config. It needs to be under plugins.entries.codex.config.
You also have:
C:\Users\ibrah\AppData\Roaming\pm\codex.cmd
but your where.exe codex showed:
C:\Users\ibrah\AppData\Roaming pm\codex.cmd
Use this shape for just the plugins section:
flat foxBOT
openclaw.json
{
"agents": {
"defaults": {
"model": {
"prima...
{
plugins: {
allow: ["openai", "codex", "google"],
entries: {
openai: {
enabled: true
},
codex: {
enabled: true,
config: {
appServer: {
command: "C:\\Users\\ibrah\\AppData\\Roaming\
pm\\codex.cmd"
}
}
},
google: {
enabled: true,
config: {
webSearch: {
apiKey: "YOUR_ROTATED_GOOGLE_KEY"
}
}
}
}
}
}
Don’t paste the whole config again with secrets. After editing, restart OpenClaw completely and run:
flat foxBOT
openclaw.json
{
"agents": {
"defaults": {
"model": {
"prima...
openclaw doctor
openclaw plugins doctor
so this is my actuall json
{
"agents": {
"defaults": {
"model": {
"primary": "openai/gpt-5.5-pro"
},
"models": {
"openai/gpt-5.5": {
"alias": "GPT"
},
"openai/gpt-5.5-pro": {}
},
"workspace": "C:\Users\ibrah\.openclaw\workspace",
"contextInjection": "always",
"bootstrapMaxChars": 50000,
"bootstrapTotalMaxChars": 300000
}
},
"tools": {
"profile": "full",
"web": {
"search": {
"provider": "gemini",
"enabled": true
}
}
},
"session": {
"dmScope": "per-channel-peer"
},
"gateway": {
"port": 18789,
"mode": "local",
"bind": "loopback",
"auth": {
"mode": "token",
"token": "d3212ad35ecf9c738903e767b98c23230975c04d1ca279d1"
},
"tailscale": {
"mode": "off",
"resetOnExit": false
},
"controlUi": {
"allowInsecureAuth": true
},
"nodes": {
"denyCommands": [
"camera.snap",
"camera.clip",
"screen.record",
"contacts.add",
"calendar.add",
"reminders.add",
"sms.send",
"sms.search"
]
}
},
"plugins": {
"entries": {
"openai": {
"enabled": true
},
"codex": {
"enabled": true
},
"google": {
"enabled": true,
"config": {
"webSearch": {
"apiKey": "AIzaSyCAOB9FEFuglX6YTYgRVErJaQ0aw0TqmWw"
}
}
}
}
},
"auth": {
"profiles": {
"openai:default": {
"provider": "openai",
"mode": "api_key"
}
}
},
"skills": {
"entries": {
"sag": {
"apiKey": "c875d66dd021cb282364287a24f8673bd44ab1cd15e6c64dea70159a998affd4"
}
}
},
"wizard": {
"lastRunAt": "2026-05-15T19:24:54.749Z",
"lastRunVersion": "2026.5.12",
"lastRunCommand": "onboard",
"lastRunMode": "local"
},
"meta": {
"lastTouchedVersion": "2026.5.12",
"lastTouchedAt": "2026-05-15T19:27:30.049Z"
}
}
flat foxBOT
so this is my actuall json
{
"agents": {
"defaults": {
"model": {
...
Yep, this actual JSON explains it.
Your codex plugin is enabled, but it does not have the explicit app-server command yet. Add it under plugins.entries.codex.config, not under openai.
Replace just your plugins block with this shape:
"plugins": {
"allow": ["openai", "codex", "google"],
"entries": {
"openai": {
"enabled": true
},
"codex": {
"enabled": true,
"config": {
"appServer": {
"command": "C:\\Users\\ibrah\\AppData\\Roaming\
pm\\codex.cmd"
}
}
},
"google": {
"enabled": true,
"config": {
"webSearch": {
"apiKey": "YOUR_ROTATED_GOOGLE_KEY"
}
}
}
}
}
Then fully restart OpenClaw/gateway.
Important: you pasted live secrets again. Please rotate them and delete/edit the message if you can. For the gateway token on PowerShell, you can generate and set a new one like:
$newToken = [System.Convert]::ToHexString([System.Security.Cryptography.RandomNumberGenerator]::GetBytes(32)).ToLower()
openclaw config set gateway.auth.token $newToken
Also rotate the Google API key and the sag API key wherever those came from. Don’t paste the full config again; paste only redacted snippets or command output.
real patio
21:38:42 [agents/harness] Codex agent harness failed; not falling back to embedded PI backend
21:38:42 [diagnostic] lane task error: lane=main durationMs=3997 error="Error: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary."
21:38:42 [diagnostic] lane task error: lane=session:agent:main:dashboard:6d2e3508-cd02-476d-91b7-6da30e420b21 durationMs=4001 error="Error: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary."
21:38:42 Embedded agent failed before reply: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary.
flat foxBOT
21:38:42 [agents/harness] Codex agent harness failed; not falling back to embedd...
Same error means OpenClaw is still not seeing the override, or the running gateway has not reloaded it.
Use the CLI to write the exact key so there’s no JSON nesting/escaping mistake:
openclaw config set plugins.allow '["openai","codex","google"]' --strict-json
openclaw config set plugins.entries.codex.config.appServer.command "C:\Users\ibrah\AppData\Roaming
pm\codex.cmd"
openclaw config get plugins.entries.codex.config.appServer.command
openclaw config validate
The config get output should be exactly:
C:\Users\ibrah\AppData\Roaming
pm\codex.cmd
Then fully quit/restart the OpenClaw app or gateway. If you are testing in the same existing dashboard/session, start a fresh session after restart too, because harness selection can be pinned on an existing session.
If it still fails after that, test whether it’s only the app/service environment by running the gateway from that same PowerShell:
$env:OPENCLAW_CODEX_APP_SERVER_BIN="C:\Users\ibrah\AppData\Roaming
pm\codex.cmd"
openclaw gateway --verbose
Then trigger one agent reply. If that works, the installed app/service is not reading the config/env we think it is.
Please don’t paste the full config again; it contains secrets. Paste only the redacted config get, config validate, and any new error lines.
real patio
21:41:06 [diagnostic] lane enqueue: lane=session:agent:main:dashboard:39831660-95b1-4ce0-a571-0d5a85c45ffe queueSize=1
21:41:06 [diagnostic] lane dequeue: lane=session:agent:main:dashboard:39831660-95b1-4ce0-a571-0d5a85c45ffe waitMs=0 queueSize=0
21:41:06 [diagnostic] lane enqueue: lane=main queueSize=1
21:41:06 [diagnostic] lane dequeue: lane=main waitMs=0 queueSize=0
21:41:06 [ws] → event tick seq=per-client clients=1
21:41:06 [agents/harness] agent harness selected
21:41:09 [plugins] loading memory-core from C:\Users\ibrah\AppData\Roaming\npm\node_modules\openclaw\dist\extensions\memory-core\index.js
21:41:09 [plugins] loaded 9 plugin(s) (1 attempted) in 37.3ms
21:41:10 [ws] → event agent seq=per-client clients=1 run=45b01e8c…ac7b agent=main session=dashboard:39831660-95b1-4ce0-a571-0d5a85c45ffe stream=codex_app_server.lifecycle aseq=1
21:41:11 [agents/harness] Codex agent harness failed; not falling back to embedded PI backend
21:41:11 [diagnostic] lane task error: lane=main durationMs=4851 error="Error: codex.cmd wrapper resolved, but no executable/Node entrypoint could be resolved without shell execution."
21:41:11 [diagnostic] lane task error: lane=session:agent:main:dashboard:39831660-95b1-4ce0-a571-0d5a85c45ffe durationMs=4856 error="Error: codex.cmd wrapper resolved, but no executable/Node entrypoint could be resolved without shell execution."21:41:11 [ws] → event agent seq=per-client clients=1 run=45b01e8c…ac7b agent=main session=dashboard:39831660-95b1-4ce0-a571-0d5a85c45ffe stream=lifecycle aseq=2 phase=error error=codex.cmd wrapper resolved, but no executable/Node entrypoint could be resolved without shell execution.
21:41:11 Embedded agent failed before reply: codex.cmd wrapper resolved, but no executable/Node entrypoint could be resolved without shell execution.
flat foxBOT
21:41:06 [diagnostic] lane enqueue: lane=session:agent:main:dashboard:39831660-9...
Good, this is progress. The config override is being read now. New error means OpenClaw is trying to spawn the Windows .cmd shim directly, and this path is blocked unless it can unwrap the shim safely.
Point OpenClaw at Codex’s JS entrypoint instead of codex.cmd:
openclaw config set plugins.entries.codex.config.appServer.command "C:\Users\ibrah\AppData\Roaming
pm
ode_modules\@openai\codex\bin\codex.js"
openclaw config get plugins.entries.codex.config.appServer.command
openclaw config validate
Then fully restart OpenClaw/gateway and start a fresh dashboard session.
If that path does not exist on your machine, check it with:
flat foxBOT
21:41:06 [diagnostic] lane enqueue: lane=session:agent:main:dashboard:39831660-9...
Test-Path "C:\Users\ibrah\AppData\Roaming
pm
ode_modules\@openai\codex\bin\codex.js"
dir "C:\Users\ibrah\AppData\Roaming
pm
ode_modules\@openai\codex\bin"
The reason: @openai/codex@0.130.0 exposes codex as bin/codex.js. OpenClaw’s Windows spawn path can run JS entrypoints via Node, but it is refusing the unresolved .cmd wrapper without shell execution.
real patio
reinstalled all from sratch and still dosent work
22:11:05 [plugins] plugins.allow is empty; discovered non-bundled plugins may auto-load: acpx (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\acpx\dist\index.js), codex (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\codex\dist\index.js), diagnostics-otel (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\diagnostics-otel\dist\index.js), diagnostics-prometheus (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\diagnostics-prometheus\dist\index.js), diffs (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\diffs\dist\index.js), google-meet (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\google-meet\dist\index.js) (+3 more). Set plugins.allow to explicit trusted ids.
22:11:10 [agents/harness] Codex agent harness failed; not falling back to embedded PI backend
22:11:10 [diagnostic] lane task error: lane=main durationMs=4655 error="Error: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary."
22:11:10 [diagnostic] lane task error: lane=session:agent:main:main durationMs=4660 error="Error: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary."
22:11:10 Embedded agent failed before reply: Managed Codex app-server binary was not found for @openai/codex. Reinstall or update OpenClaw, or run pnpm install in a source checkout. Set plugins.entries.codex.config.appServer.command or OPENCLAW_CODEX_APP_SERVER_BIN to use a custom Codex binary.
flat foxBOT
reinstalled all from sratch and still dosent work
That’s frustrating, but reinstalling probably won’t fix this if the config/session is still selecting the Codex harness and Windows still can’t spawn the app-server path.
Let’s split this into “get you running” vs “debug Codex harness”:
Fast workaround to get OpenClaw replying again: disable the Codex harness/plugin and use the normal provider path for now.
openclaw config set plugins.entries.codex.enabled false --strict-json
openclaw config unset agents.defaults.embeddedHarness
openclaw config validate
Then fully restart OpenClaw/gateway and start a fresh dashboard session.
If you want to keep debugging Codex harness, paste only these outputs, no full config/secrets:
flat foxBOT
reinstalled all from sratch and still dosent work
openclaw --version
codex --version
openclaw config get plugins.entries.codex.config.appServer.command
Test-Path "C:\Users\ibrah\AppData\Roaming
pm
ode_modules\@openai\codex\bin\codex.js"
openclaw config validate
And paste the newest error lines after reinstall. If it’s still codex.cmd wrapper resolved, then it’s still pointed at the .cmd shim. If it’s back to Managed Codex app-server binary was not found, then the override got lost or the gateway is reading a different config file.
flat foxBOT
22:11:05 [plugins] plugins.allow is empty; discovered non-bundled plugins may au...
This log shows the reinstall wiped/changed the config back: plugins.allow is empty, and the Codex app-server override is gone, so OpenClaw is back to trying its managed @openai/codex binary.
At this point I’d stop trying to fix native Codex harness in-place and first get OpenClaw working again:
openclaw config file
openclaw config set plugins.entries.codex.enabled false --strict-json
openclaw config unset agents.defaults.embeddedHarness
openclaw config validate
Then fully restart OpenClaw/gateway and start a fresh session.
If you still want Codex harness after that, set the JS entrypoint again and verify the value stuck:
openclaw config set plugins.entries.codex.config.appServer.command "C:\Users\ibrah\AppData\Roaming
pm
ode_modules\@openai\codex\bin\codex.js"
openclaw config get plugins.entries.codex.config.appServer.command
Test-Path "C:\Users\ibrah\AppData\Roaming
pm
ode_modules\@openai\codex\bin\codex.js"
openclaw config validate
If Test-Path returns False, Codex is installed somewhere else. In that case run:
npm root -g
and use that directory plus \@openai\codex\bin\codex.js.
Also, please paste the output of openclaw config file if it keeps ignoring changes. That will tell us whether the app is reading a different config than the CLI is editing.
real patio
22:18:10 [plugins] plugins.allow is empty; discovered non-bundled plugins may auto-load: acpx (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\acpx\dist\index.js), codex (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\codex\dist\index.js), diagnostics-otel (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\diagnostics-otel\dist\index.js), diagnostics-prometheus (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\diagnostics-prometheus\dist\index.js), diffs (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\diffs\dist\index.js), google-meet (C:\Users\ibrah.openclaw\npm\node_modules@openclaw\google-meet\dist\index.js) (+3 more). Set plugins.allow to explicit trusted ids.
flat foxBOT
22:18:10 [plugins] plugins.allow is empty; discovered non-bundled plugins may au...
That line alone is just the plugin allowlist warning. It means plugins.allow is still empty, not necessarily that the agent is failing.
Two checks:
openclaw config get plugins.allow
openclaw config get plugins.entries.codex.enabled
If you want to quiet that warning, set an explicit allowlist for the plugins you actually use. For your earlier setup, something like:
openclaw config set plugins.allow '["openai","google","acpx","diffs","memory-core"]' --strict-json
openclaw config validate
Leave codex out if you’re disabling Codex harness for now. Then restart the gateway.
flat foxBOT
22:18:10 [plugins] plugins.allow is empty; discovered non-bundled plugins may au...
The key question: after this warning, does OpenClaw actually reply now, or do you still get a later Codex agent harness failed error? If there’s no later harness error, this is just cleanup.
real patio
22:19:09 [diagnostic] liveness warning: reasons=event_loop_delay,cpu interval=67s eventLoopDelayP99Ms=296.7 eventLoopDelayMaxMs=58116.3 eventLoopUtilization=0.904 cpuCoreRatio=0.906 active=1 waiting=0 queued=1 recentPhases=sidecars.subagent-recovery:4ms,sidecars.main-session-recovery:3ms,sidecars.restart-sentinel:19ms,post-attach.update-sentinel:8ms,sidecars.session-locks:21ms,post-ready.maintenance:197ms work=[active=agent:main:dashboard:39831660-95b1-4ce0-a571-0d5a85c45ffe(processing,q=1,age=59s) queued=agent:main:dashboard:39831660-95b1-4ce0-a571-0d5a85c45ffe(processing,q=1,age=59s)]
22:19:43 [agent/embedded] unsupported thinking level for openai-codex/gpt-5.5-pro; retrying with off
22:19:49 [agent/embedded] embedded run failover decision: runId=78efdb6a-9542-4565-a405-73cf443ec128 stage=prompt decision=surface_error reason=none from=openai-codex/gpt-5.5-pro profile=sha256:df8169cdbcc6 rawError={"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.5-pro' model is not supported when using Codex with a ChatGPT account."}}
22:19:49 [diagnostic] lane task error: lane=main durationMs=27701 error="{"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.5-pro' model is not supported when using Codex with a ChatGPT account."}}"
22:19:49 [diagnostic] lane task error: lane=session:agent:main:dashboard:39831660-95b1-4ce0-a571-0d5a85c45ffe durationMs=27705 error="{"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.5-pro' model is not supported when using Codex with a ChatGPT account."}}"
22:19:49 Embedded agent failed before reply: {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.5-pro' model is not supported when using Codex with a ChatGPT account."}}
flat foxBOT
22:19:09 [diagnostic] liveness warning: reasons=event_loop_delay,cpu interval=67...
Now the original Codex app-server problem is gone. This is a different issue:
gpt-5.5-pro is not supported with your Codex/ChatGPT account path.
Your config is selecting:
"primary": "openai/gpt-5.5-pro"
but the run is going through openai-codex/gpt-5.5-pro, and Codex is rejecting that model for your account.
Quick fix: switch your default model away from the -pro Codex route, then restart/start a fresh session:
openclaw config set agents.defaults.model.primary "openai-codex/gpt-5.5"