#Codex Issue

1 messages · Page 1 of 1 (latest)

remote novaBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> âś… Mark Solution

small treeBOT
split fjordBOT
#

First check the version, because Codex harness packaging has changed a bit:

openclaw --version

That error means OpenClaw selected the Codex app-server harness, but it cannot find the Codex app-server executable it expects. Try this in order:

openclaw update
openclaw doctor
openclaw gateway restart

If you are running OpenClaw from a source checkout, go to that checkout and rebuild deps/assets:

# 
pnpm install
pnpm build
pnpm ui:build
openclaw doctor
openclaw gateway restart

If you installed manually with npm/pnpm, reinstall latest and restart:

npm i -g openclaw@latest
openclaw doctor
openclaw gateway restart

Also verify Codex itself is usable from the same shell/user that runs the gateway:

# 
command -v codex
codex app-server --listen stdio://

If codex exists somewhere non-standard, set it explicitly in config:

{
  plugins: {
    entries: {
      codex: {
        enabled: true,
        config: {
          appServer: {
            command: "/absolute/path/to/codex"
hollow zodiac
#

Failed to refresh gateway service environment from updated install: Error: updated install refresh failed (C:\Users\racki\AppData\Roaming\npm\node_modules\openclaw\dist\index.js): Gateway install failed: Error: schtasks create failed: FEHLER: Zugriff verweigert

split fjordBOT
# hollow zodiac Failed to refresh gateway service environment from updated install: Error: updat...

That one is Windows permissions: FEHLER: Zugriff verweigert = Access denied while OpenClaw tried to create/refresh the Windows Scheduled Task for the gateway.

First, check whether the update itself actually succeeded:

openclaw --version
openclaw gateway status
openclaw status

Then fix the daemon refresh from PowerShell. If you installed the gateway before from an elevated/admin shell, run this once in PowerShell as Administrator:

openclaw gateway install --force
openclaw gateway restart
openclaw doctor
split fjordBOT
# hollow zodiac Failed to refresh gateway service environment from updated install: Error: updat...

If you want to avoid the Scheduled Task entirely, newest OpenClaw should fall back to a per-user Startup-folder launcher when schtasks is denied, so updating again from a normal user shell can also help:

npm i -g openclaw@latest
openclaw gateway install --force
openclaw gateway restart

Docs: https://docs.openclaw.ai/gateway and https://docs.openclaw.ai/install/updating

If it still says access denied, paste the outputs of openclaw --version and openclaw gateway status --json. My suspicion is a stale Scheduled Task owned by a different/elevated user, not the Codex binary anymore.

#

This means the Scheduled Task exists, but nothing is listening on 127.0.0.1:18789. So the gateway either was not started by the task, or it starts and crashes immediately.

Run these next:

openclaw gateway restart
openclaw gateway status
openclaw logs --follow

If logs --follow does not show the crash, open the exact log file from your status output:

Get-Content "$env:LOCALAPPDATA\Temp\openclaw\openclaw-2026-05-18.log" -Tail 120

Also useful on native Windows:

#

Good, the log has the real issue now. The gateway is starting, but the Codex plugin and the OpenClaw core are out of sync:

@openclaw/codex ... expects buildAgentHookContextChannelFields
OpenClaw 2026.5.12 does not export it

Your update log says:

Before: 2026.5.12
After: 2026.5.12
Installing @openclaw/codex ...

So the plugin moved forward, but core stayed on 2026.5.12. Fix is to make them the same version.

#

Recommended path: update core, then plugin, then restart:

npm i -g openclaw@latest
openclaw --version
openclaw plugins update codex
openclaw gateway restart
openclaw gateway status

If openclaw --version still says 2026.5.12 after npm install, check whether Windows is picking up an old shim:

where openclaw
npm root -g
npm view openclaw version
#

Temporary fallback if you cannot update core yet: pin the Codex plugin back to your current core version:

openclaw plugins install @openclaw/codex@2026.5.12 --force
openclaw gateway restart

But I’d try the full core update first, because the original missing app-server binary problem may already be fixed in newer OpenClaw/Codex plugin pairing.

Docs: https://docs.openclaw.ai/install/updating and https://docs.openclaw.ai/cli/plugins

hollow zodiac
#

openclaw gateway status

OpenClaw 2026.5.18 (50a2481) — Your personal assistant, minus the passive-aggressive calendar reminders.

|
o
Service: Scheduled Task (registered)
File logs: ~\AppData\Local\Temp\openclaw\openclaw-2026-05-18.log
Command: C:\Program Files\nodejs\node.exe C:\Users\racki\AppData\Roaming\npm\node_modules\openclaw\dist\index.js gateway --port 18789
Service file: ~.openclaw\gateway.cmd
Service env: OPENCLAW_GATEWAY_PORT=18789

Config (cli): ~.openclaw\openclaw.json
Config (service): ~.openclaw\openclaw.json

Gateway: bind=loopback (127.0.0.1), port=18789 (service args)
Probe target: ws://127.0.0.1:18789
Dashboard: http://127.0.0.1:18789/
Probe note: Loopback-only gateway; only local clients can connect.

CLI version: 2026.5.18 (~\AppData\Roaming\npm\node_modules\openclaw\openclaw.mjs)
Gateway version: 2026.5.18

Runtime: running (pid 32264, state Wird ausgef�hrt, Verified gateway listener detected on port 18789 even though schtasks did not report a running task.)
Connectivity probe: ok
Capability: read-only

Troubles: run openclaw status
Troubleshooting: https://docs.openclaw.ai/troubleshooting

split fjordBOT
# hollow zodiac openclaw gateway status OpenClaw 2026.5.18 (50a2481) — Your personal assistant,...

That status is healthy now:

  • CLI and Gateway both 2026.5.18
  • Connectivity probe: ok
  • Gateway listener is live on 127.0.0.1:18789

Next test is the actual Codex harness path: send one short message to the agent from the UI/Telegram and see if it replies. If it still fails, grab only the new Codex lines:

openclaw logs --follow

The Capability: read-only line is fine for the status probe; it just means the status command verified read RPC access, not that your agent can’t use tools.

If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution so Answer Overflow indexes the fix for the next Windows/Codex person.

hollow zodiac
split fjordBOT
# hollow zodiac

This log is mostly old history. The important part is that after the update/restart I don’t see the old Codex export error repeating yet. The latest useful lines are:

18:58:22 codex is up to date (2026.5.18)
18:58:46 gateway listening
18:58:47 starting channels and sidecars

Two quick things now:

openclaw doctor --fix
openclaw gateway restart

Then hard-refresh the dashboard tab, because your older log shows the Control UI client was still v2026.5.12 for a bit:

split fjordBOT
# hollow zodiac 
Ctrl+Shift+R

After that, send a fresh test message. If it fails, paste only the newest tail after the test:

Get-Content "$env:LOCALAPPDATA\Temp\openclaw\openclaw-2026-05-18.log" -Tail 80

Right now this looks much closer: the previous hard blocker was version mismatch, and that part appears fixed.

hollow zodiac
split fjordBOT
# hollow zodiac

That’s progress: Codex is now actually being selected and starting.

Key line:

strict-agentic execution contract active ... provider=openai-codex/gpt-5.5 harness=codex

So the previous “missing app-server binary” and plugin/core mismatch are past us.

Two remaining things I see:

  1. Your dashboard is still loading an old cached UI:
client=openclaw-control-ui webchat v2026.5.12

Do a hard refresh on the dashboard tab: Ctrl+Shift+R. If it stays old, close the tab and reopen http://127.0.0.1:18789/

split fjordBOT
# hollow zodiac
  1. Telegram has a slow/timeout call:
fetch timeout ... api.telegram.org ... getMe

That may just be network latency. If Telegram replies eventually, ignore it. If Telegram does not reply, test from the dashboard first so we separate “Codex works” from “Telegram polling is flaky”.

Give the Codex run ~1-2 minutes, then if there’s still no reply paste the next tail after 21:07:38:

Get-Content "$env:LOCALAPPDATA\Temp\openclaw\openclaw-2026-05-18.log" -Tail 120
hollow zodiac
#

Something went wrong while processing your request. Please try again, or use /new to start a fresh session.

split fjordBOT
# hollow zodiac Something went wrong while processing your request. Please try again, or use /ne...

Use /new here. After changing the harness/plugin versions, the existing session can still carry bad state, and the error text is basically telling us to start a clean session.

Then send a tiny test message like hi.

If it fails again, paste the fresh log tail immediately after that failure:

Get-Content "$env:LOCALAPPDATA\Temp\openclaw\openclaw-2026-05-18.log" -Tail 160

We need the lines after the strict-agentic execution contract active entry. The current symptom is generic; the next log tail should show whether Codex app-server rejected auth, crashed, timed out, or hit a session-state problem.

hollow zodiac
split fjordBOT
# hollow zodiac

This paste cuts off before the actual failure, so it doesn’t include the Codex result after 21:07:38.

Run this instead so we get only the relevant lines and not the whole JSON log history:

Select-String -Path "$env:LOCALAPPDATA\Temp\openclaw\openclaw-2026-05-18.log" -Pattern "21:07|21:08|21:09|21:10|21:11|21:12|21:13|agent|codex|Embedded agent|error|failed" | Select-Object -Last 80

Also: the dashboard is still connecting as webchat v2026.5.12, which means browser cache is definitely stale. Close all OpenClaw dashboard tabs, open a new tab to http://127.0.0.1:18789/, then press Ctrl+F5 or Ctrl+Shift+R.

The gateway itself is healthy; we’re now chasing the final Codex run error, not the earlier install problem.

hollow zodiac
split fjordBOT
# hollow zodiac

Ok, now we’re back to the original root cause, but with the versions fixed: the Codex plugin is loading, but its managed @openai/codex app-server dependency is missing/not discoverable.

Do this next from PowerShell:

openclaw gateway stop
openclaw plugins install @openclaw/codex@2026.5.18 --force
openclaw plugins doctor
openclaw gateway restart

Then check whether the Codex binary exists in the plugin runtime:

Test-Path "$env:USERPROFILE\.openclaw
pm
split fjordBOT
# hollow zodiac 
ode_modules\.bin\codex.cmd"
Get-ChildItem "$env:USERPROFILE\.openclaw
pm
ode_modules" -Recurse -Filter "codex.cmd" -ErrorAction SilentlyContinue | Select-Object -First 10 FullName

If that still returns nothing, use the explicit-binary fallback:

npm i -g @openai/codex@latest
where codex

Then set the Codex app-server command to the codex.cmd path from where codex in your config:

split fjordBOT
# hollow zodiac 
{
  plugins: {
    entries: {
      codex: {
        enabled: true,
        config: {
          appServer: {
            command: "C:\\Users\\racki\\AppData\\Roaming\
pm\\codex.cmd"
          }
        }
      }
    }
  }
}
split fjordBOT
# hollow zodiac

Restart after that:

openclaw gateway restart

The important bit: @openclaw/codex@2026.5.18 depends on @openai/codex, but your install doesn’t have the app-server binary where the plugin expects it. Force-reinstalling the plugin should repair that; the global @openai/codex + explicit appServer.command path is the fallback.

hollow zodiac
#

Test-Path "$env:USERPROFILE.openclaw pmode_modules.bin\codex.cmd" Get-ChildItem "$env:USERPROFILE.openclaw pm ode_modules" -Recurse -Filter "codex.cmd" -ErrorAction SilentlyContinue | Select-Object -First 10 FullName
Test-Path : Es wurde kein Positionsparameter gefunden, der das Argument "Get-ChildItem" akzeptiert.
In Zeile:1 Zeichen:1

  • Test-Path "$env:USERPROFILE.openclaw pmode_modules.bin\codex.cmd" G ...
  •   + CategoryInfo          : InvalidArgument: (:) [Test-Path], ParameterBindingException
  + FullyQualifiedErrorId : PositionalParameterNotFound,Microsoft.PowerShell.Commands.TestPathCommand
split fjordBOT
split fjordBOT
hollow zodiac
#

PS C:\Users\racki> Test-Path "$env:USERPROFILE.openclaw pm ode_modules.bin\codex.cmd"; Get-ChildItem "$env:USERPROFILE.openclaw pm ode_modules" -Recurse -Filter "codex.cmd" -ErrorAction SilentlyContinue | Select-Object -First 10 FullName

False
PS C:\Users\

split fjordBOT
split fjordBOT
hollow zodiac
#

PS C:\Users\racki> cd $env:USERPROFILE
PS C:\Users\racki> cd $env:USERPROFILE
PS C:\Users\racki> cd .openclaw
PS C:\Users\racki.openclaw> cd npm
PS C:\Users\racki.openclaw\npm> cd node_modules
PS C:\Users\racki.openclaw\npm\node_modules> dir .bin\codex.cmd

Verzeichnis: C:\Users\racki\.openclaw\npm\node_modules\.bin

Mode LastWriteTime Length Name

-a---- 18.05.2026 19:55 331 codex.cmd

PS C:\Users\racki.openclaw\npm\node_modules> dir -Recurse codex.cmd

Verzeichnis: C:\Users\racki\.openclaw\npm\node_modules\.bin

Mode LastWriteTime Length Name

-a---- 18.05.2026 19:55 331 codex.cmd

split fjordBOT
# hollow zodiac PS C:\Users\racki> cd $env:USERPROFILE PS C:\Users\racki> cd $env:USERPROFILE PS...

Good, the binary exists. So the quickest fix is to stop relying on auto-discovery and point OpenClaw at it explicitly.

First test it directly:

C:\Users\racki\.openclaw
pm
ode_modules\.bin\codex.cmd app-server --listen stdio://

If it prints/starts without “not recognized” or missing module errors, stop it with Ctrl+C.

Then add this to C:\Users\racki\.openclaw\openclaw.json under your existing plugins.entries.codex config:

{
split fjordBOT
split fjordBOT
hollow zodiac
#

C:\Users\racki.openclaw : Die Benennung "C:\Users\racki.openclaw" wurde nicht als Name eines Cmdlet, einer Funktion, einer Skriptdatei oder eines ausfĂĽhrbaren Programms erkannt. ĂśberprĂĽfen Sie die
Schreibweise des Namens, oder ob der Pfad korrekt ist (sofern enthalten), und wiederholen Sie den Vorgang.
In Zeile:1 Zeichen:1

  • C:\Users\racki.openclaw
  •   + CategoryInfo          : ObjectNotFound: (C:\Users\racki\.openclaw:String) [], CommandNotFoundException
  + FullyQualifiedErrorId : CommandNotFoundException

PS C:\Users\racki.openclaw\npm\node_modules> pm
pm : Die Benennung "pm" wurde nicht als Name eines Cmdlet, einer Funktion, einer Skriptdatei oder eines ausfĂĽhrbaren Programms erkannt. ĂśberprĂĽfen Sie die Schreibweise des Namens, oder ob der Pfad korrekt
ist (sofern enthalten), und wiederholen Sie den Vorgang.
In Zeile:1 Zeichen:1

  • pm
  • ~~
    • CategoryInfo : ObjectNotFound: (pm:String) [], CommandNotFoundException
    • FullyQualifiedErrorId : CommandNotFoundException

PS C:\Users\racki.openclaw\npm\node_modules> ode_modules.bin\codex.cmd app-server --listen stdio://
ode_modules.bin\codex.cmd : Das Modul "ode_modules" konnte nicht geladen werden. Wenn Sie weitere Informationen wĂĽnschen, fĂĽhren Sie "Import-Module ode_modules" aus.
In Zeile:1 Zeichen:1

  • ode_modules.bin\codex.cmd app-server --listen stdio://
  •   + CategoryInfo          : ObjectNotFound: (ode_modules\.bin\codex.cmd:String) [], CommandNotFoundException
  + FullyQualifiedErrorId : CouldNotAutoLoadModule
split fjordBOT
split fjordBOT
hollow zodiac
#

S C:\Users\racki.openclaw\npm\node_modules> pm
pm : Die Benennung "pm" wurde nicht als Name eines Cmdlet, einer Funktion, einer Skriptdatei oder eines ausfĂĽhrbaren Programms erkannt. ĂśberprĂĽfen Sie die Schreibweise des Namens, oder ob der Pfad korrekt
ist (sofern enthalten), und wiederholen Sie den Vorgang.
In Zeile:1 Zeichen:1

  • pm
  • ~~
    • CategoryInfo : ObjectNotFound: (pm:String) [], CommandNotFoundException
    • FullyQualifiedErrorId : CommandNotFoundException

PS C:\Users\racki.openclaw\npm\node_modules> ode_modules.bin\codex.cmd app-server --listen stdio://
ode_modules.bin\codex.cmd : Das Modul "ode_modules" konnte nicht geladen werden. Wenn Sie weitere Informationen wĂĽnschen, fĂĽhren Sie "Import-Module ode_modules" aus.
In Zeile:1 Zeichen:1

  • ode_modules.bin\codex.cmd app-server --listen stdio://
  •   + CategoryInfo          : ObjectNotFound: (ode_modules\.bin\codex.cmd:String) [], CommandNotFoundException
  + FullyQualifiedErrorId : CouldNotAutoLoadModule

PS C:\Users\racki.openclaw\npm\node_modules> ..bin\codex.cmd app-server --listen stdio://
& "C:\Users\racki.openclaw
2026-05-18T19:23:45.096506Z ERROR codex_app_server_transport::transport: Failed to deserialize JSONRPCMessage: expected value at line 1 column 1
pm
2026-05-18T19:23:45.096736Z ERROR codex_app_server_transport::transport: Failed to deserialize JSONRPCMessage: expected value at line 1 column 1
ode_modules.bin\codex.cmd" app-server --listen "stdio://"
2026-05-18T19:23:45.699116Z ERROR codex_app_server_transport::transport: Failed to deserialize JSONRPCMessage: expected value at line 1 column 1

split fjordBOT
# hollow zodiac S C:\Users\racki\.openclaw\npm\node_modules> pm pm : Die Benennung "pm" wurde ni...

That output is actually useful: codex.cmd app-server --listen stdio:// started. The JSONRPC errors happened because extra pasted text was typed into the stdio app-server. That confirms the binary can run.

Press Ctrl+C to stop it, then set the command path with OpenClaw’s config CLI instead of hand-editing JSON:

openclaw config set plugins.entries.codex.config.appServer.command "C:\Users\racki\.openclaw
pm
ode_modules\.bin\codex.cmd"
openclaw config validate
openclaw gateway restart

Then start a clean session and test:

/new
split fjordBOT
hollow zodiac
#

PS C:\Users\racki> openclaw config set plugins.entries.codex.config.appServer.command "C:\Users\racki.openclaw

pm
ode_modules.bin\codex.cmd"
│
â—‡

🦞 OpenClaw 2026.5.18 (50a2481) — I'll butter your workflow like a lobster roll: messy, delicious, effective.

Updated plugins.entries.codex.config.appServer.command. Restart the gateway to apply.
PS C:\Users\racki> openclaw config validate
│
â—‡

🦞 OpenClaw 2026.5.18 (50a2481) — Gateway online—please keep hands, feet, and appendages inside the shell at all times.

Config valid: ~.openclaw\openclaw.json

split fjordBOT
# hollow zodiac PS C:\Users\racki> openclaw config set plugins.entries.codex.config.appServer.co...

That saved a broken path because the line got split again:

C:\Users\racki\.openclaw
pm
ode_modules\.bin\codex.cmd

Set it from your current directory using PowerShell to build the string, so Discord/newlines cannot mangle it:

$CodexCmd = Join-Path $env:USERPROFILE ".openclaw
pm
ode_modules\.bin\codex.cmd"
openclaw config set plugins.entries.codex.config.appServer.command $CodexCmd
openclaw config get plugins.entries.codex.config.appServer.command
split fjordBOT
hollow zodiac
#

S C:\Users\racki> $CodexCmd = Join-Path $env:USERPROFILE ".openclaw

pm
ode_modules.bin\codex.cmd"
PS C:\Users\racki> openclaw config set plugins.entries.codex.config.appServer.command $CodexCmd
│
â—‡

🦞 OpenClaw 2026.5.18 (50a2481) — I run on caffeine, JSON5, and the audacity of "it worked on my machine."

Updated plugins.entries.codex.config.appServer.command. Restart the gateway to apply.
PS C:\Users\racki> openclaw config get plugins.entries.codex.config.appServer.command

🦞 OpenClaw 2026.5.18 (50a2481) — Alexa, but with taste.

C:\Users\racki.openclaw
pm
ode_modules.bin\codex.cmd
PS C:\Users\racki> openclaw gateway restart
│
â—‡

split fjordBOT
# hollow zodiac S C:\Users\racki> $CodexCmd = Join-Path $env:USERPROFILE ".openclaw >> pm >> ode...

The config still has literal line breaks in the path. Avoid typing any Windows backslashes for this one; let PowerShell build the path from pieces.

Copy this whole block:

$S = [IO.Path]::DirectorySeparatorChar
$CodexCmd = (Join-Path $env:USERPROFILE '.openclaw') + $S + 'npm' + $S + 'node_modules' + $S + '.bin' + $S + 'codex.cmd'
$CodexCmd
openclaw config set plugins.entries.codex.config.appServer.command $CodexCmd
openclaw config get plugins.entries.codex.config.appServer.command
openclaw gateway restart

The printed $CodexCmd and config get should each be one single line. If either output has line breaks, stop there and paste just that output.

This avoids the accidental backslash+n sequence that keeps turning into a new line on your side.

hollow zodiac
#

PS C:\Users\racki> $S = [IO.Path]::DirectorySeparatorChar
PS C:\Users\racki> $CodexCmd = (Join-Path $env:USERPROFILE '.openclaw') + $S + 'npm' + $S + 'node_modules' + $S + '.bin' + $S + 'codex.cmd'
PS C:\Users\racki> $CodexCmd
C:\Users\racki.openclaw\npm\node_modules.bin\codex.cmd
PS C:\Users\racki> openclaw config set plugins.entries.codex.config.appServer.command $CodexCmd
│
â—‡

🦞 OpenClaw 2026.5.18 (50a2481) — Hot reload for config, cold sweat for deploys.

Updated plugins.entries.codex.config.appServer.command. Restart the gateway to apply.
PS C:\Users\racki> openclaw config get plugins.entries.codex.config.appServer.command

🦞 OpenClaw 2026.5.18 (50a2481) — I've read more man pages than any human should—so you don't have to.

C:\Users\racki.openclaw\npm\node_modules.bin\codex.cmd

split fjordBOT
hollow zodiac
split fjordBOT
# hollow zodiac

This is a different failure now. I don’t see the “Codex binary not found” error anymore. The gateway is freezing while processing a Telegram slash command:

event-loop starvation
active=agent:main:telegram:slash:1674282135(processing,q=1,age=76s)

Do this to isolate Codex from Telegram:

  1. Restart the gateway.
  2. Do not test from Telegram yet.
  3. Open the dashboard at http://127.0.0.1:18789/
  4. Use /new there, then send hi.

Commands:

split fjordBOT
# hollow zodiac 
openclaw gateway restart
openclaw gateway status

If the dashboard test works, Codex is fixed and the remaining issue is Telegram/event-loop starvation. If the dashboard test also freezes, paste the next tail after the dashboard test:

Select-String -Path "$env:LOCALAPPDATA\Temp\openclaw\openclaw-2026-05-18.log" -Pattern "21:28|21:29|21:30|21:31|21:32|agent|codex|app-server|harness|starvation|Embedded|failed|error" | Select-Object -Last 100

We’re close: the explicit Codex path is now accepted, but the next run is hanging hard enough to stall gateway I/O.

hollow zodiac
#

⚠️ 🛠️ "C:\WINDOWS\System32\WindowsPowerShell\v1.0\powershell.exe" -Command 'Get-ChildItem -Name BOOTSTRAP.md, MEMORY.md … (agent) failed

split fjordBOT
# hollow zodiac ⚠️ 🛠️ "C:\\WINDOWS\\System32\\WindowsPowerShell\\v1.0\\powershell.exe" -Command...

That means Codex is now far enough to run a shell tool, so the app-server path problem is likely fixed. New failure is inside the agent’s PowerShell command.

Please click/expand that failed tool entry if the UI lets you and copy the full error text after failed. The important part is usually one line below it: path not found, permission denied, execution policy, command timed out, etc.

Also run this in PowerShell from the same machine:

cd $env:USERPROFILE\.openclaw\workspace
Get-ChildItem -Name BOOTSTRAP.md, MEMORY.md -ErrorAction SilentlyContinue

If that command works manually, the failure is probably Codex’s Windows shell wrapper. If it fails manually too, paste that error and we’ll fix the workspace/path side.

hollow zodiac
#

"C:\WINDOWS\System32\WindowsPowerShell\v1.0\powershell.exe" -Command 'Get-ChildItem -Name BOOTSTRAP.md, MEMORY.md … (agent) all iget

split fjordBOT
# hollow zodiac "C:\\WINDOWS\\System32\\WindowsPowerShell\\v1.0\\powershell.exe" -Command 'Get-C...

Got it, the UI is only showing the collapsed summary, not the real stderr.

Run these directly in PowerShell and paste the output:

cd $env:USERPROFILE\.openclaw\workspace
Get-ChildItem -Name BOOTSTRAP.md, MEMORY.md -ErrorAction SilentlyContinue

Then grab the fresh Codex/tool error from logs:

Select-String -Path "$env:LOCALAPPDATA\Temp\openclaw\openclaw-2026-05-18.log" -Pattern "powershell.exe|Get-ChildItem|tool|exec|failed|stderr|BOOTSTRAP|MEMORY|agent" | Select-Object -Last 120

The good news: we moved past “Codex cannot start.” Now Codex is running and trying to inspect the workspace; we just need the exact reason that tool command failed.

hollow zodiac
split fjordBOT
# hollow zodiac

The manual workspace command returning nothing is fine: it just means neither BOOTSTRAP.md nor root MEMORY.md exists there.

That log paste is mostly older history again. The newest relevant problem from your previous tail is still:

event-loop starvation
active=agent:main:telegram:slash:1674282135

So stop testing through Telegram for one round. Telegram is currently the thing tying up the gateway.

Do this:

openclaw gateway restart
split fjordBOT
# hollow zodiac

Then open the dashboard only:

http://127.0.0.1:18789/

In the dashboard chat, run:

/new
hi

If the dashboard works, Codex is fixed and Telegram is the remaining issue. If the dashboard also hangs, run this tighter log command after the dashboard test:

Select-String -Path "$env:LOCALAPPDATA\Temp\openclaw\openclaw-2026-05-18.log" -Pattern "21:28|21:29|21:30|21:31|21:32|21:33|21:34|agent/embedded|agents/harness|app-server|codex|starvation|Embedded agent|failed" | Select-Object -Last 80
split fjordBOT
# hollow zodiac

Don’t send Telegram messages while doing this test; otherwise we can’t separate Telegram starvation from Codex execution.

hollow zodiac
#

{
"command": ""C:\\WINDOWS\\System32\\WindowsPowerShell\\v1.0\\powershell.exe" -Command 'Get-ChildItem -Name BOOTSTRAP.md, MEMORY.md -ErrorAction SilentlyContinue'",
"cwd": "C:\Users\racki\.openclaw\workspace"
}

that was the telegram error it replied tho

split fjordBOT
# hollow zodiac { "command": "\"C:\\\\WINDOWS\\\\System32\\\\WindowsPowerShell\\\\v1.0\\\\powe...

Ah, got it. That JSON is the tool call summary, not necessarily the actual error. Since it replied, Codex is functionally working now. The Get-ChildItem ... BOOTSTRAP.md, MEMORY.md command is just the agent checking for workspace context files; on your machine it returns nothing because those files aren’t there, which is okay.

So current status:

  • Gateway is healthy.
  • Codex app-server is launching.
  • Telegram replies are happening.
  • The scary PowerShell line is likely just a failed/empty context-file probe, not the original blocker.

The remaining thing I’d do is clean up the stale Telegram/session state if replies feel slow or you keep seeing starvation warnings:

openclaw gateway restart
/new

Then keep one fresh test thread. If it answers normally, you’re done.