OpenClaw `exec` tool on Windows: PowerShell commands producing no output | Friends of the Crustacean 🦞🤝 | Page 1

heady drum Feb 19, 2026, 2:57 PM

#

I'm encountering a critical problem with my OpenClaw agent running on Windows (Windows_NT 10.0.19045 (x64), node v24.13.0, shell=powershell).

The Issue: The exec tool is consistently failing to capture any output from PowerShell commands. Even the most basic commands like ls, pwd, Test-Path, and Get-ChildItem execute without returning any information. This prevents the agent from performing any file system operations (reading, writing, listing directories, or diagnosing code).

Symptoms observed:

exec(command="ls <path>") -> {"exec_response": {"output": "(no output)"}}
exec(command="pwd") -> {"exec_response": {"output": "(no output)"}}
This behavior is consistent across all PowerShell commands attempted.

Context:

Operating System: Windows_NT 10.0.19045 (x64)
Node.js Version: v24.13.0
Agent Model: google/gemini-2.5-flash
Shell: PowerShell

What I've tried/investigated:

Attempted various PowerShell commands for directory listing and path testing.
Conducted a web search which surfaced GitHub issues related to exec tool problems on Windows, including:
- CMD syntax conflicts in PowerShell (#10868)
- Requests for native PowerShell mode (#6443)
- Troubleshooting "no output / stuck" issues potentially linked to proxy environment / process inheritance (#8057)

sturdy nicheBOT Feb 19, 2026, 2:57 PM

#

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

misty ginkgoBOT Feb 19, 2026, 3:06 PM

#

A couple of important Windows-specific gotchas here.

0) First: what OpenClaw version + how is the Gateway started?

Can you paste:

openclaw --version
openclaw gateway status
and how you start the Gateway: foreground (openclaw gateway run) vs the installed Windows service/Scheduled Task (openclaw gateway install)

Reason: on Windows, how the Gateway is launched (especially via Scheduled Task / headless) can affect whether child-process stdout/stderr pipes actually connect, which shows up exactly as “(no output)” for every exec.

1) Quick “is stdout/stderr wired up?” probes

Ask your agent to run these exact exec calls:

exec(command='Write-Output "STDOUT_OK"')
exec(command='Write-Error "STDERR_OK"')

#

If both come back “(no output)”, it’s almost certainly not a PowerShell alias problem — it’s output capture.

2) Try forcing a PTY (often fixes “no output” on Windows)

Some Windows commands (and sometimes PowerShell hosting) behave much better with a pseudo-terminal:

exec(command="pwd", pty=true)
exec(command='Write-Output "PTY_STDOUT_OK"', pty=true)

If pty=true suddenly makes output appear, you’ve found the workaround (and it’s a good clue for a bug report).

Docs: https://docs.openclaw.ai/tools/exec

3) If you’re running via Scheduled Task/service: test in foreground

If you’re currently running the Gateway via the Windows service/Scheduled Task, do one controlled test:

Stop/restart so you can run it in the foreground
Run: openclaw gateway run in an interactive PowerShell window

#

Retry exec(command='Write-Output "STDOUT_OK"')

If foreground works but service-mode doesn’t, that points to a Windows headless/service stdout/stderr inheritance issue.

(And if you’re not on the latest OpenClaw, updating may already fix it.)

4) Strong recommendation: WSL2 if you can

Even if we get native Windows working, the project recommendation is still WSL2, because tooling + shell behavior is much more predictable there.

Docs: https://docs.openclaw.ai/platforms/windows

If you paste the 3 items from step 0 plus the results of the two probes (stdout/stderr) + the pty=true probe, I can tell you which bucket this falls into and what the cleanest fix is.

heady drum Feb 19, 2026, 3:54 PM

#

PS C:\Windows\system32> openclaw --version
2026.2.15
PS C:\Windows\system32>
PS C:\Windows\system32> openclaw gateway status

🦞 OpenClaw 2026.2.15 (3fe22ea) — I can't fix your code taste, but I can fix your build and your backlog.

|
o
Service: Scheduled Task (registered)
File logs: \tmp\openclaw\openclaw-2026-02-19.log
Command: C:\Program Files\nodejs\node.exe C:\Users\User\AppData\Roaming\npm\node_modules\openclaw\dist\index.js gateway --port 18789
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.

Runtime: running (last run time 19/02/2026 14:47:34)
RPC probe: ok

Listening: 127.0.0.1:18789
Troubles: run openclaw status
Troubleshooting: https://docs.openclaw.ai/troubleshooting
PS C:\Windows\system32>

#

PS C:\Windows\system32> exec(command="pwd", pty=true)
At line:1 char:19

exec(command="pwd", pty=true)
```
              ~
```

Missing argument in parameter list.
+ CategoryInfo : ParserError: (:) [], ParentContainsErrorRecordException
+ FullyQualifiedErrorId : MissingArgument

PS C:\Windows\system32> exec(command='Write-Output "PTY_STDOUT_OK"', pty=true)
At line:1 char:44

exec(command='Write-Output "PTY_STDOUT_OK"', pty=true)

Missing argument in parameter list.
+ CategoryInfo : ParserError: (:) [], ParentContainsErrorRecordException
+ FullyQualifiedErrorId : MissingArgument

PS C:\Windows\system32>

#

i was using schtasck to start it.. i have just done openclaw gateway restart. what should i do next?

#

still unable to use powershell comands to look for files on the system when i ask it to

#

and these 2 didnt work exec(command="pwd", pty=true)
exec(command='Write-Output "PTY_STDOUT_OK"', pty=true)

gave me red errors

#

please help @misty ginkgo

misty ginkgoBOT Feb 19, 2026, 4:15 PM

#

If you’re seeing “(no output)” even for basic stuff, we need to figure out which Windows execution path you’re on, because the fix differs.

1) Are you running `exec` on the Windows Gateway or on a Windows node?

If your Gateway itself is running on Windows: exec should be using powershell.exe under the hood.
If your Gateway is elsewhere (WSL/Linux/VPS) and Windows is a paired node, node exec currently tends to run via cmd.exe on Windows — and that’s exactly where “commands run but no stdout captured” can happen.

Can you paste:

openclaw --version
openclaw status
openclaw nodes status (if you use nodes)

2) Run these “stdout sanity” tests (they should always print)

Ask your agent to run:

exec(command='echo STDOUT_OK')

#

That error is just PowerShell parsing — you ran exec(command="...", pty=true) inside PowerShell, but that exec(...) syntax is not a PowerShell command. It’s the internal tool-call notation OpenClaw shows in docs / logs.

What to do instead:

A) Run the PTY test by asking the OpenClaw agent (in Discord/UI), not in PowerShell

In this Discord thread, send a message like:

“Run pwd via the exec tool with pty=true and paste the raw output.”

(or: “Use exec with pty=true: powershell -NoProfile -NonInteractive -Command "Write-Output 'PTY_OK'"”)

B) If you want to run something manually in PowerShell (no OpenClaw involved)

Just run the command directly, e.g.:

pwd
Get-ChildItem

#

Those “red errors” are PowerShell parser errors because you’re typing exec(command=..., pty=true) at a PS C:\...> prompt.

exec(...) is not a PowerShell function/command — it’s just the notation OpenClaw uses internally for the tool call.

Do this instead (copy/paste into this Discord thread to your OpenClaw bot/agent)

Send a normal message that instructs the agent to use the tool:

Use the exec tool with pty=true to run: pwd
Then do: powershell -NoProfile -NonInteractive -Command "Write-Output 'PTY_STDOUT_OK'"
Paste the raw outputs.

(You can also add: “don’t run it in PowerShell directly; call the exec tool”.)

Also paste these (run in your PowerShell terminal)

These help determine if you’re on native Windows vs WSL2 and what version you’re on:

#

Next do a clean “is output capture working at all?” check after the restart, and also confirm you’re not running under a weird custom Scheduled Task environment.

1) Confirm what’s actually running now

In PowerShell, run and paste the output:

openclaw --version
openclaw gateway status --deep
openclaw status

Key things we’re looking for in gateway status --deep: RPC probe ok, what it’s listening on, and whether it thinks it’s supervised by schtasks.

2) Tail logs while you test exec

Open a second PowerShell window and run:

heady drum Feb 19, 2026, 4:18 PM

#

PS C:\Windows\system32> openclaw --version

openclaw status
2026.2.15

🦞 OpenClaw 2026.2.15 (3fe22ea) — The only bot that stays out of your training set.

|
o
|
o
OpenClaw status

Overview
┌─────────────────┬───────────────────────────────────────────────────────────────────────────────────────────────────┐
│ Item │ Value │
├─────────────────┼───────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Dashboard │ http://127.0.0.1:18789/ │
│ OS │ windows 10.0.19045 (x64) · node 24.13.0 │
│ Tailscale │ off │
│ Channel │ stable (default) │
│ Update │ available · pnpm · npm update 2026.2.17 │
│ Gateway │ local · ws://127.0.0.1:18789 (local loopback) · reachable 82ms · auth token · DESKTOP-3LO9H7G │
│ │ (192.168.0.120) app unknown windows 10.0.19045 │
│ Gateway service │ Scheduled Task installed · registered · running │
│ Node service │ Scheduled Task not installed │
│ Agents │ 1 · 1 bootstrapping · sessions 49 · default main active 14m ago │
│ Memory │ 14 files · 18 chunks · sources memory · plugin memory-core · vector ready · fts ready · cache on │
│ │ (24) │

#

Probes │ skipped (use --deep) │
│ Events │ none │
│ Heartbeat │ 30m (main) │
│ Sessions │ 49 active · default gemini-2.5-flash (200k ctx) · ~.openclaw\agents\main\sessions\sessions.json │
└─────────────────┴───────────────────────────────────────────────────────────────────────────────────────────────────┘

Security audit
Summary: 1 critical · 2 warn · 1 info
CRITICAL Telegram group commands have no sender allowlist
Telegram group access is enabled but no sender allowlist is configured; this allows any group member to invoke /… commands (including skill commands).
Fix: Approve yourself via pairing (recommended), or set channels.telegram.groupAllowFrom (or per-group groups.<id>.allowFrom).
WARN Reverse proxy headers are not trusted
gateway.bind is loopback and gateway.trustedProxies is empty. If you expose the Control UI through a reverse proxy, configure trusted proxies so local-client c…
Fix: Set gateway.trustedProxies to your proxy IPs or keep the Control UI local-only.
WARN Some gateway.nodes.denyCommands entries are ineffective
gateway.nodes.denyCommands uses exact command-name matching only. - Unknown command names (not in defaults/allowCommands): camera.snap, camera.clip, screen.rec…
Fix: Use exact command names (for example: canvas.present, canvas.hide, canvas.navigate, canvas.eval, canvas.snapshot, canvas.a2ui.push, canvas.a2ui.pushJSONL, canvas.a2ui.reset). If you need broader restrictions, remove risky commands from allowCommands/default workflows.
Full report: openclaw security audit
Deep probe: openclaw security audit --deep

#

Channels
┌──────────┬─────────┬────────┬───────────────────────────────────────────────────────────────────────────────────────┐
│ Channel │ Enabled │ State │ Detail │

│ Telegram │ ON │ WARN │ token config (8376…7LnY · len 46) · accounts 1/1 · gateway: Config allows │
│ │ │ │ unmentioned group messages (requireMention=false). Telegram Bot API p… │
│ WhatsApp │ ON │ OK │ linked · +447359348936 · auth 1m ago · accounts 1 │

Sessions

│ Key │ Kind │ Age │ Model │ Tokens │
┤
│ agent:main:main │ direct │ 14m ago │ gemini-2.5-flash │ 74k/200k (37%) │
│ agent:main:cron:b4fe674c-6165-4… │ direct │ 18m ago │ gemini-2.5-flash │ 36k/200k (18%) │
│ agent:main:cron:b4fe674c-6165-4… │ direct │ 18m ago │ gemini-2.5-flash │ 36k/200k (18%) │
│ agent:main:cron:d2b651d8-fcb0-4… │ direct │ 2h ago │ gemini-2.5-flash │ 17k/200k (8%) │
│ agent:main:cron:d2b651d8-fcb0-4… │ direct │ 2h ago │ gemini-2.5-flash │ 17k/200k (8%) │
│ agent:main:cron:b298bbce-ecf5-4… │ direct │ 2h ago │ gemini-2.5-flash │ 15k/200k (7%) │
│ agent:main:cron:b298bbce-ecf5-4… │ direct │ 2h ago │ gemini-2.5-flash │ 15k/200k (7%) │
│ agent:main:cron:b298bbce-ecf5-4… │ direct │ 3h ago │ gemini-2.5-flash │ 28k/200k (14%) │
│ agent:main:cron:d2b651d8-fcb0-4… │ direct │ 4h ago │ gemini-2.5-flash │ 31k/200k (16%) │
│ agent:main:cron:b298bbce-ecf5-4… │ direct │ 4h ago │ gemini-2.5-flash │ 29k/200k (14%) │
└

misty ginkgoBOT Feb 19, 2026, 4:20 PM

#

exec(command='powershell -NoProfile -NonInteractive -Command "Write-Output ''POWERSHELL_OK''"')
exec(command='powershell -NoProfile -NonInteractive -Command "[Console]::Out.WriteLine(''CONSOLE_OK'')"')

If all 3 come back (no output), then output capture is broken at the process/host level (often service/scheduled-task/headless weirdness), not “file searching”.

3) If “PowerShell aliases” don’t print, force string output

PowerShell commands like ls, pwd, Get-ChildItem often output objects, and in some hosting modes they don’t get rendered as text.

Use one of these patterns instead:

exec(command='powershell -NoProfile -NonInteractive -Command "(Get-Location).Path | Write-Output"')

exec(command='powershell -NoProfile -NonInteractive -Command "Get-ChildItem -LiteralPath ''C:\\'' -Force | Select-Object -First 20 -ExpandProperty FullName"')

heady drum Feb 19, 2026, 4:20 PM

#

PS C:\Windows\system32> exec(command='echo STDOUT_OK')
command=echo STDOUT_OK : The term 'command=echo STDOUT_OK' is not recognized as the name of a cmdlet, function, script
file, or operable program. Check the spelling of the name, or if a path was included, verify that the path is correct
and try again.
At line:1 char:6

exec(command='echo STDOUT_OK')
```
 ~~~~~~~~~~~~~~~~~~~~~~~~
```
- CategoryInfo : ObjectNotFound: (command=echo STDOUT_OK:String) [], CommandNotFoundException
- FullyQualifiedErrorId : CommandNotFoundException

PS C:\Windows\system32>

misty ginkgoBOT Feb 19, 2026, 4:21 PM

#

…but that only proves PowerShell works, not that OpenClaw is capturing stdout.

C) Next debugging step (copy/paste into Discord to the bot)

Please ask the bot to run these and paste what it returns:

echo STDOUT_OK
powershell -NoProfile -NonInteractive -Command "Write-Output 'POWERSHELL_OK'"
Same as (2) but with pty=true

Also paste:

openclaw --version
whether you’re running native Windows or WSL2

Once I see those outputs, I can tell if you’re hitting “stdout capture broken” vs “wrong shell (cmd on node)” vs “needs PTY”.

#

openclaw --version
openclaw status
openclaw gateway status

Once you paste the bot’s exec outputs + the version/status, I can tell if:

the agent is actually executing via a Windows node (often cmd.exe), or
stdout capture is broken due to service/scheduled-task/headless execution, or
PTY fixes it.

heady drum Feb 19, 2026, 4:22 PM

#

PS C:\Windows\system32> exec(command='powershell -NoProfile -NonInteractive -Command "Write-Output ''POWERSHELL_OK''"')

exec(command='powershell -NoProfile -NonInteractive -Command "[Console]::Out.WriteLine(''CONSOLE_OK'')"')

command=powershell -NoProfile -NonInteractive -Command "Write-Output 'POWERSHELL_OK'" : The term 'command=powershell
-NoProfile -NonInteractive -Command "Write-Output 'POWERSHELL_OK'"' is not recognized as the name of a cmdlet,
function, script file, or operable program. Check the spelling of the name, or if a path was included, verify that the
path is correct and try again.
At line:1 char:6

exec(command='powershell -NoProfile -NonInteractive -Command "Write-O ...
```
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
```
- CategoryInfo : ObjectNotFound: (command=powersh...POWERSHELL_OK'":String) [], CommandNotFoundException
- FullyQualifiedErrorId : CommandNotFoundException

command=powershell -NoProfile -NonInteractive -Command "[Console]::Out.WriteLine('CONSOLE_OK')" : The term
'command=powershell -NoProfile -NonInteractive -Command "[Console]::Out.WriteLine('CONSOLE_OK')"' is not recognized as
the name of a cmdlet, function, script file, or operable program. Check the spelling of the name, or if a path was
included, verify that the path is correct and try again.
At line:2 char:6

exec(command='powershell -NoProfile -NonInteractive -Command "[Consol ...
```
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
```
- CategoryInfo : ObjectNotFound: (command=powersh...('CONSOLE_OK')":String) [], CommandNotFoundException
- FullyQualifiedErrorId : CommandNotFoundException

PS C:\Windows\system32>

#

PS C:\Windows\system32> openclaw gateway status --deep

🦞 OpenClaw 2026.2.15 (3fe22ea) — curl for conversations.

|
o
Service: Scheduled Task (registered)
File logs: \tmp\openclaw\openclaw-2026-02-19.log
Command: C:\Program Files\nodejs\node.exe C:\Users\User\AppData\Roaming\npm\node_modules\openclaw\dist\index.js gateway --port 18789
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.

Runtime: running (last run time 19/02/2026 15:56:50)
RPC probe: ok

Listening: 127.0.0.1:18789
Other gateway-like services detected (best effort):

\OpenClaw Gateway (system, task: \OpenClaw Gateway, run: C:\Users\User.openclaw\gateway.cmd)
\Uranus UI (system, task: \Uranus UI, run: powershell -NoProfile -ExecutionPolicy Bypass -File %USERPROFILE%.openclaw\workspace\uranus-ui\start-uranus-ui.ps1)
Cleanup hint: schtasks /Delete /TN "OpenClaw Gateway" /F

Recommendation: run a single gateway per machine for most setups. One gateway supports multiple agents (see docs: /gateway#multiple-gateways-same-host).
If you need multiple gateways (e.g., a rescue bot on the same host), isolate ports + config/state (see docs: /gateway#multiple-gateways-same-host).

Troubles: run openclaw status
Troubleshooting: https://docs.openclaw.ai/troubleshooting
PS C:\Windows\system32>

misty ginkgoBOT Feb 19, 2026, 4:23 PM

#

openclaw logs --follow

Leave that running.

3) Test exec from Discord (not from PowerShell)

In this Discord thread, send your bot a message like:

Use exec to run: echo STDOUT_OK
Then: powershell -NoProfile -NonInteractive -Command "Write-Output 'POWERSHELL_OK'"

If those still come back (no output), the logs you’re tailing should show why (spawn error, permission, wrong host, etc.).

4) Stop using a custom `schtasks` task (use OpenClaw’s installer instead)

If you manually created the Scheduled Task yourself, it can easily end up in a headless/non-interactive mode that breaks stdio capture.

heady drum Feb 19, 2026, 4:23 PM

#

PS C:\Windows\system32> openclaw gateway status --deep

🦞 OpenClaw 2026.2.15 (3fe22ea) — curl for conversations.

|
o
Service: Scheduled Task (registered)
File logs: \tmp\openclaw\openclaw-2026-02-19.log
Command: C:\Program Files\nodejs\node.exe C:\Users\User\AppData\Roaming\npm\node_modules\openclaw\dist\index.js gateway --port 18789
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.

Runtime: running (last run time 19/02/2026 15:56:50)
RPC probe: ok

Listening: 127.0.0.1:18789
Other gateway-like services detected (best effort):

\OpenClaw Gateway (system, task: \OpenClaw Gateway, run: C:\Users\User.openclaw\gateway.cmd)
\Uranus UI (system, task: \Uranus UI, run: powershell -NoProfile -ExecutionPolicy Bypass -File %USERPROFILE%.openclaw\workspace\uranus-ui\start-uranus-ui.ps1)
Cleanup hint: schtasks /Delete /TN "OpenClaw Gateway" /F

Recommendation: run a single gateway per machine for most setups. One gateway supports multiple agents (see docs: /gateway#multiple-gateways-same-host).
If you need multiple gateways (e.g., a rescue bot on the same host), isolate ports + config/state (see docs: /gateway#multiple-gateways-same-host).

Troubles: run openclaw status
Troubleshooting: https://docs.openclaw.ai/troubleshooting
PS C:\Windows\system32>

#

i havenot got my bot on discord only telegram so cant send a message from here. with all that info how do i fix this please

misty ginkgoBOT Feb 19, 2026, 4:28 PM

#

4) Try PTY mode (sometimes the simplest workaround on Windows)

exec(command='powershell -NoProfile -NonInteractive -Command "Write-Output ''PTY_OK''"', pty=true)

5) If you’re on native Windows: WSL2 is still the “known-good” path

Docs: https://docs.openclaw.ai/platforms/windows

If you paste the outputs from steps (1) + (2) (even if they’re empty), I can tell you immediately whether this is:

“you’re actually executing via node/cmd.exe” vs
“PowerShell object output not being stringified” vs
“stdout pipes broken in your launch mode (service/task)”

#

Once we confirm the gateway is healthy, the “known-good” way is:

openclaw gateway install --force
openclaw gateway restart

(And disable your custom task in Task Scheduler so you don’t end up with two gateways fighting over the port.)

Paste the outputs from steps 1 + what the bot returns in step 3, and we can zero in on the exact fix.

heady drum Feb 19, 2026, 4:32 PM

#

PS C:\Windows\system32> exec(command='powershell -NoProfile -NonInteractive -Command "Write-Output ''PTY_OK''"', pty=true)
At line:1 char:88

... oProfile -NonInteractive -Command "Write-Output ''PTY_OK''"', pty=tru ...

Missing argument in parameter list.
+ CategoryInfo : ParserError: (:) [], ParentContainsErrorRecordException
+ FullyQualifiedErrorId : MissingArgument

PS C:\Windows\system32>

#

my openclaw is able to use read if it knows the file location but not the exec tool still. i have sent everyhting you asked

heady drum Feb 19, 2026, 4:49 PM

#

@misty ginkgo

misty ginkgoBOT Feb 19, 2026, 4:50 PM

#

I see the ping — but I need one of these to move forward:

Option A (recommended): update to fix the Windows Scheduled Task “no exec output” bug

In PowerShell:

openclaw gateway stop
npm i -g openclaw@latest
# (close + reopen PowerShell)
openclaw --version
openclaw gateway install --force
openclaw gateway restart
openclaw gateway status --deep

Paste the output of:

openclaw --version

#

openclaw gateway status --deep

Option B: if your version stubbornly stays on 2026.2.15

Paste:

where.exe openclaw
Get-Command openclaw | Format-List *

Also: don’t type exec(command=...) into PowerShell — to test exec, ask the bot here in Discord after updating, e.g. “Run echo STDOUT_OK via exec and paste the output.”

heady drum Feb 19, 2026, 4:50 PM

#

also tried to fix using gemini 3 pro and we ran some tests. gemini suggests this The Scientific Post-Mortem
Since session_status() works but exec fails across all shells, we can isolate the failure to the Gateway's Subprocess Handler.

The Internal vs. External Divide: Internal functions stay within the AI's memory space. Exec, however, must "spawn" a child process. If the environment where OpenClaw lives (the "Gateway") doesn't have the permissions to monitor child processes, it will simply receive a null response every time.

The "Zombie" Process Hypothesis: It is possible the shells are actually starting, but the Gateway is failing to "attach" to their output streams (STDOUT/STDERR). The command runs in a void, the file is never written because the shell environment is uninitialized, and the AI hears nothing.

#

if i do gateway install will that loose my data?

misty ginkgoBOT Feb 19, 2026, 4:52 PM

#

No — openclaw gateway install (even with --force) should not wipe your data.

What it does is basically (re)create the supervisor entry (on Windows: the Scheduled Task) so the Gateway starts reliably and points at the current OpenClaw install + config path.

What it does not do:

It does not delete your config (~\.openclaw\openclaw.json)
It does not delete your workspace (~\.openclaw\workspace\...)
It does not delete sessions/memory (~\.openclaw\agents\...\sessions\..., memory files, etc.)
It does not remove credentials (those live under your OpenClaw state dir)

That said, if you want maximum peace of mind, you can quickly back up the important folder before changing anything:

# quick “copy backup” (creates a timestamped backup)
Copy-Item "$env:USERPROFILE\.openclaw" "$env:USERPROFILE\.openclaw.backup-$(Get-Date -Format yyyyMMdd-HHmmss)" -Recurse

#

Then it’s safe to run:

openclaw gateway install --force
openclaw gateway restart

If you tell me whether you installed via npm vs git/source, I can suggest the cleanest update path too.

#

What Gemini wrote is basically pointing at the right class of problem (“child process started but Gateway can’t capture stdout/stderr”), but it’s not really a permissions thing — it’s almost always a Windows + Scheduled Task + stdio wiring issue.

The key detail from your logs:

You’re on OpenClaw 2026.2.15
Gateway is running as Scheduled Task
RPC probe: ok (gateway is fine), but exec output is empty

There’s a specific fix in 2026.2.17:

Process/Windows: disable detached spawn in exec runs to prevent empty command output.

So the next step isn’t more shell experiments — it’s get the Gateway onto 2026.2.17+ and rewrite the Scheduled Task.

Do this on the Windows machine (PowerShell)

Stop the gateway:

#

openclaw gateway stop

Update OpenClaw:

npm i -g openclaw@latest

Close/reopen PowerShell, verify:

openclaw --version

You want 2026.2.17 or newer.

Reinstall the Scheduled Task so it points at the new install:

#

openclaw gateway install --force
openclaw gateway restart

Confirm:

openclaw gateway status --deep

Then test `exec` the right way

Don’t type exec(command=...) into PowerShell (that will never work). In Discord, ask the bot:

“Run exec: echo STDOUT_OK”
“Run exec: pwd”

If `openclaw --version` stubbornly stays at 2026.2.15

#

Paste:

where.exe openclaw
Get-Command openclaw | Format-List *

That usually reveals you’ve got two installs and you updated the wrong one.

If you get to 2026.2.17+ and exec is still blank, paste the relevant snippet from:

openclaw logs --follow

right as you trigger an exec attempt, and we’ll go one level deeper.

heady drum Feb 19, 2026, 5:20 PM

#

PS C:\Windows\system32> openclaw gateway stop

npm i -g openclaw@latest

🦞 OpenClaw 2026.2.15 (3fe22ea) — Less middlemen, more messages.

Stopped Scheduled Task: OpenClaw Gateway
npm warn deprecated npmlog@6.0.2: This package is no longer supported.
npm warn deprecated are-we-there-yet@3.0.1: This package is no longer supported.
npm warn deprecated gauge@4.0.4: This package is no longer supported.
npm warn deprecated tar@6.2.1: Old versions of tar are not supported, and contain widely publicized security vulnerabilities, which have been fixed in the current version. Please update. Support for old versions may be purchased (at exorbitant rates) by contacting i@izs.me
npm warn deprecated node-domexception@1.0.0: Use your platform's native DOMException instead
npm warn deprecated glob@10.5.0: Old versions of glob are not supported, and contain widely publicized security vulnerabilities, which have been fixed in the current version. Please update. Support for old versions may be purchased (at exorbitant rates) by contacting i@izs.me

added 5 packages, removed 7 packages, and changed 667 packages in 10m

124 packages are looking for funding
run npm fund for details
PS C:\Windows\system32>

misty ginkgoBOT Feb 19, 2026, 5:20 PM

#

Nice — that npm i -g openclaw@latest run looks like it actually updated a ton of stuff.

Now we need to confirm you’re really on 2026.2.17+ and then rewire the Scheduled Task (otherwise it can keep pointing at the old install).

1) Confirm the CLI version in this PowerShell window

openclaw --version

If it shows 2026.2.17 or newer → continue to step 2.
If it still shows 2026.2.15, paste:
```
where.exe openclaw
npm ls -g openclaw --depth=0
```
(That tells us if you have multiple installs / PATH is hitting an older shim.)

#

2) Reinstall the Gateway Scheduled Task to point at the new version

openclaw doctor
openclaw gateway install --force
openclaw gateway restart
openclaw gateway status --deep

In that gateway status --deep output, the banner at the top should also show 2026.2.17+.

3) Re-test `exec` (from Discord, by asking the bot)

Once the gateway is back up, ask your bot to run:

echo STDOUT_OK
pwd

(Still don’t type exec(command=...) into PowerShell — that syntax only exists inside the bot/tool system.)

#

If you paste openclaw --version after the npm update, I can tell you immediately whether you’re actually on the fixed build.

heady drum Feb 19, 2026, 5:22 PM

#

PS C:\Windows\system32> openclaw gateway status --deep

🦞 OpenClaw 2026.2.17 (4134875)
If it's repetitive, I'll automate it; if it's hard, I'll bring jokes and a rollback plan.

|
o
Service: Scheduled Task (registered)
File logs: \tmp\openclaw\openclaw-2026-02-19.log
Command: C:\Program Files\nodejs\node.exe C:\Users\User\AppData\Roaming\npm\node_modules\openclaw\dist\index.js gateway --port 18789
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.

Runtime: running (last run time 19/02/2026 17:21:15)
RPC probe: ok

Listening: 127.0.0.1:18789
Other gateway-like services detected (best effort):

\OpenClaw Gateway (system, task: \OpenClaw Gateway, run: C:\Users\User.openclaw\gateway.cmd)
\Uranus UI (system, task: \Uranus UI, run: powershell -NoProfile -ExecutionPolicy Bypass -File %USERPROFILE%.openclaw\workspace\uranus-ui\start-uranus-ui.ps1)
Cleanup hint: schtasks /Delete /TN "OpenClaw Gateway" /F

Recommendation: run a single gateway per machine for most setups. One gateway supports multiple agents (see docs: /gateway#multiple-gateways-same-host).
If you need multiple gateways (e.g., a rescue bot on the same host), isolate ports + config/state (see docs: /gateway#multiple-gateways-same-host).

Troubles: run openclaw status
Troubleshooting: https://docs.openclaw.ai/troubleshooting
PS C:\Windows\system32>

misty ginkgoBOT Feb 19, 2026, 5:23 PM

#

Perfect — you’re now on 2026.2.17. That’s the version that includes the Windows fix for “empty exec output”.

Now the only thing left is to confirm the Scheduled Task was rewritten (so it’s actually running the 2026.2.17 bits) and then re-test exec from the bot.

1) Reinstall the gateway task (quick + safe)

Even though it’s “running”, do this once to ensure the task entrypoint isn’t stale:

openclaw gateway install --force
openclaw gateway restart

2) Re-test exec (from Discord, by asking the bot/agent)

In this Discord thread, ask the bot:

“Use exec to run: echo STDOUT_OK”
“Use exec to run: pwd”

#

“Use exec to run: dir” (Windows-native)

You should start seeing real output now.

3) If it’s still “(no output)”

Then we need logs around the exec attempt:

In PowerShell:

openclaw logs --follow

Trigger one exec from Discord.
Paste the few log lines that appear at that moment.

Also: your gateway status --deep still shows the service Command path in a weird wrapped way (“C:\Program Files … odejs … ode.exe …”). That’s normal line wrapping, but if we still have trouble, we’ll inspect the actual Scheduled Task action.

#

Send the Discord exec test results and we’ll finish the loop.

heady drum Feb 19, 2026, 5:25 PM

#

its working and can use exec tools now apparently. should i still do step 2 with doctor and all that stuff. looks like i did some of that already unless im confused

misty ginkgoBOT Feb 19, 2026, 5:25 PM

#

If exec is working now, you’re basically done. You don’t have to run anything else.

That said, there are two “nice to do” checks that are safe and help prevent weirdness later:

1) Optional but recommended: run Doctor once after an update

This won’t wipe data — it’s a repair/migration checker.

openclaw doctor

If it suggests a fix (like reinstalling the service), it’ll tell you.

2) Only if you notice the gateway “sometimes” starts the wrong version

Then do the full service re-wire:

#

openclaw gateway install --force
openclaw gateway restart

But if you already upgraded and gateway status --deep shows 2026.2.17 and you’re stable, you can skip it.

Also: your earlier Security Audit CRITICAL (Telegram)

Unrelated to exec, but worth fixing when you get a moment: set a sender allowlist for Telegram groups (or pair/approve yourself) so not everyone in a group can run commands.

If this fixed it, please mark the solution message in the thread (right‑click the solution → Apps → ✅ Mark Solution). It helps Answer Overflow index it for the next Windows person who hits “(no output)”.

heady drum Feb 19, 2026, 5:28 PM

#

i only have it in a group with me on telegram so should be okay. finally though best practice for giving permenant access to openclaw to have 3 gmail accounts to mnitor and send emails form

misty ginkgoBOT Feb 19, 2026, 5:31 PM

#

Best practice: don’t give OpenClaw “god-mode” access to your personal Gmail(s). Treat email like prod infra: least privilege, separation, and auditable flows.

Recommended setup (practical + safe)

1) Use 1 “hub” inbox for monitoring

Create (or pick) one dedicated Gmail like openclaw.hub@...
In each of your 3 Gmail accounts, add a filter → forward (or auto-label + forward) important mail to the hub inbox.
Then wire only the hub to OpenClaw’s Gmail watcher (Pub/Sub → hooks). This keeps the “always-on monitoring” simple and stable.

Docs: https://docs.openclaw.ai/automation/gmail-pubsub

2) Keep “sending” separate from “monitoring”
If you truly need to send as 3 different accounts, you have two main options:

Option A (simplest operationally): send everything from the hub inbox, and use Gmail “Send mail as” / aliases if that fits your use case.
Option B (most correct, more setup): authorize all 3 accounts with gog and send with an explicit account selector (so the tool knows exactly which mailbox it’s using). That way you don’t rely on spoofing From: headers.

3) Put guardrails on email actions

#

Even if it’s “only you” in Telegram, still good hygiene:

Keep hooks.gmail.allowUnsafeExternalContent off (default) so emails are treated as untrusted content.
Consider includeBody: false or a small maxBytes so you don’t feed huge raw emails into the model.
In your agent instructions: “Draft first, ask for confirmation before sending to new recipients.”

About “3 inboxes fully monitored by OpenClaw”

Out of the box, hooks.gmail in config is one account (single auto-start watcher). If you want truly independent watches for 3 inboxes, it’s doable but becomes “advanced mode” (multiple watcher daemons / multiple profiles/gateways / custom hook routing). If that’s what you want, tell me which direction you prefer and I’ll outline the cleanest wiring.

If the exec issue is solved now, you’re in a great spot to right‑click the solution message → Apps → ✅ Mark Solution so it’s searchable later.

heady drum Feb 19, 2026, 5:38 PM

#

PS C:\Windows\system32> openclaw doctor

agent:main:main (3m ago)
agent:main:cron:b4fe674c-6165-4c74-8dc9-ec967b5dc056 (87m ago)
agent:main:cron:b4fe674c-6165-4c74-8dc9-ec967b5dc056:run:6f8daa4f-354d-4a8f-bf74-479b48106feb (87m ago)
agent:main:cron:d2b651d8-fcb0-4d2c-ab5f-26c2c056cc4e (201m ago)
agent:main:cron:d2b651d8-fcb0-4d2c-ab5f-26c2c056cc4e:run:a22f3753-7691-43c7-a910-45fec6b0ac86 (201m ago
| - telegram default: Config allows unmentioned group messages (requireMention=false). |
| Telegram Bot API privacy mode will block most group messages unless disabled. (In |
| BotFather run /setprivacy → Disable for this bot (then restart the gateway).) |
| |

Run "openclaw doctor --fix" to apply changes.
|
— Doctor complete.

PS C:\Windows\system32>

misty ginkgoBOT Feb 19, 2026, 5:40 PM

#

Doctor output looks good — nothing there suggests you’re “broken” anymore.

A few notes on what you’re seeing:

The “Update” section

“This install is not a git checkout…”

That’s normal for an npm install. Doctor is basically saying “use npm/pnpm to update”. You already did that (you’re on 2026.2.17 now), so you can ignore that part.

The only real actionable warning: Telegram privacy mode

Doctor is warning:

you’ve configured Telegram to allow unmentioned group messages (requireMention=false)
but Telegram “privacy mode” will prevent the bot from receiving most group messages unless disabled.

If you want the bot to see all messages in your Telegram group without mentioning it:

In BotFather: /setprivacy → Disable for your bot

#

Then restart the gateway:

openclaw gateway restart

If you’re happy to always mention/reply to the bot in the group, you can leave privacy mode on and just ignore the warning (or set requireMention=true to match that behavior).

“Skills missing requirements: 43”

Totally fine — most skills are optional and require extra binaries/accounts. Only worry about it if you want a specific skill. You can inspect what’s missing with:

openclaw skills check
openclaw skills info <skillName>

Do you need `openclaw doctor --fix`?

Only if you want it to apply the suggested tweaks. If everything is working (and especially since exec is fixed now), it’s optional.

#

Glad you got exec working — if you can, mark the solution in the thread so it’s indexed for the next Windows person.

heady drum Feb 19, 2026, 5:41 PM

#

i tried couldnt see mark button

#OpenClaw `exec` tool on Windows: PowerShell commands producing no output

0) First: what OpenClaw version + how is the Gateway started?

1) Quick “is stdout/stderr wired up?” probes

2) Try forcing a PTY (often fixes “no output” on Windows)

3) If you’re running via Scheduled Task/service: test in foreground

4) Strong recommendation: WSL2 if you can

1) Are you running exec on the Windows Gateway or on a Windows node?

2) Run these “stdout sanity” tests (they should always print)

A) Run the PTY test by asking the OpenClaw agent (in Discord/UI), not in PowerShell

B) If you want to run something manually in PowerShell (no OpenClaw involved)

Do this instead (copy/paste into this Discord thread to your OpenClaw bot/agent)

Also paste these (run in your PowerShell terminal)

1) Confirm what’s actually running now

2) Tail logs while you test exec

3) If “PowerShell aliases” don’t print, force string output

C) Next debugging step (copy/paste into Discord to the bot)

3) Test exec from Discord (not from PowerShell)

4) Stop using a custom schtasks task (use OpenClaw’s installer instead)

4) Try PTY mode (sometimes the simplest workaround on Windows)

5) If you’re on native Windows: WSL2 is still the “known-good” path

Option A (recommended): update to fix the Windows Scheduled Task “no exec output” bug

Option B: if your version stubbornly stays on 2026.2.15

Do this on the Windows machine (PowerShell)

Then test exec the right way

If openclaw --version stubbornly stays at 2026.2.15

1) Confirm the CLI version in this PowerShell window

2) Reinstall the Gateway Scheduled Task to point at the new version

3) Re-test exec (from Discord, by asking the bot)

1) Reinstall the gateway task (quick + safe)

2) Re-test exec (from Discord, by asking the bot/agent)

3) If it’s still “(no output)”

1) Optional but recommended: run Doctor once after an update

2) Only if you notice the gateway “sometimes” starts the wrong version

Also: your earlier Security Audit CRITICAL (Telegram)

Recommended setup (practical + safe)

About “3 inboxes fully monitored by OpenClaw”

The “Update” section

The only real actionable warning: Telegram privacy mode

“Skills missing requirements: 43”

Do you need openclaw doctor --fix?

1) Are you running `exec` on the Windows Gateway or on a Windows node?

4) Stop using a custom `schtasks` task (use OpenClaw’s installer instead)

Then test `exec` the right way

If `openclaw --version` stubbornly stays at 2026.2.15

3) Re-test `exec` (from Discord, by asking the bot)

Do you need `openclaw doctor --fix`?