Not working
Hi everyone, I need help with my OpenClaw setup on a Windows server.
Here is what I already did:
- Installed OpenClaw successfully
- Gateway starts and health shows OK
- Node host is running
- Telegram bot is connected and shows OK
- Dashboard opens on localhost
- Workspace is set to my project folder:
C:\Users\Administrator\projects\automation
- I cloned a real repo into that workspace (
- I changed tools profile to
- I fixed the AGENTS.md loop so it no longer blocks onboarding
-
- runtime: direct
- mode: off
- elevated allowedByConfig: true
What is still not working:
- The agent still says the
- It cannot run commands like
- It cannot inspect the local repo by itself
- It behaves like a chat/workspace assistant, not like a real coding/execution agent
- Sometimes the dashboard also gets stuck with auth/rate-limit issues (
What I want:
I want the OpenClaw agent to work as a real coding agent on the server:
- inspect my project files
- run shell commands
- edit files
- help me code inside the actual workspace
- basically “vibe code” locally on the server
Main issue:
Everything looks connected (gateway, node, telegram, workspace), but the agent still says it cannot execute shell commands and asks me to run them manually.
My question:
What exact config or component is still missing to make the agent use exec/process/file tools on the host machine?
Is this a tool-policy issue, an exec host issue, a permissions issue, or something else?
Thanks.
Here is what I already did:
- Installed OpenClaw successfully
- Gateway starts and health shows OK
- Node host is running
- Telegram bot is connected and shows OK
- Dashboard opens on localhost
- Workspace is set to my project folder:
C:\Users\Administrator\projects\automation
- I cloned a real repo into that workspace (
acpx- I changed tools profile to
coding- I fixed the AGENTS.md loop so it no longer blocks onboarding
-
openclaw sandbox explain- runtime: direct
- mode: off
- elevated allowedByConfig: true
What is still not working:
- The agent still says the
exec- It cannot run commands like
dir- It cannot inspect the local repo by itself
- It behaves like a chat/workspace assistant, not like a real coding/execution agent
- Sometimes the dashboard also gets stuck with auth/rate-limit issues (
too many failed authentication attemptstoken mismatchWhat I want:
I want the OpenClaw agent to work as a real coding agent on the server:
- inspect my project files
- run shell commands
- edit files
- help me code inside the actual workspace
- basically “vibe code” locally on the server
Main issue:
Everything looks connected (gateway, node, telegram, workspace), but the agent still says it cannot execute shell commands and asks me to run them manually.
My question:
What exact config or component is still missing to make the agent use exec/process/file tools on the host machine?
Is this a tool-policy issue, an exec host issue, a permissions issue, or something else?
Thanks.
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
134,827Members
Resources
Similar Threads
Was this page helpful?
