Prevent sending environment variables
✅Solved
I am using a Skill to access a server via a script. Credentials are currently stored in .openclaw/.env. However, the session logs indicate that OpenClaw is exposing the .env file to the model. How can I prevent this behavior? Is it possibly caused by the Skill description:"
Solution
That skill description is not what “exposes” your
What’s usually happening is one of these:
1) The model used a tool to read it (e.g.
2) Your script printed secrets (directly or indirectly) and that output got logged.
3) You’re looking at logs that show “dotenv loaded from …” (path), which is not the file contents being sent to the LLM.
OpenClaw does load env vars from
How to prevent the model from seeing
This depends on what tools you’re allowing.
Security baseline shows the pattern (deny runtime/fs tools, workspace-only FS):
~/.openclaw/.envSKILL.mdWhat’s usually happening is one of these:
1) The model used a tool to read it (e.g.
read ~/.openclaw/.envcat ~/.openclaw/.envprintenv2) Your script printed secrets (directly or indirectly) and that output got logged.
3) You’re looking at logs that show “dotenv loaded from …” (path), which is not the file contents being sent to the LLM.
OpenClaw does load env vars from
~/.openclaw/.envHow to prevent the model from seeing .env
contents
This depends on what tools you’re allowing..envA) If you don’t need shell/file tools: deny them
If the agent can’t run commands or read files, it can’t “leak” env vars /.envSecurity baseline shows the pattern (deny runtime/fs tools, workspace-only FS):
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
134,426Members
Resources
Recent Announcements
Similar Threads
Was this page helpful?
