Block specific commands from being executed by Ona Agent. Use deny lists to prevent dangerous operations, enforce security policies, and maintain compliance. This is part of Guardrails.Documentation Index
Fetch the complete documentation index at: https://ona.com/docs/llms.txt
Use this file to discover all available pages before exploring further.

How it works
- User provides input to Ona Agent
- Agent decides to execute a command
- System checks command against deny list
- Command is executed (if allowed) or blocked with error message
Pattern matching
| Pattern | Effect |
|---|---|
shutdown | Blocks exactly “shutdown” |
shutdown* | Blocks “shutdown”, “shutdown -h”, “shutdown now” |
rm * | Blocks all rm commands with arguments |
Slash commands (
/clear, /support-bundle) are not blocked by deny lists. They are converted to prompts before reaching the agent. Bash commands (prefixed with !) are subject to deny list filtering.Configuration
Go to Settings → Agents → Policies. Only administrators can access.- Add patterns to Command Deny List (one per line)
- Save changes
Example patterns
Effect on users
When blocked, users see:- Manual commands unaffected: Users can still run commands directly in terminal
- Agent only: Only Ona Agent execution is restricted
- No retries: Agent is instructed not to retry blocked commands
Security considerations
Protects against:- Accidental destructive commands
- Malicious prompt injection
- Compliance violations
- Resource abuse
- Direct user commands in terminal
- Application-level actions
- Slash commands (cannot be blocked via deny lists)
Best practices
- Start with broad patterns (
aws *instead of listing variants) - Test in a non-production environment first
- Document why patterns were added
- Review and update periodically
Testing
- Create a new environment
- Ask Ona Agent to run a blocked command
- Verify the error message appears