See how one tool proposal changes across three modes.
Compare the same fixed proposal in Observe, Audit, and Enforce. In Audit and Enforce, an enabled, scoped policy evaluates the sample rule. No Staso account is needed.
Redacted example trace. No account required.
Follow one agent run from failure to policy decision.
Replay a redacted coding run across Observe, Audit, and Enforce.
Using LiteLLM? Explore a separate fixed Python tool-call example.
See the LiteLLM demoProduction command control
Fix the pricing tests and verify the patch.
Proposed tool call
run_shell(command="rm -rf /tmp/build-cache/*")- Environment
- production
- Policy
- Command safety (enabled)
- Configured rule
- Block destructive shell commands on production hosts
- Rule type
- LLM judge
- Policy scope
- Production · run_shell
The captured debugging path stays visible.
The trace links captured model turns, file edits, a failed test, and the proposed shell command.
- Trace
- Captured
- Tool proposal
- Found
- Tool execution
- Not confirmed
The policy flags the command without stopping it.
The enabled policy records this rule match as would-block. The call remains allowed.
- Mode
- Audit
- Rule match
- Would block
- Final action
- Allowed
The same match stops before dispatch.
The same rule match under an enabled Enforce policy blocks this supported proposal before tool execution.
- Mode
- Enforce
- Rule match
- Block
- Final action
- Blocked before execution
Now replaying
Reading the request
Trace spans
Selected span
{
"request": "Fix the pricing tests and verify the patch.",
"environment": "production"
}Ready to see your own trace?
Choose the integration you use now. Staso will return you to that setup after sign-in.