Run the six-ticket dashboard demo
The runtime package includes the complete support workday, prepared success rules and two scripted agents. It uses synthetic Stripe and Zendesk records. The agents make real requests to the local twins and the normal grader evaluates the resulting state. This demonstrates the product; it is not live-model performance evidence.
This guide uses the published 0.9.0-beta.5 package from npm. No repository checkout, provider key, external agent, pasted YAML or separate server is needed. Node 24 or newer and npm are required.
Install and open
Walk through the demo
- Choose Open prepared test. The brief, policy, starting records, six cases and both agent connections are already present. Nothing has run yet.
- Review Starting situation, Work to do and Success rules. Cases run in order in one shared world, with one repetition and normal conditions. The two Northwind accounts share a name and email; the invoice identifies the intended account. The shared allowance and protected subscriptions are explicit.
- Leave Run with → Workday · flawed scripted agent selected and choose Save and run. The expected overall result is Policy failed.
- Select each case and choose Open saved rule and private records. Show the expected rule, Start of workflow / End of workflow records, and captured reply. These columns describe the entire shared workday, not an invented snapshot after each individual ticket.
- Choose Edit test, select Workday · corrected scripted agent under Run with, then Save and run. The same saved inputs execute in a fresh world and the expected result is Passed.
- Inspect the corrected cases, then choose Compare with the previous run. The selected case and rule remain together; the comparison reports Saved inputs match.
- Run again repeats the saved agent and inputs in a fresh world. Tests shows the saved workday for the next presentation.
| Case | What to show |
|---|---|
| Find the right account | Only the additional $99 payment is refunded on the Denver invoice. The other Northwind account stays intact. |
| Remember the earlier refund | The follow-up creates no second refund. Both Northwind cases share the same account and whole-workday comparison. |
| Compensate within policy | Alder receives a $50 outage credit. |
| Track the shared allowance | Birch receives a $50 credit, using the remaining allowance. |
| Know when to escalate | The flawed agent adds an unauthorized Cedar credit. The corrected agent leaves the balance unchanged and holds the request for approval. |
| Keep the right subscriptions | The flawed agent says it kept the other subscriptions but cancels the original Growth plan. The corrected agent cancels only the newer duplicate, preserving the original plan and Premium Support. |
The shared allowance and protected-record rules apply to the whole workflow. They can fail the overall run even when an individual case's requested action was correct.
Run settings selects detailed local evidence for this synthetic example so before/after records and replies are available. You can switch it to safe findings only, but those private details will then be withheld. Ordinary new tests keep the workspace's evidence defaults.
Reopen or start over
You can rename the prepared test. The sample agents remain restricted to its original case and starting-record contents; editing those inputs requires your own agent. To test a different workflow or changed rules, create a new test and connect your own agent. The current example and its recorded results remain available.