Support Ticket Triage¶
Use Codex to build a support-ticket triage flow in Maestro Flow, then improve it twice from real review decisions and trace feedback.
Overview¶
In this workshop you build a support-ticket triage automation in UiPath Maestro Flow, with Codex doing the building. A ticket arrives as flow input. An inline agent classifies it and drafts a reply grounded in the HomeFix knowledge base. A human reviews the draft in Action Center and approves, edits, or rejects it. You then feed your review decisions and trace feedback back to Codex, and it builds a second and third version that route urgent tickets, drop noise, and finally auto-resolve the safe cases.
Three things work together¶
- Codex as the agent running the terminal: it reads files, edits flow definitions, runs the UiPath CLI, queries the tenant, and reports results.
- Maestro Flow orchestrating an inline agent, a knowledge index, and a human task side by side.
- An improvement loop driven by evidence: run real tickets, collect human decisions, leave trace feedback, revise, and add automation only where the evidence supports it.
What you will touch today¶
- A Maestro Flow solution named
triage-plus your username, for exampletriage-user24, with three flows:triage-user24v1,triage-user24v2, andtriage-user24v3. - The shared SupportKB knowledge index in the Shared folder.
- Action Center review tasks assigned to you, with outcomes Approve, Modify & Send, and Reject.
- Trace feedback on weak runs, and a final V1/V2/V3 scoreboard.
How the prompts work¶
Every prompt in this workshop is typed exactly as written. There are six of them, plus one word to release the final batch. Codex looks up your username, email, and tenant itself, so the same text works for everyone in the room.
Your team code is triage- plus your username, the part of your email before the first underscore. A login of user24_internaltestfusion20ymux5i@uipathstaging.onmicrosoft.com gives triage-user24. Codex derives it for you in the next prompt, so you never type it.
Expect a loop, not a single prompt: build V1, run test tickets, review tasks, leave feedback, ask Codex to infer weaknesses, build V2, run a second ticket set, complete tasks, build V3 with safe automation, run final tickets, compare results.
| Step | Focus |
|---|---|
| Connect to Your Environment | Log into the VM and Automation Cloud, and confirm the workshop tenant in Studio Web |
| Prepare the Workspace | Create your working folder, start Codex and sign in the UiPath CLI, then Prompt 1: set the rules, get your team code, preflight the data |
| Build and Deploy V1 | Prompt 2: build the first flow, then deploy it from Studio Web |
| Run and Review Batch A | Prompt 3: run eight tickets through V1 and review what it gets wrong |
| Build V2 from Feedback | Prompt 4: leave trace feedback, build V2, and deploy it |
| Run Batch B and Build V3 | Prompt 5: run batch B, delegate the review, build V3 with safe auto-resolution |
| Run Batch C and Compare | Prompt 6: run the final batch and build the V1/V2/V3 scoreboard |
Something not working? If something breaks lists the known symptoms and their fixes.
Training Environment
Log in at cloud.uipath.com/tpenlabs inside your workshop VM and confirm the tenant picker shows AgenticAdvancedWorkshop before you start.