fix(bpmn): fix AI Email Support Agent layout and lint issues

[HanselIdes]

Summary

Fixes two issues reported in Slack (thread) with the AI Email Support Agent blueprint.

Logic fix: Swapped "Review case resolution" and "Final answer to customer" so the human reviewer approves the agent's response before it is sent to the customer. The previous order inverted the human-in-the-loop step.

Layout and DI fixes:

• Moved "Case solved" end event to the right of both tasks; resolved overlapping sequence flow edges in that area
• Added missing <bpmndi:BPMNLabel /> on unnamed join gateway Gateway_0kp1fqr — this element was omitted by the modeling tool for unnamed gateways, which surfaced a serialization bug
• Added explicit XOR join gateway (Gateway_join_specialist) before "Specialist answered" inside the ad-hoc sub-process to fix a fake-join lint violation
• Added explicit end event downstream of the escalation throw to satisfy no-implicit-end
• Added BPMNShape/BPMNEdge for TextAnnotation_loan_approval_intro and its association, which were in the process XML but missing from the diagram (no-bpmndi errors)
• Consolidated three duplicate escalation elements and two duplicate error elements; removed unused Message_0fr5ip9

Result: 0 bpmnlint errors, 0 warnings (down from 2 errors, 8 warnings).

Test fix: Updated test cases to complete the user task ("Review case resolution") before the service task ("Final answer to customer") to match the corrected flow order.

Test plan

• npx bpmnlint --config .bpmnlintrc "solutions/bank-ai-loan-approval/AI Email Support Agent.bpmn" — exits 0, 0 errors, 0 warnings
• Diagram flow reads left-to-right: Review case resolution → Final answer to customer → Case solved
• "Case solved" end event is rightmost element in the top path
• Camunda Modeler shows "No problems found"
• mvn test passes in CI

🤖 Generated with Claude Code

[HanselIdes]

Diagram screenshot (post-fix)

Rendered via camunda-viewer + Playwright after all visual and lint fixes were applied.

Screenshot is saved locally at `/tmp/ai-email-support-agent.png` — attach it here by dragging into the GitHub web editor if you want it embedded.

What to verify visually:

• Happy path (top): Customer email received → Handle customer request → Save customer interaction → Agent as a Judge → Review case resolution → Final answer to customer → Case solved (left to right, no overlap)
• "Finalize case manually" branch terminates cleanly at "Case resolution applied by human"
• "Human control required" event subprocess sits below the main flow, clearly separated
• No overlapping elements or crossing flow lines

[Copilot]

Pull request overview

Fixes the AI Email Support Agent blueprint by correcting the human-in-the-loop ordering (review happens before sending the final customer email), and resolves multiple bpmnlint/diagram (DI) issues to reach a clean lint state.

Changes:

• Reordered the “Review case resolution” user task to occur before “Final answer to customer”.
• Added/adjusted BPMN elements and DI (explicit join gateway, explicit end event, missing labels/shapes/edges) to eliminate bpmnlint errors/warnings.
• Updated JSON process test cases to match the corrected execution order.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
solutions/bank-ai-loan-approval/AI Email Support Agent.bpmn	Corrects task sequencing and fixes DI/lint issues (explicit join/end, labels, shapes/edges, de-dup of escalation/error elements).
solutions/bank-ai-loan-approval/test/src/test/resources/test-cases/AI Email Support Agent.test.json	Updates test instruction order to complete the review user task before the final email service task, adding assertions for expected active elements.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[philippfromme]

Is this correct?

[HanselIdes]

Is this correct?

@philippfromme No, that's a weird arrow, thanks for catching it.

Fix is live