Skip to content

gen-ui-interrupt.md

Latest commit

 

History

History
75 lines (53 loc) · 4.7 KB

gen-ui-interrupt.md

File metadata and controls

75 lines (53 loc) · 4.7 KB
Raw

QA: In-Chat HITL via useInterrupt — LangGraph (Python)

Prerequisites

  • Demo is deployed and accessible at /demos/gen-ui-interrupt on the dashboard host
  • Agent backend is healthy (/api/health); OPENAI_API_KEY is set on Railway; LANGGRAPH_DEPLOYMENT_URL points at a LangGraph deployment exposing the interrupt_agent graph
  • Note: The picker card is rendered INLINE inside the chat transcript via useInterrupt({ renderInChat: true }), wired to langgraph's interrupt() primitive from the backend schedule_meeting tool.

Test Steps

1. Basic Functionality

  • Navigate to /demos/gen-ui-interrupt; verify the page renders within 3s with the CopilotChat centered in a max-w-4xl container filling full viewport height, with rounded (rounded-2xl) styling
  • Verify the CopilotChat input placeholder is visible and the transcript is empty on first load
  • Send "Hello" and verify the agent responds with a text-only reply (no time picker rendered — the agent only calls schedule_meeting when explicitly asked to book/schedule)

2. Feature-Specific Checks

Suggestions

  • Verify both suggestion pills are visible with verbatim titles:
    • "Book a call with sales"
    • "Schedule a 1:1 with Alice"

Interrupt Trigger + Inline Render (useInterrupt low-level primitive)

  • Click the "Book a call with sales" suggestion (or prompt equivalently)
  • Within 20s verify the agent invokes schedule_meeting, the backend hits interrupt({topic, attendee}), and a time-picker card renders INLINE inside the chat transcript with data-testid="time-picker-card"
  • Confirm the card IS a descendant of the chat transcript container (NOT portaled to <body>, unlike the hitl-in-app modal) — inspect in DevTools to verify the card sits between chat message bubbles
  • Verify the card header shows eyebrow "Book a call" and a topic heading reflecting the agent-supplied topic (e.g. contains "sales" / "pricing")
  • Verify the "Pick a time:" subheading and a 2x2 grid of exactly 4 slot buttons with the default labels:
    • "Tomorrow 10:00 AM"
    • "Tomorrow 2:00 PM"
    • "Monday 9:00 AM"
    • "Monday 3:30 PM"
  • Verify a "None of these work" ghost button is rendered below the slot grid

Pick-a-Slot Resume Path

  • Click one of the four time slot buttons (e.g. "Monday 9:00 AM")
  • Verify the card immediately switches to the confirmed state: data-testid="time-picker-picked", green-tinted border/background, and text "Booked for " with the label in bold
  • Verify all slot buttons disable (opacity reduced) and cannot be re-clicked
  • Verify the agent resumes within 10s and produces a chat reply confirming the meeting was scheduled for the chosen label (backend returns "Meeting scheduled for {chosen_label}: {topic}")

Cancel Path

  • Send a second prompt "Schedule a 1:1 with Alice next week to review Q2 goals."
  • Verify a fresh time-picker card renders inline (data-testid="time-picker-card") — attendee line "With Alice" should appear under the topic
  • Click the "None of these work" button
  • Verify the card switches to data-testid="time-picker-cancelled" with the text "Cancelled — no time picked."
  • Verify the agent resumes and replies that the meeting was NOT scheduled / the user cancelled

Multi-Turn

  • After completing either the pick or cancel path, send one more prompt "Book another call tomorrow morning"
  • Verify a new, independent time-picker card renders inline (previous card stays in its resolved state), the interrupt lifecycle repeats cleanly, and a second resume works end-to-end

Contract Check — Interrupt Is Low-Level

  • Confirm only the tool-triggered path renders the picker: sending a plain conversational message ("What's the weather?") should NOT render a picker
  • Confirm no approval-dialog-style modal appears at any point (this demo is inline, not modal)

3. Error Handling

  • Send an empty message; verify it is a no-op
  • Attempt to double-click a slot button rapidly; verify only one selection is committed (button disables on first click)
  • Verify no uncaught console errors across any pick / cancel / multi-turn flow above

Expected Results

  • Chat loads within 3 seconds; plain-text response within 10 seconds
  • Time picker card renders inline in the chat within 20 seconds of a schedule/booking prompt
  • Picker resolves via either a slot button (emits {chosen_time, chosen_label}) or the "None of these work" button (emits {cancelled: true}); post-resolution the card is read-only
  • Agent resume produces a confirmation message that references the chosen slot label or the cancellation
  • No UI layout breaks, no uncaught console errors, no duplicate pickers from a single interrupt