QA: Sub-Agents — LlamaIndex

Prerequisites

• Demo is deployed and accessible at /demos/subagents on the dashboard host
• Agent backend is healthy (/api/health); OPENAI_API_KEY is set; LlamaIndex agent_server has the subagents_router mounted at /subagents

Test Steps

1. Basic Functionality

• Navigate to /demos/subagents; verify the page renders within 3s with a left-side delegation log and a right-side CopilotChat pane
• Verify data-testid="delegation-log" is visible with heading "Sub-agent delegations"
• Verify data-testid="delegation-count" shows "0 calls" before any interaction
• Verify the empty-state copy reads "Ask the supervisor to complete a task. Every sub-agent it calls will appear here."
• Verify the chat input placeholder is "Give the supervisor a task..."
• Verify all 3 suggestion pills are visible with verbatim titles: "Write a blog post", "Explain a topic", "Summarize a topic"

2. Feature-Specific Checks

Multi-Step Delegation (research -> writing -> critique)

• Click the "Write a blog post" suggestion; verify the supervisor running indicator (data-testid="supervisor-running") appears within 2s and shows "Supervisor running"
• Within 60s, verify three data-testid="delegation-entry" rows appear in order with data-status="completed" and these badges (in order):
  • Research (blue badge)
  • Writing (emerald badge)
  • Critique (purple badge)
• Verify data-testid="delegation-count" reads "3 calls"
• Verify each entry shows a "Task:" line and an output box; the research output is a bulleted list of facts, the writing output is a paragraph, and the critique output is 2-3 bullet critiques
• Verify the supervisor's final chat message in the right-side pane summarizes the result (1-3 sentences)
• After the run completes verify the running indicator is hidden

Live "running" State

• Send "Write a short blog post about AI agents. Research first, then write."; while the supervisor is in flight, verify at least one entry briefly shows data-status="running" with the spinner and "Sub-agent is working…" copy before flipping to completed

Status Coloring & Ordering

• Click the "Explain a topic" suggestion; verify entries appear top-down in invocation order (research first, writing second, critique last)
• Verify running entries use amber styling, completed entries use neutral styling, and any failed entries use red styling

Multi-Turn Accumulation

• After one full delegation run, send "Now do the same for cold plunges"; verify the delegation log grows (previous entries preserved, 3 new entries appended) and delegation-count reads "6 calls"
• Reload the page; verify the delegation log resets to empty (state is per-session)

3. Error Handling

• Send an empty message; verify it is a no-op (no user bubble, no assistant response, no new delegation entries)
• If a sub-agent invocation fails (e.g. transient OpenAI 5xx), verify the corresponding entry shows data-status="failed" with red styling and a [sub-agent failed] message; verify the supervisor's chat reply briefly surfaces the failure rather than fabricating a result
• Verify DevTools -> Console shows no uncaught errors during any flow above

Expected Results

• Page loads within 3 seconds
• A typical research/write/critique sequence completes within 60 seconds
• The delegation log updates live as each sub-agent finalizes (entries flip from running to completed/failed)
• The supervisor running indicator reflects agent.isRunning
• Delegation entries persist across multi-turn runs within a session and reset on reload
• No UI layout breaks, no uncaught console errors