Skip to content

subagents.md

Latest commit

 

History

History
60 lines (42 loc) · 3.34 KB

subagents.md

File metadata and controls

60 lines (42 loc) · 3.34 KB

QA: Sub-Agents — MS Agent Framework (Python)

Prerequisites

  • Demo is deployed and accessible at /demos/subagents on the dashboard host
  • Agent backend is healthy (/api/health); OPENAI_API_KEY is set on Railway; the agent server has the subagents supervisor mounted at /subagents

Test Steps

1. Basic Functionality

  • Navigate to /demos/subagents; verify the page renders within 3s with the delegation log on the left and the CopilotChat pane on the right
  • Verify data-testid="delegation-log" is visible with heading "Sub-agent delegations"
  • Verify data-testid="delegation-count" shows "0 calls" on first render
  • Verify the empty-state copy "Ask the supervisor to complete a task. Every sub-agent it calls will appear here." is visible
  • 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

Single Delegation — Research

  • Send "Give me 3 facts about reusable rockets."
  • Within 15s verify at least one data-testid="delegation-entry" appears
  • Verify the entry's badge reads "Research" (research_agent)
  • Verify data-testid="delegation-count" reads "1 calls" (or higher)
  • Verify the Task: ... line shows the research brief and the body contains a bulleted list of facts

Full Pipeline — Research -> Write -> Critique

  • Click the "Write a blog post" suggestion (sends a brief about cold exposure training)
  • Within 30s verify the delegation log contains at least 3 entries: one Research, one Writing, one Critique (in that order)
  • Verify each entry's status reads "completed"
  • Verify the supervisor's chat reply summarises the work in 1-2 sentences (it should NOT dump the full draft inline; the draft lives in the log)

Live Updates While Running

  • Send "Explain how LLMs do tool calling. Research, write a paragraph, then critique."
  • While the supervisor is running, verify data-testid="supervisor-running" badge ("Supervisor running") appears next to the title
  • Watch the delegation log: verify entries arrive incrementally as each sub-agent finishes (NOT all at once at the end) — confirms state_update(...) is emitted as a StateSnapshotEvent per tool call
  • After the supervisor finishes, verify the running badge disappears

Multiple Turns

  • Send a second prompt "Summarize the current state of reusable rockets in 1 polished paragraph, with research and critique."
  • Verify the delegation log GROWS — prior entries are preserved and new ones append (no clobber). Confirms each delegation tool reads the prior list out of the agent's current_state-driven contextvar before pushing.

3. Error Handling

  • Send an empty message; verify it is a no-op
  • Send "Hello" (no delegation needed); verify the supervisor replies without delegating, and the delegation log stays empty
  • Verify DevTools -> Console shows no uncaught errors during any flow above

Expected Results

  • Page loads within 3 seconds
  • Single research-only prompts complete within 15 seconds
  • Full research -> write -> critique pipelines complete within 30 seconds
  • Delegation entries arrive incrementally and persist across turns
  • No UI layout breaks, no uncaught console errors