Skip to content

TEST-SCRIPT.md

Latest commit

 

History

History
116 lines (74 loc) · 3.76 KB

TEST-SCRIPT.md

File metadata and controls

116 lines (74 loc) · 3.76 KB
Raw

TEST-SCRIPT.md — Autonomous Pipeline Demo Walkthrough

Step-by-step script for presenting the autonomous pipeline demo at MVP Summit.

Pre-Demo Setup

  1. Build the sample:

    cd samples/autonomous-pipeline
npm install
npm run build

  2. (Optional) Start Aspire dashboard for live telemetry:

    docker run -d \
  -p 18888:18888 \
  -p 4317:18889 \
  -e DASHBOARD__FRONTEND__AUTHMODE=Unsecured \
  -e DASHBOARD__OTLP__AUTHMODE=Unsecured \
  mcr.microsoft.com/dotnet/aspire-dashboard:latest

  3. Verify build:

    npm run start

    You should see the full pipeline run with colored output.

Demo Script (3-5 minutes)

Act 1: The Cast (30 seconds)

Say: "Squad doesn't use agent-1, agent-2. It casts a team with real personalities."

Show: The CASTING section — four agents from The Usual Suspects universe, each with a name, role, and personality trait.

Key point: "These identities persist across sessions. You build a relationship with your agents."

Act 2: The Queue (30 seconds)

Say: "Here's a realistic development pipeline — 10 tasks from API design to security audit."

Show: The TASK QUEUE with progress bar and task list.

Key point: "Each task has a role affinity. The SDK knows which agent should handle it."

Act 3: Tier Selection (30 seconds)

Say: "Not every task needs the same resources. The SDK selects the right tier automatically."

Show: The RESPONSE TIER SELECTION section — different tasks get different tiers.

Key point: "A security audit gets the full multi-agent tier. A doc update gets lightweight. This saves cost."

Act 4: Autonomous Execution (90 seconds)

Say: "Now watch. The agents work autonomously."

Show: The scrolling execution log — agents picking up tasks, completing them, showing token counts.

Highlight these moments:

  • 📚 Skill matching — "The SDK matched the JWT auth skill to this task automatically"
  • 🔀 squad_route — "McManus just routed a follow-up task to Fenster — the tester"
  • 📋 squad_decide — "Keyser recorded a team decision: 'Use JWT with RS256'"
  • 🧠 squad_memory — "Fenster saved a learning about token refresh behavior"

Key point: "This is programmatic orchestration. Not prompt engineering. Code enforces the routing."

Act 5: The Dashboard (30 seconds)

Show: The final dashboard — agent scoreboard, cost breakdown, timeline.

Key point: "Every token tracked. Every decision logged. Full observability."

Act 6: OTel / Aspire (30 seconds, if Aspire is running)

Say: "And all of this feeds into Aspire."

Show: Switch to browser at http://localhost:18888. Show traces and metrics.

Key point: "Same OTel pipeline used in production. One line of code to enable."

Talking Points

  • SDK vs Prompt orchestration: "Prompts suggest. Code enforces."
  • squad_route: "Agents hand off work to each other — like a real team."
  • squad_decide: "Decisions propagate to every agent. No context silos."
  • squad_memory: "Agents learn. Next session, they remember."
  • Cost tracking: "Every token is accounted for. Per-agent. Per-session."
  • OTel native: "One call: initSquadTelemetry(). Traces and metrics flow to Aspire."

If Something Goes Wrong

  • No colors: Terminal doesn't support ANSI. Try Windows Terminal or iTerm2.
  • Import errors: Run npm install from the sample directory.
  • OTel not showing: Check that OTEL_EXPORTER_OTLP_ENDPOINT is set and Aspire is running on port 4317.
  • TypeScript errors: Run npm run build to check for compilation issues.

Run Tests

npm test

All tests validate SDK integration patterns — they don't require live Copilot access.