SUPERSEDED by docs/SRPVDAL_BOSS_AGENT_DEMO_PLAYBOOK.md

MIZ OKI 3.0 Demo Playbook: SRPVDAL + Boss Agent + Subagents

Purpose

This playbook gives a practical, end-to-end demo path for MIZ OKI 3.0 focused on:

1. SRPVDAL (Sense → Reason → Plan → Verify → Decide → Act → Learn)
2. Boss Agent orchestration with the 32-cell mesh and specialist agents
3. Tooling showcase (MCP tools, orchestrators, and guardrails)
4. Dual interface demo in both VSCode and Chrome

1) What to showcase (core narrative)

A. “From static automation to governed autonomy”

Open with the platform framing:

• MIZ OKI is positioned as a cloud-native autonomous decision platform.
• Boss Agent is the canonical control plane and source of truth.
• SRPVDAL extends SRPVDAL by adding Plan and Verify before execution.

B. Why SRPVDAL matters

Use this one-liner:

“We don’t just decide and act; we plan, verify, and only then execute under policy and rollback control.”

The srpvdal_plan_verify.py module explicitly defines the 7-stage pipeline and describes what PLAN and VERIFY add (DAG planning, contingencies, safety/policy validation, Devil’s Advocate checks).

C. Why Boss Agent matters

Boss Agent is where everything converges:

• Unified FastAPI runtime
• Agent + cell registry
• Legacy-compatible and modern APIs
• MCP tool registration and invocation
• SRPVDAL/SRPVDAL lifecycle telemetry and controls

2) Architecture talking points for demo

Use these as speaking bullets while screens are visible:

1. Boss Agent is canonical (boss_agent_core.py) and wired for orchestration + tools.
2. AGENT_REGISTRY includes orchestrators (moa, moe, rewoo, coding_moa) and 32 production cells.
3. SRPVDAL is integrated (PLAN + VERIFY hooks and keywords are present in core runtime).
4. APIs are broad and demo-ready (/health, /status, /api/v1/chat, /api/v1/agents, /api/v1/cells, /api/v1/srpvdal/*, legacy /process, /run, etc.).
5. UI and extension are both first-class surfaces: - Command Center UI for browser-driven operations - VSCode extension for developer-native orchestration

3) Suggested 20-minute live demo flow

Demo Segment 1 (3 min): “System is alive”

Goal

Prove the platform is online and observable.

Steps

1. Call Boss Agent health/status endpoint.
2. Show agent registry counts and available classes of agents.
3. Show cells health endpoint.

Example commands

curl -s "$BOSS_URL/health" | jq
curl -s "$BOSS_URL/status" | jq
curl -s "$BOSS_URL/api/v1/agents" | jq '.total_agents, .by_type'
curl -s "$BOSS_URL/api/v1/cells/health" | jq

Narration

• “We have orchestration, specialists, and cell services all discoverable from one runtime.”

Demo Segment 2 (5 min): “SRPVDAL in action”

Goal

Demonstrate the 7-stage policy-safe flow.

Steps

1. Send a business goal to /api/v1/chat (or /process for compatibility).
2. Ask explicitly for SRPVDAL stage trace and confidence by stage.
3. Show verify/governance behavior before action.

Example prompts

• “Reallocate paid media budget for next 7 days with minimal risk. Show full SRPVDAL stage outputs.”
• “What is current SRPVDAL confidence and what blockers are present in VERIFY?”

Example commands

curl -s -X POST "$BOSS_URL/api/v1/chat" \
  -H 'Content-Type: application/json' \
  -d '{"message":"Reallocate paid media budget for next 7 days; run SRPVDAL with explainability."}' | jq

curl -s -X POST "$BOSS_URL/process" \
  -H 'Content-Type: application/json' \
  -d '{"query":"Show pipeline status and SRPVDAL confidence by stage"}' | jq

Narration

• “PLAN decomposes tasks and contingencies, VERIFY runs safety/policy/DA checks, then DECIDE/ACT proceed only when cleared.”

Demo Segment 3 (5 min): “Boss + subagents + tools”

Goal

Show delegation to orchestrators/specialists with tool-backed execution.

Steps

1. Query /api/v1/agents and highlight orchestrator + specialist mix.
2. Trigger MOA or MOE route endpoint for the same question and compare behavior.
3. Trigger one operational tool path (example: cell invoke or orchestration invoke).

Example commands

curl -s -X POST "$BOSS_URL/api/v1/moa/collaborate" \
  -H 'Content-Type: application/json' \
  -d '{"task":"Diagnose drop in CVR and propose interventions with confidence"}' | jq

curl -s -X POST "$BOSS_URL/api/v1/moe/route" \
  -H 'Content-Type: application/json' \
  -d '{"task":"Diagnose drop in CVR and propose interventions with confidence"}' | jq

curl -s -X POST "$BOSS_URL/api/v1/cells/invoke" \
  -H 'Content-Type: application/json' \
  -d '{"cell_id":"cell_8","task":"Run GraphRAG query for conversion-drop hypotheses"}' | jq

Narration

• “The Boss chooses and coordinates subagents rather than forcing one model path for every task.”

Demo Segment 4 (4 min): “Show boss agent in Chrome”

Goal

Demonstrate business-user workflow in browser.

Surface

MIZ OKI Command Center UI (Next.js), route: /services/orchestration.

Steps

1. Open the orchestration page in Chrome.
2. Switch orchestrators (MOA/MOE/REWOO/Coding MOA/Auto-Select).
3. Run one prompt and show response + operational context.
4. Open Simulation Lab (/simulation) to show SRPVDAL/journey/system metrics.

Narration

• “Chrome demo is for operators and growth teams: chat, orchestration control, and pipeline telemetry in one panel.”

Demo Segment 5 (3 min): “Show boss agent in VSCode”

Goal

Demonstrate developer-native control plane.

Surface

vscode-boss-agent-extension.

Steps

1. Open VSCode with project workspace.
2. Launch Boss Agent Chat panel (Ctrl/Cmd+Shift+B).
3. Show orchestration mode switch (direct/moa/moe).
4. Open MCP tools browser and invoke a read-only tool.
5. Use context menu on selected code (“Explain Selected Code”).

Narration

• “VSCode is the same Boss Agent brain but optimized for engineering workflows and rapid iteration.”

4) Can we showcase Boss Agent in both VSCode and Chrome?

Yes — and you should.

• Chrome (Command Center UI): best for business operations, live orchestration, SRPVDAL/SRPVDAL visibility.
• VSCode extension: best for technical users, code-aware interaction, and MCP/tool operations from developer context.

Recommended positioning: - “Same agentic backend, two persona-specific frontends.”

5) Reliability + governance proof points (for Q&A)

If stakeholders ask “is this safe?”, emphasize:

1. VERIFY stage before action execution in SRPVDAL.
2. Guardrails and policy checks in the pipeline.
3. Rollback and audit patterns available in Boss endpoints/tooling.
4. Compatibility endpoints exist, reducing migration risk.

6) Practical prep checklist (day before demo)

• Confirm Boss URL and auth headers are configured.
• Test /health, /status, /api/v1/agents, /api/v1/cells/health.
• Pre-load 2 demo prompts:
• Budget reallocation with constraints
• Conversion drop diagnosis with causal explanation
• Open both environments:
• Chrome tab at /services/orchestration
• VSCode extension panel connected to same Boss URL
• Keep one fallback: legacy /process endpoint.

7) Demo close (30-second script)

“What you saw is MIZ OKI 3.0 operating as an autonomous yet governed system: SRPVDAL reasoning, Boss Agent orchestration across cells/subagents, and execution through tool and policy layers. We can run this from operator UI in Chrome or engineer UI in VSCode, with the same backend intelligence and controls.”

8) Run the automated live demo harness

Use the new script to execute a real endpoint sequence and generate a markdown report:

python3 scripts/demo/run_srpvdal_demo.py \
  --boss-url https://boss-agent-adk-698171499447.us-central1.run.app \
  --report artifacts/srpvdal_demo_report.md

Optional auth:

export BOSS_AGENT_TOKEN=your_token
python3 scripts/demo/run_srpvdal_demo.py \
  --boss-url https://boss-agent-adk-698171499447.us-central1.run.app

The script validates: - /health, /status, /api/v1/agents, /api/v1/cells/health, /api/v1/srpvdal/state - Chat and compatibility execution via /api/v1/chat and /process - Orchestrator routes via /api/v1/moa/collaborate and /api/v1/moe/route

It exits non-zero if any step fails, so it can be used in CI/demo readiness checks.