// Deploys optional A2A capability agents (Knowledge Base, CRM) that extend the voice agent with new skills. Explains the A2A architecture, enables the capability registry, deploys agent stacks, and verifies discovery. Use after the core deployment and Daily setup are complete.
Scaffold a new A2A capability agent with Python application, Dockerfile, requirements.txt, and CDK stack following the project's established patterns
Scaffold a new local tool for the voice agent pipeline with capability-based registration, executor function, ToolDefinition, and tests
Run an ECS scaling validation test with parallel monitoring, baseline enforcement, and automated result documentation
Sets up Daily.co phone number and webhook for PSTN dial-in. Guides through API key verification, phone number purchase, pinless dial-in configuration, and secrets sync. Use after deploying infrastructure, when setting up a phone number, or when configuring dial-in.
Scaffolds a new A2A capability agent with Python application, Dockerfile, requirements.txt, and CDK stack. Use when adding a new remote tool or service that the voice agent discovers via CloudMap.
Scaffolds a new local tool for the voice agent pipeline with capability-based registration, executor function, and tests. Use when adding a tool that runs inside the voice agent container and may need transport or SIP session access.
| name | deploy-capability-agents |
| description | Deploys optional A2A capability agents (Knowledge Base, CRM) that extend the voice agent with new skills. Explains the A2A architecture, enables the capability registry, deploys agent stacks, and verifies discovery. Use after the core deployment and Daily setup are complete. |
You are guiding the user through extending their voice agent with A2A (Agent-to-Agent) capability agents. This is optional — the voice agent works without them, but they add powerful features.
/configure-daily and wants to extend the agentThe core deployment must be complete:
/deploy-cloud-api or /deploy-sagemaker has been run/configure-daily has been run (phone number active)Before deploying anything, set context for the user:
Current voice agent capabilities:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
• Natural conversation — powered by Claude on Bedrock
• get_current_time — "What time is it?"
• hangup_call — "Goodbye" / ask to end the call
Tool calling is enabled by default. The agent can hold conversations
and use the two built-in tools above. No other tools are active yet.
Explain the architecture before deploying:
A2A Architecture:
━━━━━━━━━━━━━━━━
┌─────────────────┐
│ Voice Agent │
│ (ECS hub) │
└────────┬────────┘
│ polls every 30s
┌────────▼────────┐
│ CloudMap │
│ (discovery) │
└──┬──────────┬──┘
│ │
┌────────▼──┐ ┌───▼────────┐
│ KB Agent │ │ CRM Agent │
│ (ECS spoke)│ │(ECS spoke) │
└────────────┘ └────────────┘
Key points:
Verify the core deployment is healthy and check whether agents are already deployed:
# Core health
aws ssm get-parameter --name "/voice-agent/ecs/cluster-arn" --query 'Parameter.Value' --output text 2>/dev/null
# Capability registry status
aws ssm get-parameter --name "/voice-agent/config/enable-capability-registry" --query 'Parameter.Value' --output text 2>/dev/null || echo "not-set (defaults to false)"
# Agent stacks
aws cloudformation describe-stacks --stack-name VoiceAgentKbAgent --query 'Stacks[0].StackStatus' --output text 2>/dev/null || echo "not-deployed"
aws cloudformation describe-stacks --stack-name VoiceAgentCrmAgent --query 'Stacks[0].StackStatus' --output text 2>/dev/null || echo "not-deployed"
Present the available agents and let the user choose:
| Agent | What It Does | Stacks Deployed | Test Prompt |
|---|---|---|---|
| Knowledge Base | RAG search over documents in resources/ (FAQ, policies, product info) | VoiceAgentKnowledgeBase + VoiceAgentKbAgent | "What's your return policy?" |
| CRM | Customer lookup, case creation, case updates via DynamoDB | VoiceAgentCRM + VoiceAgentCrmAgent | "Can you look up my account?" |
Ask: "Which agents would you like to deploy? You can deploy one or both."
This is required once before deploying any agents. Skip if already enabled.
aws ssm put-parameter \
--name "/voice-agent/config/enable-capability-registry" \
--value "true" --type String --overwrite
The voice agent reads this flag at the start of each call. No container restart is needed — the next incoming call will start the CloudMap polling loop.
Get confirmation before deploying. Show the account ID and region.
USE_CLOUD_APIS=true npx cdk deploy VoiceAgentKnowledgeBase VoiceAgentKbAgent --require-approval never
This deploys:
resources/search_knowledge_base skill via A2ATakes ~5-8 minutes. The KB agent uses a direct tool executor (bypasses inner LLM) for low-latency queries (~300ms).
USE_CLOUD_APIS=true npx cdk deploy VoiceAgentCRM VoiceAgentCrmAgent --require-approval never
This deploys:
Takes ~5-8 minutes.
Seed the CRM with demo data — the DynamoDB tables are created empty. The CRM API has a /admin/seed endpoint that loads demo customers and cases:
CRM_URL=$(aws ssm get-parameter --name "/voice-agent/crm/api-url" --query 'Parameter.Value' --output text)
curl -s -X POST "$CRM_URL/admin/seed" | python3 -m json.tool
Expected response: {"message": "Demo data seeded successfully", "customers_seeded": 3, "cases_seeded": 2}
The seed data includes:
| Customer | Phone | Account Type | Test Prompt |
|---|---|---|---|
| John Smith | 555-0100 | premium | "Look up the account for 555-0100" |
| Sarah Johnson | 555-0101 | basic | "Can you find Sarah Johnson?" |
| Michael Chen | 555-0102 | enterprise | "Look up customer 555-0102" |
There are also 2 demo support cases (a billing dispute for John Smith and a service outage for Michael Chen).
Without seeding, all CRM lookups return "customer not found." This step is required for testing.
To reset and re-seed later: DELETE /admin/reset followed by POST /admin/seed.
After deployment, wait ~1 minute for the agents to register in CloudMap and be discovered. Then verify:
# Check CloudMap namespace for registered services
NAMESPACE=$(aws ssm get-parameter --name "/voice-agent/ecs/cloudmap-namespace" --query 'Parameter.Value' --output text 2>/dev/null)
if [ -n "$NAMESPACE" ]; then
NAMESPACE_ID=$(aws servicediscovery list-namespaces --query "Namespaces[?Name=='$NAMESPACE'].Id" --output text)
aws servicediscovery list-services --filters "Name=NAMESPACE_ID,Values=$NAMESPACE_ID,Condition=EQ" \
--query 'Services[].Name' --output text
fi
You should see the agent service names listed. The voice agent will discover them on the next CloudMap poll (within 30 seconds).
Prompt the user to make a test call:
Capability agents deployed:
━━━━━━━━━━━━━━━━━━━━━━━━━━
[status] Knowledge Base — "What's your return policy?"
[status] CRM — "Look up the account for 555-0100"
Call [phone number] and try one of the prompts above.
The voice agent discovers new agents automatically — no restart needed.
If the CRM agent was deployed, confirm the seed data is loaded:
CRM_URL=$(aws ssm get-parameter --name "/voice-agent/crm/api-url" --query 'Parameter.Value' --output text)
curl -s "$CRM_URL/customers" | python3 -c "
import json, sys
data = json.load(sys.stdin)
if not data:
print('WARNING: No customers found. Run POST /admin/seed to load demo data.')
else:
print(f'{len(data)} customer(s) loaded:')
for c in data:
print(f' {c.get(\"name\", \"?\")} — {c.get(\"phone\", \"?\")}')
"
If a tool isn't responding on the first call, remind the user:
Capability Agent Deployment Complete:
✅ Capability registry enabled
[status] Knowledge Base agent — deployed / skipped
[status] CRM agent — deployed / skipped
✅ Agents discoverable via CloudMap
Your voice agent now has [N] additional tools available.
Run /verify-deployment to confirm full system health.
Remind user to use the destroy-project skill to tear down all resources when done.