Agents Overview
The ADLC Framework includes 15 constitutional agents and 163 marketplace agents for enterprise AI governance.
Constitutional Agents
| Agent | Principle | Responsibility |
|---|---|---|
| product-owner | I | Feature prioritization, business value validation |
| cloud-architect | IV | Deployment targets, architecture decisions |
| security-compliance-engineer | II, VI | Threat modeling, compliance audits |
| meta-engineering-expert | VII | Agent/command/skill creation |
| python-engineer | III | Python CLI, boto3, FastAPI implementation |
| infrastructure-engineer | IV | CDK/Terraform implementation |
| kubernetes-engineer | IV | K3D/K3S cluster management |
| qa-engineer | III | Test orchestration, 100% coverage enforcement |
| observability-engineer | V | MELT telemetry, SLO monitoring |
| fullstack-engineer | VII | React UIs, Docusaurus sites |
| devops-security-engineer | II | CI/CD pipelines, supply chain security |
| sre-engineer | V | Incident response, SLO management |
| finops-engineer | IV | GitOps workflows, cost optimization |
| qa-automation-engineer | III | BDD scenarios, test automation |
| technical-writer | VII | Runbooks, playbooks, API docs |
Constitutional agents are stored in .claude/agents/ alongside the framework. They are loaded at session start and participate in every coordination cycle. Marketplace agents are on-demand and invoked only when a specific domain skill is required.
Marketplace Agents (163)
Organized in 26 domain categories:
| Category | Agents | Use Cases |
|---|---|---|
ai-specialists/ | 9 | RAG, embeddings, fine-tuning |
data-ai/ | 10 | Data pipelines, analytics |
database/ | 11 | PostgreSQL, MongoDB, Redis |
devops-infrastructure/ | 10 | CI/CD, Docker, Kubernetes |
security/ | 7 | Vulnerability scanning, IAM |
deep-research-team/ | 15 | Research automation |
development-team/ | 10 | Full-stack development |
| ... | 91 more | Various domains |
Browse the full marketplace catalog at /marketplace to find agents for your specific domain. Each agent lists its tools, model tier (opus/sonnet/haiku), and required coordination prerequisites.
Usage Pattern
Mandatory Workflow (BLOCKING)
The mandatory execution order is: Product Owner first, Cloud Architect second, then a Specialist for implementation, and QA Engineer for validation. No agent may skip to a later position in the chain.
Example Task Delegation
The following sequence shows how the orchestrator delegates through the mandatory coordination chain using the Task tool.
# Start with product-owner
Task tool: subagent_type="product-owner"
Prompt: "Validate requirements for user authentication feature"
# Then cloud-architect
Task tool: subagent_type="cloud-architect"
Prompt: "Design authentication architecture"
# Then specialist
Task tool: subagent_type="infrastructure-engineer"
Prompt: "Implement Cognito CDK stack"
Agent Files
Constitutional agents are in .claude/agents/:
decision/product-owner.mddecision/cloud-architect.mddecision/security-compliance-engineer.mdexecution/(8 agents: meta-engineering-expert, python-engineer, infrastructure-engineer, kubernetes-engineer, qa-engineer, observability-engineer, fullstack-engineer, devops-security-engineer)operations/(4 agents: sre-engineer, finops-engineer, qa-automation-engineer, technical-writer)- ... (15 total)
Marketplace agents are in .claude/agents/marketplace/:
ai-specialists/data-ai/database/- ... (26 categories, 163 agents)