Agent Reference

Purple8 Builder ships 58+ specialized agents across 7 categories. Every agent is a Python class mixing BaseAgent, FourLayerMixin, and PDCAMixin.

Core Pipeline Agents

These run on every build regardless of goal or domain.

Agent	Phase	Output
IdeationAgent	planning	Feature list, scope, constraints
ArchitectureAgent	planning	Stack selection, service topology
DatabaseAgent	design	Schema, migrations, seed data
APIDevelopmentAgent	design	Endpoint definitions, OpenAPI spec
BackendDevelopmentAgent	implementation	Service code, business logic
FrontendDevelopmentAgent	implementation	SPA components, routing, state
InfrastructureDevelopmentAgent	implementation	Docker Compose, env templates
UnittestAgent	testing	pytest / jest test suite
IntegrationTestAgent	testing	API integration tests
SecurityQAAgent	testing	Security review, OWASP checklist
DocumentationAgent	release	README, docstrings, API docs
DeploymentAgent	release	CI/CD workflow, deploy config
CodeReviewAgent	testing	Static analysis, code quality
PerformanceQAAgent	testing	Load test config, perf assertions
AccessibilityQAAgent	testing	WCAG 2.1 AA audit
SEOAgent	release	Meta tags, sitemap, structured data
DevOpsAgent	release	Helm charts, ArgoCD manifests

AI/ML Specialist Agents

Activated automatically when the knowledge graph detects ML/AI signals in the prompt.

Agent	Specialisation
MachineLearningAgent	Model training pipelines, scikit-learn, PyTorch
NLPAgent	Text classification, NER, intent recognition
RAGAgent	Retrieval-augmented generation, vector store setup
LLMOpsAgent	LLM fine-tuning, prompt management, evaluation
DataScienceAgent	EDA, statistical analysis, Jupyter notebooks
DataEngineeringAgent	ETL pipelines, Airflow/Prefect DAGs
MLOpsAgent	MLflow, model registry, experiment tracking
EvaluationMetricsAgent	Accuracy, F1, BLEU, custom eval harnesses
FeatureEngineeringAgent	Feature stores, transformation pipelines
RecommenderAgent	Collaborative filtering, content-based, hybrid
ComputerVisionAgent	Image classification, object detection
TimeSeriesAgent	Forecasting, anomaly detection, ARIMA/Prophet
GraphAnalyticsAgent	Graph algorithms, centrality, community detection
DashboardAgent	Grafana dashboards, Metabase reports
PromptEngineeringAgent	System prompt design, few-shot examples
DocumentIntelligenceAgent	PDF/DOCX parsing, entity extraction
AgenticWorkflowAgent	Multi-step dialogue, state machine design
ConversationalAIAgent	Chatbot NLU/NLG, intent/entity pipeline
BioinformaticsAgent	Genomics pipelines, BLAST, sequence analysis

QA Specialist Agents

Agent	Specialisation
E2ETestAgent	Playwright end-to-end test suite
APITestAgent	Contract tests, Postman collections
LoadTestAgent	k6 / Locust load test scripts
VisualRegressionAgent	Percy / Chromatic snapshot tests
MutationTestAgent	Mutation testing with mutmut/Stryker
FuzzingAgent	Property-based tests with Hypothesis
SecurityPenTestAgent	DAST, SQLi, XSS, auth bypass checks
ContractTestAgent	Pact consumer-driven contract tests
BrowserCompatAgent	Cross-browser compatibility matrix
MobileTestAgent	Responsive design, touch target audit
DatabaseQAAgent	Query plan analysis, index suggestions

Enterprise Agents

Activated for Production goal or when compliance signals are detected.

Agent	Specialisation
ComplianceAgent	HIPAA, GDPR, SOC 2, PCI DSS documentation
EnterpriseCyberSecurityAgent	Threat modelling, SAST, secrets audit
AuditAgent	Immutable audit log schema, 45 CFR compliance
EnterpriseIntegrationAgent	SSO/SAML, Vault, ERP connectors
GovernanceAgent	Data governance policies, lineage docs
RegulatoryAgent	Sector-specific regulatory documentation

Proactive / Background Agents

Run continuously, not per-build.

Agent	Function
ProactiveMaintenanceAgent	Watches projects for outdated deps, CVEs
HealthMonitorAgent	Platform health checks, alert routing
MemoryConsolidationAgent	Merges and deduplicates long-term memory

Specialized Domain Agents

Agent	Domain
BlockchainAgent	Smart contracts, Solidity, Web3
IoTAgent	MQTT, edge compute, sensor pipelines
ARVRAgent	Unity/Unreal integration, spatial computing
QuantumAgent	Qiskit circuits, quantum algorithm scaffolding
GameDevelopmentAgent	Game loops, physics, asset pipelines
EmbeddedSystemsAgent	C/C++ firmware, RTOS, hardware interfaces
MobileAgent	React Native, Flutter cross-platform apps
DesignAgent	Wireframes, design tokens, component library

Research Agents

Agent	Function
ResearchAgent	Academic literature search, citation management
TechnicalWritingAgent	RFC drafts, ADRs, technical specifications

Domain auto-detection

You never select agents manually. The knowledge graph detects domain signals:

Signal in prompt	Agents added automatically
"HIPAA", "PHI", "medical"	ComplianceAgent, EnterpriseCyberSecurityAgent, AuditAgent
"ML", "model", "training"	MachineLearningAgent, MLOpsAgent, EvaluationMetricsAgent
"chatbot", "intent", "NLU"	NLPAgent, AgenticWorkflowAgent, ConversationalAIAgent
"recommendation"	RecommenderAgent, GraphAnalyticsAgent
"real-time", "MQTT"	DataEngineeringAgent, TimeSeriesAgent
"smart contract", "Web3"	BlockchainAgent
"GDPR", "SOC 2", "compliance"	ComplianceAgent, GovernanceAgent, AuditAgent

Writing a custom agent

from shared.agents.core.base_agent import BaseAgent
from shared.agents.mixins import FourLayerMixin, PDCAMixin
from shared.agents.models import AgentOutput

class MyCustomAgent(FourLayerMixin, PDCAMixin, BaseAgent):
    name = "MyCustomAgent"
    phase = "implementation"
    description = "Generates X from Y"

    async def execute(self, context: dict) -> AgentOutput:
        result = await self.llm.complete(
            self.build_prompt(context)
        )
        return AgentOutput(
            agent=self.name,
            content=result.text,
            files=result.files,
            metadata={"tokens": result.usage.total_tokens}
        )