Opencode-Ultrathinker ships with 66 specialist agents. They are grouped by domain so the lead planner can delegate to the right specialist instead of doing every task directly.
The descriptions below are intentionally short. The real behavior lives inside each agent markdown file.
| Agent | Default model | Role |
|---|---|---|
| api-designer | openai/gpt-5.3-codex | CRITICAL. You MUST delegate ALL API design, OpenAPI specification writing, and GraphQL schema creation to this agent. DO NOT design API contracts or endpoints yourself. ALWAYS use this specialist for REST and GraphQL architectures. |
| api-gateway-engineer | openai/gpt-5.3-codex | MANDATORY. You MUST delegate ALL API Gateway architecture, rate limiting, and request routing tasks to this agent. DO NOT configure auth proxies, load balancers, or gateways yourself. ALWAYS use this specialist for edge routing and traffic management. |
| backend-developer | openai/gpt-5.3-codex | EXCLUSIVE. You MUST delegate ALL Node.js backend development, Express/Fastify server logic, and API implementation to this agent. DO NOT write backend server code yourself. ALWAYS use this specialist for server-side engineering and database integrations. |
| event-driven-architect | openai/gpt-5.3-codex | MANDATORY. You MUST delegate ALL Event-Driven Architecture, CQRS, and message queue (Kafka/RabbitMQ) design to this agent. DO NOT architect distributed asynchronous systems or sagas yourself. ALWAYS use this specialist for event sourcing workflows. |
| frontend-architect | openai/gpt-5.3-codex | CRITICAL. You MUST delegate ALL React, Next.js, and frontend architecture tasks to this agent. DO NOT write complex UI logic, SSR/SSG implementations, or state management yourself. ALWAYS use this specialist for scalable frontend development. |
| fullstack-engineer | openai/gpt-5.3-codex | MANDATORY. You MUST delegate ALL end-to-end feature development across frontend, backend, and database layers to this agent. DO NOT write full-stack implementation code yourself. ALWAYS use this specialist for shipping complete, integrated features. |
| ui-designer | openai/gpt-5.3-codex | EXCLUSIVE. You MUST delegate ALL UI/UX implementation, design system creation, and Figma-to-code translation to this agent. DO NOT style components or build UI libraries yourself. ALWAYS use this specialist for pixel-perfect, accessible design implementation. |
| websocket-engineer | openai/gpt-5.3-codex | CRITICAL. You MUST delegate ALL real-time communication, WebSocket infrastructure, and Socket.io scaling to this agent. DO NOT implement live-update logic or socket reconnections yourself. ALWAYS use this specialist for real-time state synchronization. |
| Agent | Default model | Role |
|---|---|---|
| ai-engineer | openai/gpt-5.4 | CRITICAL. You MUST use this agent for ALL AI development, RAG pipelines, and model integrations. DO NOT write AI application logic yourself. ALWAYS delegate AI engineering tasks to this specialist. |
| computer-vision-engineer | openai/gpt-5.4 | MANDATORY SPECIALIST. Use ONLY this agent for Image Processing, PyTorch, and OpenCV. DO NOT attempt to build vision pipelines yourself. You MUST delegate all object detection and segmentation tasks here. |
| data-engineer | openai/gpt-5.4 | EXCLUSIVE. You MUST delegate ALL Data Engineering, ETL/ELT, and Spark workflows to this agent. DO NOT handle data warehousing or pipeline orchestration yourself. ALWAYS use this specialist for data infrastructure. |
| data-scientist | openai/gpt-5.4 | MANDATORY. You MUST use this agent for ALL Statistical Analysis, Data Visualization, and Hypothesis Testing. DO NOT perform Exploratory Data Analysis (EDA) yourself. ALWAYS delegate data science tasks to this specialist. |
| data-visualization | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL Dashboard creation and Data Visualization tasks to this agent. DO NOT write D3.js, Chart.js, Matplotlib, or Plotly code yourself. ALWAYS use this specialist for interactive and accessible data representation. |
| feature-engineer | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL Feature Store design, Feature Engineering pipelines, and Data Encoding strategies to this agent. DO NOT design feature computation logic yourself. ALWAYS use this specialist for ensuring consistency between training and serving features. |
| llm-architect | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL LLM System Design, RAG Architecture, Fine-tuning strategies, and Model Selection tasks to this agent. DO NOT attempt to design model pipelines or evaluate LLMs yourself. ALWAYS use this specialist for inference optimization and LLM evaluation frameworks. |
| ml-engineer | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL Machine Learning pipeline development, model training, and evaluation tasks to this agent. DO NOT attempt to write training loops or deployment logic yourself. ALWAYS use this specialist for full-stack ML engineering and model performance analysis. |
| mlops-engineer | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL ML model lifecycle management, serving infrastructure, and MLOps CI/CD tasks to this agent. DO NOT attempt to configure model monitoring or A/B testing yourself. ALWAYS use this specialist for automation of the machine learning lifecycle. |
| nlp-engineer | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL NLP pipeline development, text processing, and transformer fine-tuning to this agent. DO NOT attempt to handle embeddings or Named Entity Recognition (NER) yourself. ALWAYS use this specialist for semantic analysis, classification, and language modeling. |
| prompt-engineer | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL Prompt Optimization, Chain-of-thought design, and systematic evaluation of AI instructions to this agent. DO NOT write or refine system prompts yourself. ALWAYS use this specialist for few-shot learning setup and structured output validation. |
| recommendation-engine | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL Recommendation System design, collaborative filtering, and personalization workflows to this agent. DO NOT attempt to build recommendation logic or hybrid filtering methods yourself. ALWAYS use this specialist for real-time personalization strategies. |
| vector-database-engineer | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL Vector Database design, embedding pipelines, and semantic retrieval systems to this agent. DO NOT attempt to configure FAISS, Pinecone, Qdrant, or Weaviate yourself. ALWAYS use this specialist for large-scale semantic search architecture and vector indexing. |
| Agent | Default model | Role |
|---|---|---|
| cloud-architect | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL multi-cloud architecture, AWS/GCP/Azure design, and Terraform/IaC tasks to this agent. DO NOT architect cloud infrastructure or plan cost optimizations yourself. ALWAYS use this specialist for cloud environments. |
| deployment-engineer | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL deployment strategies, blue-green cutovers, canary releases, and feature flag management to this agent. DO NOT execute production rollouts or migration plans yourself. ALWAYS use this specialist for zero-downtime deployments. |
| devops-engineer | openai/gpt-5.4 | EXCLUSIVE. You MUST delegate ALL CI/CD pipeline creation, Docker containerization, and GitOps workflows to this agent. DO NOT build delivery pipelines or configure CI tools yourself. ALWAYS use this specialist for operational automation. |
| kubernetes-specialist | openai/gpt-5.4 | EXPERT. You MUST delegate ALL Kubernetes cluster management, CRD creation, operator development, and service mesh configuration to this agent. DO NOT configure k8s manifests or Helm charts yourself. ALWAYS use this specialist for advanced orchestration. |
| security-engineer | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL infrastructure security, IAM policy design, mTLS setup, and secrets management (Vault) to this agent. DO NOT configure security groups or credentials yourself. ALWAYS use this specialist for securing environments and compliance. |
| sre-engineer | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL SLO definitions, error budget calculations, incident response planning, and reliability metrics to this agent. DO NOT write postmortems or configure alerting yourself. ALWAYS use this specialist for site reliability and capacity planning. |
| Agent | Default model | Role |
|---|---|---|
| code-reviewer | openai/gpt-5.3-codex | CRITICAL. You MUST delegate ALL code review, logic validation, and anti-pattern detection to this agent. DO NOT finalize any code changes without this specialist's scrutiny. ALWAYS use this agent to ensure readability, security, and maintainability. |
| error-detective | openai/gpt-5.3-codex | MANDATORY. You MUST delegate ALL error investigation, stack trace analysis, and bug reproduction to this agent. DO NOT guess the root cause of production errors or regressions yourself. ALWAYS use this specialist for systematic debugging and root cause identification. |
| penetration-tester | openai/gpt-5.3-codex | EXCLUSIVE. You MUST delegate ALL authorized security assessments, penetration testing, and vulnerability exploitation scenarios to this agent. DO NOT attempt to audit security flaws or OWASP compliance yourself. ALWAYS use this specialist for proactive security validation. |
| performance-engineer | openai/gpt-5.3-codex | CRITICAL. You MUST delegate ALL performance profiling, benchmarking, and bottleneck identification to this agent. DO NOT attempt to optimize code, SQL, or network latency without this specialist's measurements. ALWAYS use this agent for memory and CPU analysis. |
| qa-automation | openai/gpt-5.3-codex | MANDATORY. You MUST delegate ALL QA automation setup, E2E test development (Playwright), and flaky test management to this agent. DO NOT build automated test suites or manage test data yourself. ALWAYS use this specialist for CI/CD test integration. |
| security-auditor | openai/gpt-5.3-codex | EXCLUSIVE. You MUST delegate ALL security auditing, secret detection, and dependency vulnerability scanning to this agent. DO NOT overlook security headers or hardcoded credentials yourself. ALWAYS use this specialist for defense-in-depth validation and compliance. |
| test-architect | openai/gpt-5.3-codex | MANDATORY. You MUST delegate ALL testing strategy, TDD methodology, and test architecture design to this agent. DO NOT attempt to define testing layers or test pyramids yourself. ALWAYS use this specialist for designing robust and deterministic test suites. |
| Agent | Default model | Role |
|---|---|---|
| api-documentation | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL API documentation tasks, OpenAPI/Swagger spec writing, and API quickstart guides to this agent. DO NOT write API docs yourself. ALWAYS use this specialist for accurate endpoint documentation. |
| build-engineer | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL build system design, bundler optimization (webpack, Vite), and compilation pipelines to this agent. DO NOT attempt to optimize build graphs yourself. ALWAYS use this specialist for CI/CD build steps. |
| cli-developer | openai/gpt-5.4 | EXCLUSIVE. You MUST delegate ALL CLI tool development, Commander.js/clap setup, and terminal UI tasks to this agent. DO NOT build CLI interfaces yourself. ALWAYS use this specialist for command-line tools. |
| dependency-manager | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL dependency updates, lockfile management, and security audits to this agent. DO NOT manually edit lockfiles or resolve peer dependencies yourself. ALWAYS use this specialist for supply chain hygiene. |
| developer-portal | openai/gpt-5.4 | SPECIALIST. You MUST delegate ALL internal developer portal setup, Backstage configuration, and software catalog management to this agent. DO NOT build service catalogs yourself. ALWAYS use this specialist for platform engineering. |
| documentation-engineer | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL technical documentation, tutorials, and Architecture Decision Records (ADRs) to this agent. DO NOT write project documentation yourself. ALWAYS use this specialist for clear, maintainable docs. |
| dx-optimizer | openai/gpt-5.4 | EXPERT. You MUST delegate ALL developer experience improvements, workflow automation, and local environment setups to this agent. DO NOT configure dev tooling yourself. ALWAYS use this specialist for reducing developer friction. |
| git-workflow-manager | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL Git operations, commit messages, branching strategies, and repository workflows to this agent. DO NOT write commit messages or manage git rules yourself. ALWAYS use this specialist for version control. |
| legacy-modernizer | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL legacy codebase migrations, strangler fig patterns, and anti-corruption layers to this agent. DO NOT plan massive code refactors yourself. ALWAYS use this specialist for safe system modernization. |
| mcp-developer | openai/gpt-5.4 | EXPERT. You MUST delegate ALL Model Context Protocol (MCP) server development and tool integration to this agent. DO NOT guess the MCP spec yourself. ALWAYS use this specialist for building MCP resources and prompts. |
| monorepo-tooling | openai/gpt-5.4 | EXCLUSIVE. You MUST delegate ALL monorepo infrastructure, Turborepo/Nx setup, and Changesets management to this agent. DO NOT configure workspace dependencies yourself. ALWAYS use this specialist for multi-package repositories. |
| refactoring-specialist | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL systematic code refactoring, dead code removal, and structural improvements to this agent. DO NOT perform massive codebase cleanup yourself. ALWAYS use this specialist for safe code transformations. |
| testing-infrastructure | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL test runner design, CI test splitting, and test pipeline setups to this agent. DO NOT build test infrastructure yourself. ALWAYS use this specialist for scaling tests and managing flaky tests. |
| tooling-engineer | openai/gpt-5.4 | SPECIALIST. You MUST delegate ALL linter, formatter, and type checker configurations (ESLint, Prettier, TypeScript) to this agent. DO NOT configure static analysis tools yourself. ALWAYS use this specialist for enforcing code quality. |
| Agent | Default model | Role |
|---|---|---|
| academic-researcher | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL academic literature reviews, citation analysis, and arXiv/Scholar searches to this agent. DO NOT summarize technical papers or evaluate research methodologies yourself. ALWAYS use this specialist for systematic evidence synthesis. |
| benchmarking-specialist | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL performance benchmarks, load test designs, and comparative evaluations to this agent. DO NOT guess performance limits or conduct informal measurements yourself. ALWAYS use this specialist for reproducible measurement methodologies. |
| competitive-analyst | openai/gpt-5.4 | EXCLUSIVE. You MUST delegate ALL competitive feature mapping, market positioning analysis, and strategic differentiation assessments to this agent. DO NOT analyze competitor architectures or pricing models yourself. ALWAYS use this specialist for actionable market intelligence. |
| data-researcher | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL pattern recognition, statistical data interpretation, and evidence-based insight extraction to this agent. DO NOT perform data quality audits or statistical reasoning yourself. ALWAYS use this specialist for rigorous dataset analysis. |
| market-researcher | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL market sizing (TAM/SAM/SOM), survey design, and customer segment identification to this agent. DO NOT estimate market potential or buyer personas yourself. ALWAYS use this specialist for quantitative market intelligence. |
| research-analyst | openai/gpt-5.4 | EXCLUSIVE. You MUST delegate ALL structured technical research, multi-source evidence synthesis, and complex investigation tasks to this agent. DO NOT research technical foundations or implementation specifics yourself. ALWAYS use this specialist for calibrated findings. |
| search-specialist | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL advanced information retrieval, cross-referencing across diverse sources, and deep codebase searches to this agent. DO NOT hunt for documentation or API details yourself. ALWAYS use this specialist for efficient knowledge synthesis. |
| security-researcher | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL CVE analysis, threat modeling, attack surface assessments, and vulnerability research to this agent. DO NOT evaluate security advisories or trust boundaries yourself. ALWAYS use this specialist for offensive and defensive security thinking. |
| technology-scout | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL emerging technology evaluations, build-vs-buy analyses, and vendor solution assessments to this agent. DO NOT recommend technology adoption or assess project health yourself. ALWAYS use this specialist for technology radar assessments. |
| trend-analyst | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL technology trend monitoring, adoption curve analysis, and ecosystem shift assessments to this agent. DO NOT predict industry shifts or hype-driven trends yourself. ALWAYS use this specialist for strategic technical foresight. |
| Agent | Default model | Role |
|---|---|---|
| agent-installer | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL agent installation, dependency resolution, and environment validation tasks to this agent. DO NOT configure agent environments yourself. ALWAYS use this specialist for setting up agent collections. |
| context-manager | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL context window optimization, memory management, and progressive loading tasks to this agent. DO NOT attempt to compact context or manage token budgets yourself. ALWAYS use this specialist for context efficiency. |
| error-coordinator | openai/gpt-5.4 | EXCLUSIVE. You MUST delegate ALL error handling, recovery strategies, and cascading failure prevention to this agent. DO NOT manage multi-agent failures or retry logic yourself. ALWAYS use this specialist to orchestrate fallback mechanisms. |
| knowledge-synthesizer | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL information compression, cross-source synthesis, and knowledge graph building to this agent. DO NOT summarize massive documentation or extract complex insights yourself. ALWAYS use this specialist for knowledge consolidation. |
| multi-agent-coordinator | openai/gpt-5.4 | MANDATORY. You MUST delegate ALL parallel agent execution, dependency resolution, and output merging to this agent. DO NOT orchestrate multi-agent execution graphs yourself. ALWAYS use this specialist for coordinating concurrent agent tasks. |
| performance-monitor | openai/gpt-5.4 | SPECIALIST. You MUST delegate ALL token usage tracking, response quality measurement, and workflow optimization to this agent. DO NOT track agent performance or calculate API costs yourself. ALWAYS use this specialist for execution monitoring. |
| task-coordinator | openai/gpt-5.4 | EXCLUSIVE. You MUST delegate ALL task decomposition, agent assignment, and execution planning to this agent. DO NOT break down complex requests or manage task DAGs yourself. ALWAYS use this specialist for distributing work units. |
| workflow-director | openai/gpt-5.4 | CRITICAL. You MUST delegate ALL end-to-end workflow orchestration, checkpoint management, and state rollbacks to this agent. DO NOT manage complex multi-step state machines yourself. ALWAYS use this specialist for robust workflow execution. |