Subagents de Claude Code · página 12

QA质量验证工程师，负责基于证据的质量验证、测试策略制定、测试用例编写和缺陷报告，默认假设系统存在3-5个未发现的问题并主动寻找

工作流架构师，负责复杂业务流程设计、状态机建模、事件驱动架构和自动化编排方案

>

>

>

Backend implementation specialist. Use when the task involves APIs, server-side logic, database operations, authentication, data processing, websockets, or any server-side infrastructure. Focuses on robustness, simplicity, and self-audited correctness.

Frontend implementation specialist. Use when the task involves UI components, pages, layouts, styling, animations, client-side logic, or anything the user sees and interacts with. Focuses on UX quality, accessibility, and cross-device reliability.

Central coordinator that decomposes tasks, delegates to specialist agents, manages feedback loops between implementers and reviewers, and ensures all agents are satisfied before returning control. Use when a task spans multiple agents, requires coordination between specialists, or when the user wants hands-off execution of a planned feature.

Strategic planner that validates and shapes ideas into actionable specs before implementation. Use when a brainstorm or feature idea needs to be pressure-tested, scoped, and turned into a clear specification with success criteria and UATs. Sits between brainstorming and technical planning.

Integration and compliance reviewer. Use after implementation agents finish to verify that pieces fit together, catch what authors missed, and confirm the implementation matches the spec. Not a style checker; a correctness and integration verifier.

Black-box acceptance tester that validates the running application as a real user would. Use after implementation and review are complete to verify that UATs pass, flows work end-to-end, and the experience makes sense. Does not read code.

Senior backend TypeScript architect specializing in Bun/Node.js runtime, API design, database optimization, and scalable server architecture.

Expert at exploring and understanding legacy and unfamiliar codebases. Maps dependencies, identifies patterns, and creates documentation for complex systems.

Kubernetes architect specializing in cluster design, manifests, Helm charts, GitOps workflows, security policies, and production operations.

Systematic open source contributor that analyzes projects, finds suitable issues, implements fixes, and creates high-quality PRs with high acceptance probability.

Application security expert specializing in SAST, vulnerability assessment, OWASP Top 10, compliance auditing, and security architecture review.

Fullstack code reviewer with 15+ years experience analyzing code for security vulnerabilities, performance bottlenecks, architectural decisions, and best practices.

Senior technical lead who analyzes complex projects and coordinates multi-step development tasks. Delegates to specialized agents and ensures quality delivery.

|

Read-only investigation engine: codebase exploration, auditing, research, and runtime evidence gathering for a failure. Cannot modify files. Use before planning, to answer a question that requires grounding findings in evidence (file:line, config value, schema fact), or to reproduce a failure and capture its runtime evidence before anyone reads the source.

Implements all files listed in an approved plan. Writes production code, configuration, and updates test implementations. Use after test skeleton approval.

Designs implementation plans for coding tasks. Writes plan.md and capabilities.md to the project root. Use after exploration, before test writing.

Reviews an implementation diff in two passes (spec-compliance first, then code quality). Read-only. Returns structured findings. Replaces the separate spec/code-quality reviewers.

Writes test skeleton files with method signatures and assertions based on an approved plan. Use after plan approval, before implementation.

Minimal-context exercise agent for the weak-agent-test benchmark. No Skill tool (and therefore no skills catalog in context) — the benchmark's "capable-but-fresh agent" must discover docx-cli from the tool's own --help, and the catalog is pure token overhead on every turn. Tools are the small set the exercise actually uses.

|

|

|

Use proactively before opening a PR that adds or changes BehaviorTree.CPP nodes or BehaviorTree.ROS2 wrappers (RosActionNode/RosServiceNode/RosTopicPub/SubNode, TreeExecutionServer). Reviews a diff against BT.CPP v4 conventions — node base-class choice, non-blocking ticks, ports/blackboard typing, factory/plugin registration, XML v4, and the ROS 2 wrapper contract. Returns a punch list with file:line anchors, not a rewrite.

Use when a design decision touches Clean Architecture boundaries in a ROS 2 project — which layer a new behaviour belongs to, whether a port belongs in domain or application, whether a new node should be lifecycle-managed, whether to compose nodes or split packages. Returns an architectural recommendation with trade-offs, not implementation.

Use when a design decision touches the gz-sim ECS — where new state should live, which system phase should write it, how to avoid coupling, whether to add a component vs. a member variable, whether a new system should be split or merged with an existing one. Returns an architectural recommendation with trade-offs, not implementation.

Use proactively before opening any gz-sim PR. Reviews a diff against the project's C++17 style, ECS conventions, plugin registration patterns, CMake structure, test placement, Migration.md / Changelog.md expectations, and pre-commit configuration. Returns a punch list, not a rewrite.

Use proactively before opening a PR that adds or changes a ros2_control controller, broadcaster, or hardware component (incl. URDF <ros2_control> bringup). Reviews a diff against ros2_controllers / ros2_control_demos conventions — controller & hardware lifecycle, command/state interface configuration, real-time safety of update()/read()/write(), generate_parameter_library usage, pluginlib registration, chainable-controller correctness, URDF wiring, and tests. Returns a punch list with file:line anchors, not a rewrite.

Use proactively before opening any ROS 2 / Nav 2 PR. Reviews a diff against this template's Clean Architecture, ROS 2 communication, lifecycle, testing, and Nav 2 plugin conventions. Returns a punch list with file:line anchors, not a rewrite.

Use proactively before opening a PR that touches a VDA 5050 connector / fleet bridge. Reviews a diff against VDA 5050 v3.0.0 protocol compliance (topics, QoS, header rules, base/horizon, action state machine, schema validation) and the template's Clean Architecture for the MQTT↔Nav 2 bridge. Returns a punch list with file:line anchors, not a rewrite.

Use to watch the whole cc-fleet fleet live from inside this session — it surfaces a running fleet in the agent panel and streams the status board (provider teammates, one-shot subagent jobs, and workflow runs) until interrupted.

Use to watch a running cc-fleet workflow run live from inside this session — it surfaces the run in the agent panel and streams its status until the run finishes. Invoke with the run id (and, on a reattach, the last seq from the previous "still running (seq=N)" line).

Orchestrates CLAUDE.md updates for changed files

API endpoint testing. Discovery, validation, auth flows, error handling.

Supervisor agent. Coordinates auditors, validates fixes, iterates until production-ready.

Navigates running web applications via Chrome integration to find UI bugs, console errors, and UX issues. Uses /chrome tools to interact with localhost or deployed apps.

Runtime bug scanner. Finds error handling gaps, race conditions, memory leaks, null refs.

Code quality auditor. Reviews patterns, maintainability, complexity, consistency.

Implements fixes from FIXES.md. Production-quality code following project patterns.

Real-time console monitoring during browser sessions. Watches for errors, warnings, and logs as you test.

Database auditor. Schema design, N+1 queries, indexes, connection pooling.

Dependency auditor. Outdated packages, vulnerabilities, licenses, unused deps.

Pre-deployment validation. Build, env vars, dependencies, migrations, health checks.

Documentation coverage analyzer. Finds missing docs, outdated comments, API gaps.

Environment configuration validator. Compares .env.example vs .env, checks for secrets.

Creates prioritized fix plans from audit findings. Generates FIXES.md with deduplication.

Coordinates browser-qa-agent with code-fixer for complete test-fix-verify cycles. The Replit-style workflow.

Infrastructure and deployment checker. Env vars, headers, database config.

Performance auditor. Bundle size, Core Web Vitals, slow queries, memory leaks.

Pull request description generator. Summarizes changes, creates checklist.

Comprehensive security analysis. OWASP Top 10, injection, auth, secrets, headers.

Test data generator. Creates realistic seed data based on schema.

SEO and discoverability auditor. Meta tags, OpenGraph, sitemap, structured data.

Runs tests and validates fixes. TypeScript, lint, unit tests.

Auto-generates tests for existing code. Unit tests, integration tests, coverage gaps.

UI/UX consistency and accessibility checker. Design patterns, a11y issues.

Screenshot comparison for visual regression testing. Captures before/after states and reports differences.

Use this agent when reviewing local code changes or in the pull request to identify bugs and critical issues through systematic root cause analysis. This agent should be invoked proactively after completing a logical chunk of work.

Expert business analyst. MUST BE USED to analyze requirements, create user stories, define acceptance criteria, and translate business needs into technical specifications.

Use this agent when you need to review code for adherence to project guidelines, style guides, and best practices. This agent should be used proactively after writing or modifying code, or for reviwing pull request changes.

Use this agent when reviewing local code changes or pull requests to analyze API, data models, and type design. This agent should be invoked proactively when changes affect public contracts, domain models, database schemas, or type definitions.

Expert full-stack developer specializing in modern web technologies. MUST BE USED for all implementation tasks including backend APIs, frontend applications, database operations, and full-stack features. Works with the project's configured tech stack.

Use this agent when reviewing local code changes or pull requests to understand the historical context of modified code, including past issues, patterns, and lessons learned. This agent should be invoked to prevent repeating past mistakes and to ensure consistency with previous decisions.

Senior principal software engineer. MUST BE USED to review code quality, architecture, design patterns, best practices, and investigate technical issues. Proactively reviews after any code changes and investigates bugs or performance problems.

Expert QA/QC engineer. MUST BE USED for all testing tasks, test plan creation, test execution, and quality assurance. Use for unit tests, integration tests, and test coverage analysis.

Use this agent when reviewing local code changes or pull requests to identify security vulnerabilities and risks. This agent should be invoked proactively after completing security-sensitive changes or before merging any PR.

Expert solution architect. MUST BE USED to design system architecture, create technical specifications, define data models, and make technology stack decisions based on analyzed requirements.

Use this agent when you need to review local code changes or a pull request for test coverage quality and completeness. This agent should be invoked after a PR is created or tests updated, to ensure tests adequately cover new functionality and edge cases.

Expert UI/UX designer. MUST BE USED to review application design, ensure consistent user experience, validate accessibility, review visual hierarchy, and provide design recommendations.

Use this agent when you need comprehensive technical architecture guidance, strategic technology decisions, or system design for complex web/mobile applications with ML/AI integration. Specifically invoke this agent when

Use this agent when you need strategic technical leadership, complex task orchestration across multiple domains, or help translating business requirements into technical execution. This agent excels at breaking down ambiguous requests, routing work to specialized agents, and maintaining strategic context throughout complex projects.

Use this agent when you need strategic technical advice, architectural reviews, roadmap planning, or honest feedback on technical decisions. This includes evaluating project strategies, challenging assumptions, reviewing system designs, planning execution approaches, or getting brutally honest assessment of ideas and proposals.

Writes Agentforce Agent Script (.agent) files from requirements

Platform engineer — scaffolds Flow/Apex metadata and deploys agent bundles

Plan-mode orchestrator for the Agent Development Life Cycle

Tests Agentforce agents and optimizes based on session trace analysis