接入你的智能体
从发布者处安装。不是 PlugMCP 端点。
远程 URL 还是本机命令？ 两种安装方式的区别
AI Dev Kit is a complete plugin-oriented developer workspace for AI coding assistants. It ships everything needed to run professional-grade engineering workflows — TDD, code review, security audit, CI/CD, ML pipelines, infrastructure-as-code, code quality analysis, and debt tracking — across Claude Code, Codex, OpenCode, and Gemini CLI.
Think of it as a system prompt that scales to 59 specialized skills and 33 domain agents, with lifecycle hooks and automated validation.
/plugin marketplace add noah-sheldon/ai-dev-kit
/plugin install ai-dev-kit@ai-dev-kit
Or install directly via HTTPS:
/plugin marketplace add https://github.com/noah-sheldon/ai-dev-kit
/plugin install ai-dev-kit@ai-dev-kit
codex extensions add https://github.com/noah-sheldon/ai-dev-kit
Or clone and discover automatically:
git clone https://github.com/noah-sheldon/ai-dev-kit
The marketplace at .agents/plugins/marketplace.json will be discovered automatically.
cd .opencode/plugins && npm install && npm run build
Or install via HTTPS:
npx opencode plugin install https://github.com/noah-sheldon/ai-dev-kit
gemini extensions link https://github.com/noah-sheldon/ai-dev-kit
Or install via HTTPS directly:
gemini extensions install https://github.com/noah-sheldon/ai-dev-kit
copilot plugin marketplace add https://github.com/noah-sheldon/ai-dev-kit
copilot plugin install ai-dev-kit@ai-dev-kit
Or install via HTTPS shorthand:
copilot plugin install https://github.com/noah-sheldon/ai-dev-kit
qwen extensions install https://github.com/noah-sheldon/ai-dev-kit
Or install locally:
qwen extensions install /path/to/ai-dev-kit
./install.sh
ai-dev-kit/
├── agents/ 33 specialized agents (planner, architect, code-reviewer, debt analyzers, ...)
├── skills/ 59 skill playbooks (TDD, security, ML, infra, web, data, ...)
├── commands/ 41 workflow commands and shims
├── hooks/ lifecycle automation (pre-tool, post-tool, session events)
├── rules/ language-specific guidance (common, python, typescript, web)
├── manifests/ install manifests for deterministic setup
├── schemas/ JSON schemas for validation
├── docs/ architecture, design decisions, troubleshooting
├── examples/ reusable templates
├── scripts/ install, validate, sync, and template helpers
├── tests/ smoke tests and surface validation
├── .claude-plugin/ Claude Code plugin manifest + marketplace
├── .codex-plugin/ Codex plugin manifest
├── .agents/ Codex marketplace catalog
├── .gemini/ Gemini CLI extension
└── .opencode/ OpenCode project config + TypeScript plugin
agentic-engineering api-design api-integrations backend-patterns frontend-patterns frontend-design hexagonal-architecture coding-standards codebase-onboarding
tdd-workflow code-review security-review security-scan e2e-testing python-testing eval-harness verification-loop
claude-api openai-api langchain-llamaindex mlops-workflow mlops-rag pytorch-patterns deep-research exa-search search-first iterative-retrieval autonomous-agent-harness autonomous-loops continuous-agent-loop context-prune token-budget-advisor prompt-optimizer mcp-server-patterns
data-pipelines data-pipelines-ai database-migrations postgres-patterns document-processing
aws-devops aws-deployment docker-patterns deployment-patterns ci-pipeline github-ops observability-telemetry multi-agent-git-workflow
architecture-decision-records dmux-workflows documentation-lookup git-workflow skill-authoring backlog-management workflow-status
codebase-report technical-debt-report
wxt-chrome-extension
Build & quality: build-fix code-review doctor eval ml-review quality-gate review review-pr test-coverage validate verify
Workflow: checkpoint feature-dev plan project-template promote resume-session save-session sessions skill-create skill-health
Git & multi-agent: git-agent multi-agent-status
DevEx: context-budget context-prune install uninstall update-codemaps update-docs
Hook automation: hookify hookify-configure hookify-list hookify-help
Loop & continuous: loop-start loop-status
ML: e2e launch
Learning: learn learn-eval
Request → Planner → Architect → Domain Agents → AI-Judge → Implementation → Code Review → Security Review → Merge
planner agent before touching code.tdd-guide for RED/GREEN/REFACTOR discipline.ai-judge — rubric covers completeness, correctness, security, feasibility, testability.code-reviewer and security-reviewer before merge.git-agent-coordinator for clean branch management.MCP servers are not bundled with this kit to avoid corporate network/proxy issues. Add your own MCP servers as needed:
Configure MCP servers in your own .mcp.json at the project root.
Q: How do I adapt this for my project?
Start from docs/examples/project-guidelines-template.md and customize the skills you need.
Q: What if a model is unavailable?
Every agent has a fallback_model setting. If the specified model is down, it falls back to the workspace default.
Q: How do I validate everything is correct?
npm test # smoke tests
node scripts/validate-surface.js # production surface check
Q: Can I install only certain skills? Yes — the kit supports selective install. Each skill is self-contained. Copy only what you need.
Q: How do I contribute?
See CONTRIBUTING.md. Run npm test before submitting.
security-reviewer.See SECURITY.md for the full policy.
MIT — see LICENSE.
Noah Sheldon — noahsheldon.dev
Production-ready AI dev workflow scaffold with 59 skills, 33 agents, hooks, and rules for Claude Code, Codex, Gemini CLI, Copilot CLI, Qwen Code, and OpenCode.
工具
暂无工具列表。远程服务器在首次健康检查后会获得工具列表。
常见问题
在本页复制适用于 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 配置。它使用发布者的安装方式，而不是 PlugMCP 代理。
此记录未列出必需的环境变量。
Ai Dev Kit 通过 stdio 在本地运行。
上次检查时间：2026-10-28。 Production-ready AI dev workflow scaffold with 59 skills, 33 agents, hooks, and rules for Claude Code, Codex, Gemini CLI, Copilot CLI, Qwen Code, and OpenCode.