DEV-AI Agents

🤖 What is DEV-AI Agents?

A production-focused engineering agent ecosystem for turning development work into repeatable, agent-assisted workflows.

Agent ecosystem

• Agents — 30+ engineering agents plus ECC-inspired enterprise specialists covering .NET, Angular, full-stack, QA, accessibility, security, architecture, debugging, API, database, performance, refactoring, testing, domain analysis and legacy .NET.
• Skills — 16 reusable engineering guidance packs that can be composed into agent workflows.
• Plugins — selected integrations and workflows aligned with GitHub's awesome-copilot plugin ecosystem.
• MCP — curated developer-tool MCP JSON definitions adapted from the devtools collection in claude-code-templates.
• Prompts — 8 reusable task prompts for ADO-to-full-stack, ECC-inspired orchestration, .NET/Web API, Angular UI, QA and accessibility workflows.
• Settings — reusable runtime/VS Code setup presets.
• ECC-inspired layer — planning, architecture, TDD, database, E2E, security, build repair and PR-review patterns adapted for enterprise .NET + Angular + ADO.
• Registry — registry/marketplace.json is the machine-readable catalog used by DEV-CLI and the npm/npx installer.

🧠 Agentic engineering flow

Story / Requirement
       ↓
   Specification
       ↓
 Agent Planning
       ↓
   Implementation
       ↓
 Review + Security
       ↓
 QA + Accessibility
       ↓
 Evidence + PR

The repository is designed around specialized agents working as a coordinated engineering system, rather than treating every task as a single generic coding prompt.

🧩 Repository architecture

| Layer | Purpose | |---|---| | Agents | Specialized engineering personas and workflows | | Skills | Reusable domain and implementation knowledge | | Plugins | Integrations and developer workflows | | MCP | Tool connectivity and developer automation | | Registry | Machine-readable discovery and installation metadata |

📦 Install with npm / npx

DEV-AI Agents is designed as a zero-setup CLI, following the installation model of Claude Code Templates: users can install a selected component without cloning this repository.

Recommended one-command install

For a developer who wants the primary engineering agent, there is no separate SDD subcommand:

npx dev-ai-agents

This installs the single primary Dev-AI SDD agent plus the project-local SDD templates needed to run it.

After installation:

VS Code Chat → Agents → Dev-AI SDD
@Dev-AI SDD initialize
@Dev-AI SDD run story #126433

The CLI also supports the full component catalog for developers who want additional specialized agents, skills, prompts, MCP templates or settings.

Install any agent (registry-backed commands)

The registry now covers all 70 VS Code agent files under agents/. Every agent card can use the same verified command pattern; the ID must match the registry exactly.

# List every available component and target
npx dev-ai-agents --list

# Install one agent
npx dev-ai-agents --agent ai-change-impact-analyzer
npx dev-ai-agents --agent dev-ai-sdd
npx dev-ai-agents --agent dev-ai-code-explorer

# Install every VS Code agent (agents only; not MCPs, skills, plugins or settings)
npx dev-ai-agents --all-agents

# Install several agents together
npx dev-ai-agents --agent dev-ai-sdd --agent dev-ai-fullstack-engineer --agent dev-ai-qa

# Alternative npm exec form, without a global install
npm exec --yes --package=dev-ai-agents -- dev-ai-agents --agent dev-ai-sdd

# Optional project dependency, then run the CLI locally
npm install --save-dev dev-ai-agents
npx dev-ai-agents --agent dev-ai-sdd

Agents install into .github/agents/<agent-file>.agent.md. Existing files are protected by default; use --force only when you intentionally want to overwrite them. The DEV-AI app should read each agent's installCommand and npmExecCommand from registry/marketplace.json rather than constructing package names for agents that do not have separate npm packages.

A separate package command such as npx @chetan11fb/dev-ai-super-fullstack-agent is available only for the Super Agent package that is actually published separately; it is not the default distribution model for every Markdown agent.

Individual installation

Every registry component can be installed individually. You do not need to install the complete DEV-AI stack.

The CLI resolves the component from registry/marketplace.json, downloads the exact source file from this repository, and writes it to that component's declared target.

Agent

npx dev-ai-agents --agent dev-ai-fullstack-engineer

Other examples:

npx dev-ai-agents --agent dev-ai-dotnet
npx dev-ai-agents --agent dev-ai-qa
npx dev-ai-agents --agent dev-ai-accessibility
npx dev-ai-agents --agent dev-ai-ado

Default target examples:

.github/agents/dev-ai-fullstack-engineer.agent.md
.github/agents/dev-ai-dotnet.agent.md
.github/agents/dev-ai-qa.agent.md

⭐ Primary Dev-AI SDD agent

The SDD capability is not a separate CLI/package. It is the primary global engineering agent installed by the normal Dev-AI command:

npx dev-ai-agents

This installs:

• .github/agents/dev-ai-sdd.agent.md
• .specify/memory/constitution.md
• specs/_template/spec.md
• specs/_template/plan.md
• specs/_template/tasks.md
• specs/_template/sdd-state.md
• specs/_template/decisions.md

Then open VS Code Chat → Agents → Dev-AI SDD and run:

@Dev-AI SDD initialize
@Dev-AI SDD run story #126433

The workflow pauses for human review at Specification, Plan and Tasks, then continues through Analyze → Implement → Validate → Converge → Final Review.

Important: SDD does not require a new ADO MCP. If the consuming VS Code workspace already has Azure DevOps MCP configured in .vscode/mcp.json, Dev-AI SDD uses that existing MCP connection to read and analyze the story. The installer does not overwrite or duplicate MCP configuration.

For a repository with no ADO MCP, SDD can still work from local requirement text or another available integration; it never invents missing story data.

Skill

npx dev-ai-agents --skill dotnet-development
npx dev-ai-agents --skill angular-development
npx dev-ai-agents --skill accessibility

Default target:

.github/skills/<skill>/SKILL.md

Prompt

Reusable prompts can also be installed independently:

npx dev-ai-agents --prompt ado-story-to-fullstack
npx dev-ai-agents --prompt backend-dotnet-webapi
npx dev-ai-agents --prompt angular-ui-fullstack
npx dev-ai-agents --prompt qa-review-gate

Default target:

docs/prompts/<prompt>.md

MCP

npx dev-ai-agents --mcp github
npx dev-ai-agents --mcp azure-devops
npx dev-ai-agents --mcp playwright
npx dev-ai-agents --mcp figma

Default target:

.vscode/mcp/<mcp>.json

Plugin

npx dev-ai-agents --plugin ai-team-orchestration

Plugins may contain multiple files and are installed as a directory.

Setting

npx dev-ai-agents --setting vscode-fullstack
npx dev-ai-agents --setting runtime-bootstrap

Install multiple selected components

You can combine component types in one command:

npx dev-ai-agents \
  --agent dev-ai-fullstack-engineer \
  --skill dotnet-development \
  --mcp azure-devops \
  --prompt ado-story-to-fullstack

Preview the complete catalog

npx dev-ai-agents --list

The list shows each component's ID and installation target, so you can choose only what your project needs.

Backward compatibility

Older scripts using:

npx dev-ai-agents sdd

continue to work, but the recommended command is simply:

npx dev-ai-agents

Browse the catalog

npx dev-ai-agents --list

Install everything

npx dev-ai-agents --all

Use --force to intentionally overwrite an existing component:

npx dev-ai-agents --agent dev-ai-fullstack-engineer --force

The CLI reads registry/marketplace.json, downloads the selected component from its real GitHub source path, and installs it into the declared target. The registry remains the single source of truth for the installer.

Supported individual component types are agent, skill, prompt, MCP, plugin and setting. This means a developer can install only the capability they need—for example, just the .NET agent, just the ADO MCP, or just the ADO-to-full-stack prompt—without pulling the rest of the ecosystem.

Publishing note: the public package is dev-ai-agents. After the npm package release is published, developers can install the primary agent with npx dev-ai-agents or add it to a project with npm install --save-dev dev-ai-agents.

For the full individual-agent command contract, registry validation, and npm publishing steps, see docs/npm-publishing-and-dev-ai-app.md.

⚡ ECC-Inspired Enterprise Engineering Layer

DEV-AI now includes a native adaptation layer inspired by the engineering patterns in affaan-m/ECC. The goal is not to turn DEV-AI into ECC or copy unrelated language-specific assets; it is to bring the useful plan → build → test → security → database → review → PR discipline into the existing .NET + Angular + ADO ecosystem.

Added native agents

• ECC Planner
• ECC Architect
• ECC Code Reviewer
• ECC Database Reviewer
• ECC TDD Guide
• ECC E2E Runner
• ECC Security Reviewer
• ECC Build Error Resolver
• ECC TypeScript Reviewer
• ECC Refactor Cleaner
• ECC Performance Optimizer
• ECC Silent Failure Hunter
• ECC PR Test Analyzer
• ECC Spec Miner
• ECC Documentation Updater

Added reusable skills

• ECC Orchestration
• ECC TDD Workflow
• ECC Database Review
• ECC Security Review

Added workflow prompts

• ADO → ECC-Inspired Full-Stack Workflow
• ECC-Inspired PR Review Gate

Enterprise flow

ADO Story
   ↓
Spec / SDD
   ↓
Planner → Architect → Code Explorer
   ↓
.NET / EF Core + Database
   ↓
Angular / TypeScript
   ↓
TDD → Integration → Playwright E2E
   ↓
Accessibility → Security
   ↓
Code Review → PR Evidence → Pull Request

This layer is deliberately native to DEV-AI. Existing ADO, .NET, Angular, accessibility, MCP and DevOps agents remain the enterprise specialization layer, while ECC contributes reusable engineering workflow patterns.

Individual installation

npx dev-ai-agents --agent dev-ai-ecc-code-reviewer
npx dev-ai-agents --agent dev-ai-ecc-database-reviewer
npx dev-ai-agents --agent dev-ai-ecc-tdd-guide
npx dev-ai-agents --skill ecc-orchestration
npx dev-ai-agents --skill ecc-database-review
npx dev-ai-agents --prompt ado-ecc-fullstack-workflow
npx dev-ai-agents --prompt ecc-pr-review-gate

Source and attribution

Selected patterns are adapted from affaan-m/ECC, MIT licensed. The imported/adapted files identify ECC as their source. Unrelated ECC language-specific agents are intentionally not copied into DEV-AI.

🔗 Upstream references

• GitHub awesome-copilot instructions: https://github.com/github/awesome-copilot/tree/main/instructions
• GitHub awesome-copilot plugins: https://github.com/github/awesome-copilot/tree/main/plugins
• Claude Code Templates: https://github.com/davila7/claude-code-templates
• Claude Code Templates uses the same zero-setup npx model for installing individual agents, commands, skills, hooks and MCPs.

📦 Source URL rule

A component's GitHub source URL must use its real repository path, for example:

https://github.com/chetan11fb/dev-ai-agents/blob/main/agents/dev-ai-fullstack-engineer.agent.md

Do not convert the source path into .github/agents. The .github/agents path is an installation target for a consuming VS Code repository.

🚀 Installation target

For agents, the default consuming-project target is:

.github/agents/<agent-file>.agent.md

DEV-CLI must read registry/marketplace.json and use:

• path for source/GitHub links
• target for installation

🔐 Safety

Upstream content is used as engineering reference. Do not copy secrets or repository-specific credentials. MCP configuration files containing environment-variable placeholders must keep those placeholders; users provide their own credentials locally.

🧰 Real-time Windows Runtime Bootstrap

If a VS Code terminal reports that Node.js/npm are not installed, DEV-AI includes a real-time Windows bootstrap:

Set-ExecutionPolicy -Scope Process Bypass
.\scripts\bootstrap-node.ps1

The script:

• detects whether Node.js and npm are available;
• installs Node.js LTS through winget when missing;
• refreshes PATH in the current PowerShell session;
• verifies node -v, npm -v, where.exe node and where.exe npm;
• clearly asks for a new VS Code terminal when PATH refresh cannot be applied.

It never reports success without runtime verification.

🧬 Full-Stack Engineering Reference Pack

I also maintain a clearly separated upstream reference pack from Shubhamsaboo/awesome-llm-apps. The upstream project currently describes 100+ open-source AI agents, agent skills and RAG applications, including agent skills, advanced coding/architecture examples, MCP agents and generative UI projects. citeturn0search0

Imported upstream engineering assets

| Area | Imported reference | |---|---| | Git history / RCA | Commit Archaeologist | | Dependency analysis | Dependency Doctor | | PR/change hygiene | Scope Creep Detector | | Multi-agent orchestration | Advisor Orchestrator Worker | | Architecture | AI System Architect | | Coding | Multimodal Coding Agent Team | | GitHub MCP | GitHub MCP Agent | | Browser MCP | Browser MCP Agent | | MCP routing | Multi-MCP Agent Router |

The imported source files are kept under upstream/awesome-llm-apps/ and reusable skills under skills/. They are intentionally separated from DEV-AI's native agents so upstream code can be tracked without confusing it with Chetan's original implementation.

Important: these are upstream Apache-2.0 assets. Their original attribution/license metadata is preserved. The imported projects may use Python/other runtimes and are reference implementations; they do not automatically become VS Code/Copilot agents.

DEV-AI adaptation layer

For actual enterprise full-stack work, use the native DEV-AI agents for:

• Azure DevOps / ADO user stories and acceptance criteria
• .NET / C# / ASP.NET Core Web API
• Angular / TypeScript UI
• backend and microservices
• QA / tester / Playwright
• accessibility / NVDA / WCAG
• GitHub / PR / code review
• MCP configuration
• runtime/bootstrap and CI/CD
• domain/codebase discovery

The goal is upstream capability + DEV-AI enterprise specialization, not a blind replacement of the existing architecture.

🧰 Selected VS Code + Full-Stack Development Pack

A focused set of development assets from davila7/claude-code-templates is integrated directly into DEV-AI's native structure.

Agents

The imported agents are available directly under agents/ as VS Code-compatible .agent.md files, using the original upstream content:

• claude-backend-architect.agent.md
• claude-backend-developer.agent.md
• claude-code-architect.agent.md
• claude-code-explorer.agent.md
• claude-frontend-developer.agent.md
• claude-fullstack-developer.agent.md
• claude-accessibility-tester.agent.md
• claude-code-reviewer.agent.md
• claude-debugger.agent.md
• claude-playwright-tester.agent.md
• claude-refactoring-specialist.agent.md
• claude-csharp-expert.agent.md
• claude-csharp-developer.agent.md
• claude-dotnet-core-expert.agent.md
• claude-angular-architect.agent.md

Skills

The selected reusable skills are integrated directly under skills/:

• skills/development/agent-development/
• skills/development/agent-md-refactor/
• skills/development/create-plan/
• skills/development/dispatching-parallel-agents/
• skills/development/subagent-driven-development/
• skills/development/writing-skills/
• skills/utilities/playwright-skill/
• skills/web-development/react-state-management/
• skills/web-development/shadcn/

The imported assets are kept separate by filename where needed to avoid overwriting DEV-AI's existing agents, while sharing the same native agents/ and skills/ architecture. The temporary upstream/claude-code-templates/ copy has been removed. Original upstream content and attribution should be respected.

🔭 Dynatrace Expert + MCP

DEV-AI includes the Dynatrace Expert agent from GitHub's awesome-copilot ecosystem, integrated into the native agents/ structure.

Agent

npx dev-ai-agents --agent dynatrace-expert

Installs to:

.github/agents/dynatrace-expert.agent.md

It covers incident RCA, deployment impact, production error triage, performance regression, release validation, security/vulnerability analysis and DQL.

VS Code MCP

Use the included template:

mcp/dynatrace-vscode.json

For a consuming project, configure .vscode/mcp.json with the Dynatrace remote MCP endpoint and an environment-backed COPILOT_MCP_DT_API_TOKEN. Never commit the real token.

Documentation

• Dynatrace setup
• Copilot prompt cookbook
• Troubleshooting

Official Dynatrace MCP documentation: https://docs.dynatrace.com/docs/dynatrace-intelligence/dynatrace-mcp