Enterprise AI Development Platform

Full agentic workflow from ticket to production — requirements, planning, implementation, testing, verification, and PR — fully autonomous. 76+ skills across 10 AI platforms — Claude Code, Copilot CLI, VS Code Chat, Codex, Cursor, Windsurf, and more. Integrates with Azure DevOps, Jira, Figma, and live AEM data.

76
Skills
25
Copilot Agents
4
Plugins
10
AI Platforms*
13
Specialized Agents
Why

Why This Exists

Enterprise teams run complex sprints across multiple repos, multiple IDEs, and multiple team members. They need a system that handles the full lifecycle — requirements analysis, implementation planning, code generation, testing, verification, documentation, and PR creation — with governance at every step.

KAI is a structured development workflow built as a plugin system for enterprise teams. It encodes your entire sprint lifecycle — requirements, planning, execution, review, PR — into skills that orchestrate multi-agent pipelines across every major AI platform. A single command like /dx-req pulls the ticket from Azure DevOps or Jira, validates readiness against your DoR, distills developer requirements, researches the codebase with parallel subagents, and generates a team summary. Each skill chains specialized agents (Opus for deep review, Sonnet for execution, Haiku for lookups), gathers context from multiple sources (tickets, config, codebase, Figma designs, live AEM content), and writes structured output that the next skill picks up automatically.

Config-Driven

Your build commands, branch names, and conventions live in one config file. Every skill reads it. No hardcoded values, no repeated instructions.

Persistent Memory

Each skill writes structured output to local files. The next skill picks it up automatically. Sessions can end and resume without losing context.

Multi-Source Context

Skills don’t just read source code. They pull ADO/Jira tickets, Figma designs, AEM content, and browser screenshots through MCP integrations.

Autonomous Mode

The same skills that run locally also run unattended as ADO pipeline agents, triggered by webhooks. Tag a ticket, get a verified bugfix with a PR.

Every AI Platform

Same 76 skills work across 10 AI platforms — Claude Code, Copilot CLI, VS Code Chat, Codex, Cursor, Windsurf, and more. Install once, use everywhere.

Enterprise Governance

DoR validation, 6-phase verification gate, autonomous PR review, secret scanning, and branch protection. Ship with confidence.

Get Started

Ready to Try?

Design-to-Code

Figma to Production Code

One command turns a Figma URL into a production AEM component. Powered by the included aem-mcp-server for live JCR access — dialogs, components, pages, and content queries are available to AI agents in real time.

One Command /dx-figma-all <figma-url>
Extract Design context, tokens, screenshots from Figma via MCP
/dx-figma-extract
Prototype HTML/CSS using your project conventions and tokens
/dx-figma-prototype
Verify Chrome screenshot, compare against Figma, auto-fix
/dx-figma-verify
Implement Production AEM component from verified prototype
/dx-plan + /dx-step-all

For Designers & PMs

Design fidelity is verified before production code starts. Works with well-structured Figma files, adapts to hybrid designs, and even handles screenshot-based files.

For Developers

No manual pixel-matching. AI maps Figma tokens to your project’s existing SCSS variables, uses your BEM naming conventions, and generates WCAG-compliant ARIA patterns.

See the full Figma-to-Code pipeline →

Sprint Lifecycle

Every Phase Covered

AI assistance from ticket assignment through documentation -- no phase left behind.

Requirements

Fetch ticket (ADO or Jira), validate DoR, distill dev requirements, research codebase, generate team summary. Hours to minutes.

/dx-req

Design-to-Code

Extract Figma designs, generate convention-aware prototypes, verify against reference. Design to implementation-ready code.

/dx-figma-all

Planning

Step-by-step implementation plan with dependency tracking, validation against requirements, and risk resolution.

/dx-plan

Execution

Execute plan steps with test, review, commit loop. Self-healing with corrective steps on failure.

/dx-step-all

Code Review

6-phase verification gate: compile, lint, test, secret scan, architecture, AI code review with 80%+ confidence.

/dx-step-verify

Pull Request

Auto-generated PR description from team summary, work item linking (ADO/Jira), branch management.

/dx-pr
Plugins

Four Plugins, Independently Installable

Core workflow for any stack. AEM specialization for teams that need it — 12 skills and 6 agents covering the full AEM development lifecycle.

dx-core

Full development lifecycle — requirements, planning, execution, review, PR, accessibility, documentation. Works with any stack.

49 skills 7 agents

dx-aem

AEM full-flow — component dialogs, JCR content, editorial QA, browser automation, demo capture. The complete AEM development lifecycle.

12 skills 6 agents

dx-automation

24/7 autonomous agents — DoR, PR review, bug fix, DevAgent, QA, estimation. AWS Lambda + ADO pipelines.

11 skills
Engineering

What Makes This Different

Built for enterprise development teams.

Config-Driven, Never Hardcoded

One config.yaml holds all project values. Same plugin codebase runs across 4 repos and 50+ potential projects without a code change.

Memory via Markdown

Each skill writes findings to a .md file. No database, no API — just files on disk. Survives session crashes. Next skill picks up automatically.

Direct Skill Invocation

Coordinators invoke skills directly via Skill() calls. No intermediary agent needed. Self-healing with 3 layers: fix, heal, retry loop.

Every AI Platform

Same 76 skills work across 10 AI platforms — Claude Code, Copilot CLI, VS Code Chat, Codex, Cursor, Windsurf, and more. Install once, use everywhere.

Model Tier Strategy

Opus for deep code review. Sonnet for execution. Haiku for simple lookups. Cost scales with complexity, not volume.

See all design decisions →

Explore

Dive Deeper

*Fully tested with Claude Code, Copilot CLI, and VS Code Chat. Built on open standards (OpenSkills, OpenPlugins, AGENTS.md, MCP) so all compatible platforms work out of the box. Some agents that haven’t fully adopted these standards yet are supported with platform-specific tweaks.

KAI by Dragan Filipovic