Kratu — Azure DevOps Autopilot
Turn your Azure DevOps board into reviewed pull requests, using the AI coding agent you already use.
Kratu syncs your Bugs and Tasks, ranks them by priority, and has your coding CLI (Claude Code, OpenAI Codex, GitHub Copilot CLI, Gemini CLI, Cursor Agent, Aider, or your own) analyze each item, explain the root cause and blast radius, fix it in an isolated workspace, verify it with your tests, and open a pull request linked to the work item.
Kratu never merges, never pushes to protected branches, and never touches your own working tree. Every fix ends in a pull request for human review.
Why Kratu
- Your board in the sidebar. A ranked queue (Priority → Severity → oldest) with triage risk and status. Details open in one click.
- Bring your own agent. Kratu orchestrates; the agent edits. It detects installed CLIs automatically, including the Claude Code binary inside the Claude Code extension, and falls back to another agent if one isn't signed in.
- Explains before it edits. Every item gets a root cause, a plan, the files involved and a blast radius (low / medium / high). Anything above your limit waits for your approval.
- Safe by construction. Isolated git worktrees, blocked paths, secret scanning, size limits, protected branches, and credentials kept off command lines.
- Verified. Detects Node, .NET, Java (Maven/Gradle), Go, Rust, Python and Make. Checks run before and after each change, and new failures go back to the agent for repair.
- New or existing branch. You choose where each fix goes. Unattended runs always use a fresh branch per work item.
- Clock in. Kratu works through your queue during your working hours, in the foreground or as a background job (Windows Task Scheduler / cron).
- Clear errors. Every problem, from a missing agent to an expired token or a wrong project name, comes with the exact fix.
Requirements
- Git, and a repository whose work is tracked in Azure DevOps (Services or Server). Code can live in Azure Repos or GitHub.
- At least one AI coding CLI: Claude Code, Codex, Copilot CLI, Gemini CLI, Cursor Agent or Aider. Run Kratu: Show AI Coding Agents to see install commands.
- An Azure DevOps sign-in: the Azure CLI (
az login), Git Credential Manager, or a Personal Access Token (Work Items: Read & write, Code: Read & write).
Node.js is not required: Kratu runs on VS Code's built-in runtime.
Getting started
- Open your repository in VS Code and click the Kratu icon in the Activity Bar.
- Click Set up Kratu, paste your board URL, sign in, and choose your agent, policy and hours.
- Run Kratu: Check Setup to confirm everything is ready.
- Hover an item in the queue and press 🪄 Fix, or 🔍 Analyze for an explanation only.
- Press ▶ Clock in to let Kratu work through the queue.
The Get started with Kratu walkthrough (Help → Welcome) covers the same steps.
Commands
| Command |
What it does |
| Kratu: Set Up |
Connect board, sign-in, agent, policy and working hours |
| Kratu: Check Setup |
Readiness check with exact fixes (optionally tests agent sign-in) |
| Kratu: Analyze |
Root cause, plan and blast radius; no changes |
| Kratu: Fix and Raise Pull Request |
Analyze → fix → verify → branch → PR → update the work item |
| Kratu: Approve for Automatic Fix |
Let the next run fix an item that was held for approval |
| Kratu: Clock In / Clock Out |
Work the queue during your hours |
| Kratu: Run One Pass Now |
Process the top ready items once |
| Kratu: Run in Background During Working Hours |
Install the OS scheduler job |
| Kratu: Open Team Settings |
Edit .kratu/config.json (shared, contains no secrets) |
Settings
| Setting |
Default |
Description |
kratu.autoRefreshMinutes |
10 |
Refresh the queue while VS Code is open (0 = off) |
kratu.showStatusBar |
true |
Show queue and clock status in the status bar |
kratu.cliPath |
"" |
Use a separately installed Kratu CLI instead of the bundled one |
Project behaviour (filters, blast-radius limit, branch policy, verification commands, agent models, custom agents) lives in .kratu/config.json, shared by the VS Code extension and the kratu CLI.
Privacy
Kratu has no servers and collects no telemetry. It talks only to your Azure DevOps organization, your git remote and the AI agent CLI you choose, all under your own accounts. Tokens stay on your machine, and every run's transcript is kept locally in ~/.kratu/runs.
Feedback
Kratu is in preview. Please use the Q&A tab on this page for questions, bugs and ideas.
| |
