Claude Code, Cursor, Codex, Gemini CLI — your team is already using them. Today we’re making them work for UiPath projects too.
Why
Coding agents could already read XAML. They couldn’t build, troubleshoot, or manage a project like a UiPath developer would. No PascalCase variables, no in_/out_/io_ argument prefixes, no Try/Catch around RetryScope, hardcoded values instead of Config.xlsx. The output ran, but it didn’t ship.
We wanted to fix that without forcing you to switch agents. The goal: any coding agent can scaffold, edit, and review a UiPath automation that follows the same conventions a senior RPA developer would — with Studio still the place you visualize, debug, and govern what gets shipped.
How
The uip CLI installs the UiPath toolset into your agent. It wires up the MCP servers, drops the skill definitions in the right place, and configures the agent to talk to Studio, Orchestrator, and your local project.
​ npm install -g @uipath/cli ​
​ uip skills install ​
Pick your agent — they all work on day one.
Under the hood:
@uipath/cli— the npm package that installs the toolset and configures your agentUiPath/skills— the public skills repo with the instructions, examples, and patterns.
Studio or Studio Web stays the visual layer — every change the agent makes is something you can open, inspect, and debug there.
Heads up:
Autopilot is getting a massive upgrade soon and will use the same skills and tools we’re shipping for third-party coding agents. Same conventions, same prompts, same patterns — whether you’re using Claude Code, Cursor, Copilot, Gemini CLI, or Autopilot.
What you can build
Five things we’ve already shipped internally — not a ceiling:
- Invoice intake — REFramework process, SharePoint → Document Understanding → vendor validation → Slack
- Migration & refactoring — convert hardcoded values to
Config.xlsx, wrap external calls in Try/Catch + RetryScope, feed it a 5-year-old project, ask it to rewrite for Maestro. - Selector hardening — swap brittle selectors for Unified Target with fallbacks
- Test generation — generate a Test Manager suite covering happy path and edge cases
- PR review — wire your agent into GitHub so UiPath PRs get reviewed against your conventions before a human looks
Sky’s the limit. If you can describe it, the agent can probably scaffold it.
UI Automation works too
The thing that makes RPA RPA — driving real applications through their UI — is supported on day one:
- Build new UI automations end-to-end with the agent
- Targets stored in the Object Repository so they’re reusable across workflows and survive UI changes
- Strict selectors (find-one) for deterministic target creation — no fuzzy matches, no surprises at runtime
- All major selector technologies supported:
wnd,html,webctrl,aa,uia,java,sap
This is the part that’s hard for general-purpose coding agents — they don’t know what a selector is, let alone the difference between AA and UIA, or why you’d reach for the SAP scripting interface before falling back to image-based targeting. The skills bake that knowledge in.
Demo: building with UIAutomation
More than building
Building is the headline, but it’s just part of the story. Once an automation is in production, the same agent — same skills, same tools — handles the rest of the lifecycle:
Troubleshooting failed jobs
- Pull the failed queue items from Orchestrator, read the exception traces, group them by root cause, and propose fixes
- “Why did the invoice process fail at 3am?” → agent reads the logs, identifies the broken selector, opens a PR with the fix
- Replay a stuck transaction in a sandbox, watch where it diverges, patch the workflow
- Diagnose flaky test runs — is it the environment, the data, or a real regression?
Managing deployments
- Promote a package from dev → test → prod across tenants, with governance checks at each gate
- Bulk-update connections, assets, and queues when you migrate environments
- Audit what’s actually running in Orchestrator vs what’s in your repo — find the drift before it bites you
- Generate release notes from the Git diff so your CAB meeting takes 5 minutes instead of 50
Operating the estate
- “Which processes haven’t been touched in 6 months?” → agent queries Orchestrator and gives you the deprecation candidates
- Reconcile licenses against actual robot utilization
- Find duplicate processes across folders — agent reads the XAML, clusters by similarity, flags the candidates for consolidation
Same agent. Same uip skills install. The conventions and tools that work for building also work for everything that comes after.
Try it
​ npm install -g @uipath/cli ​
​ uip skills install ​
Pick your agent, point it at a project, tell it what to build.
If it worked: reply in this thread with what you built, which agent, and the prompt. Best examples end up in the docs.
If it broke: use /uipath-feedback from inside your agent. It captures the intent, and what the agent tried, and routes it to the team. No screenshots, no repro steps.
Future is here
