🔍

UiPath Coding Agents
Implementation Guide

Enable AI coding agents — Claude Code, Cursor, GitHub Copilot, Gemini CLI, Codex, OpenCode — to build, operate, and troubleshoot UiPath automations through natural language using the uip CLI.

🗓 June 2026 📦 UiPath CLI — Standalone Latest 🎯 Product Support | Architects | Developers

1. What Are UiPath Coding Agents?

UiPath Coding Agents connects AI coding agents to the UiPath platform through the uip CLI. The agent runs uip commands on your machine, using the session established by uip login. It inherits your permissions exactly.

The key concept is Skills — instruction bundles that teach the agent when to pack a Solution, how to chain publish → deploy run, when to wait for a job, and how to triage a faulted RPA process.

🔨

Build & Deploy

Scaffold, validate, pack, publish, and deploy UiPath solutions and agents from natural-language asks.

⚙️

Operate

Drive any Orchestrator task — folders, jobs, queues, assets, audit logs — through plain English.

🔍

Troubleshoot

Conversational root-cause analysis for RPA failures, Orchestrator issues, and Maestro process faults.

ℹ️
Skills are not tools. Tools extend the CLI with new commands; skills extend the agent with UiPath-specific knowledge. You do not need to install tools before installing skills — the agent can auto-install them on first use.

2. Prerequisites Checklist

Verify all of the following before starting the installation.

  • @uipath/cli installed on the machine — npm install -g @uipath/cli
  • Node.js LTS installed (required for npm-based install)
  • Valid UiPath account (Cloud or on-premises) with appropriate permissions
  • At least one supported coding agent installed (see Section 4)
  • Outbound HTTPS to cloud.uipath.com and GitHub (for skill catalog)
  • uip login session established — must run BEFORE skills install

Required Orchestrator Permissions

The coding agent inherits your session permissions exactly. Ensure your account has:

PermissionRequired For
Read access to Orchestrator foldersAgent inspection of jobs, assets, queues
Write access to target foldersAgent deployment, resource management
Package upload rightsPublishing NuGet packages
Role assignment rightsManaging user-folder assignments

3. Installation — Step by Step

1

Install UiPath CLI

Install via npm. The package is @uipath/cli, binary is uip.

npm install -g @uipath/cli
uip --version
2

Authenticate with UiPath

Run uip login once per machine. All agent commands run under this session.

# Interactive login (opens browser OAuth)
uip login

# Service identity / External App (for CI/CD)
uip login --client-id <app-id> --client-secret <secret> --tenant <name>

# Check session status
uip login status
3

Install Your Coding Agent

Each agent has its own installer. Quick commands:

# Claude Code
curl -fsSL https://claude.ai/install.sh | bash

# Gemini CLI — see github.com/google-gemini/gemini-cli for install steps
# OpenCode — see opencode.ai for install steps

# Cursor / GitHub Copilot / Codex — download from their respective websites
4

Install UiPath Skills

Install skills into your coding agent. You pick the agent; all UiPath skills install at once.

# Claude Code (global only)
uip skills install --agent claude

# Cursor — global
uip skills install --agent cursor

# Cursor — project-local (tracked with repo)
uip skills install --agent cursor --local

# Other agents
uip skills install --agent copilot
uip skills install --agent gemini
uip skills install --agent codex
uip skills install --agent opencode

# Interactive — choose agent(s) from a checkbox
uip skills install
ℹ️
Claude Code supports global scope only. Using --local with --agent claude produces a ValidationError.
5

Verify the Integration

Open your coding agent and ask a UiPath-specific task. The agent should propose correct uip command chains.

AgentVerification AskExpected Response
Claude Code"Pack this Solution and deploy it to the Shared folder"Proposes: uip solution pack → uip solution publish → uip solution deploy run
Cursor"Create a new UiPath Agent project and pack it"Proposes: uip agent init → validate → pack
GitHub CopilotAsk any UiPath task in ChatCites the installed skill rules
Gemini CLIAsk for a UiPath deployment taskProposes correct uip commands
Codex / OpenCodeAsk for a UiPath taskFollows the skill recipe

If Claude Code doesn't respond correctly, restart it to reload plugins, then retry.

4. Supported Coding Agents

All agents share the same three prerequisites: uip CLI installed, a logged-in session, and the coding agent installed.

🤖 Claude Code
--agent claude
📍 Global only
Skills install via the Claude Code plugin system. User-scoped — --local not supported.
⚡ Cursor
--agent cursor
📍 Global or Local
Skills at .cursor/skills/<skill>/. Loads per-workspace.
🐙 GitHub Copilot
--agent copilot
📍 Global or Local
Skills at .github/skills/<skill>/. Works in VS Code, JetBrains, Visual Studio.
💎 Gemini CLI
--agent gemini
📍 Global or Local
Skills at .gemini/skills/<skill>/.
🧠 Codex
--agent codex
📍 Global or Local
Skills at .agents/skills/<skill>/ — open Agent Skills standard directory.
🌐 OpenCode
--agent opencode
📍 Global or Local
Global: ~/.config/opencode/skills/. Local: .opencode/skills/.

5. Understanding Skills

ConceptTools (uip tools)Skills (uip skills)
What they doExtend the CLI with new commandsExtend agents with UiPath knowledge
Who uses themThe uip CLI itselfAI coding agents (Claude, Cursor, etc.)
Exampleuip rpa pack, uip or jobs listGuides agent to the right command sequence
Install commanduip tools install ...uip skills install --agent <name>

Skills Lifecycle Commands

# Install
uip skills install --agent <name>
uip skills install --agent <name> --local

# Update (also removes skills dropped from catalog)
uip skills update --agent <name>

# Uninstall
uip skills uninstall --agent <name>

# After every CLI upgrade — run BOTH:
uip skills update --agent <name>
uip tools update

Global vs. Local Scope

ScopeFlagLocationUse Case
Global(default)User's agent profileAvailable in every project on the machine
Local--localCurrent directoryPin skills to a specific version; ship to all contributors

Local installs automatically add the skill store directory to .gitignore so cached bundles are not committed.

6. Operating the UiPath Platform

A coding agent with skills installed can drive any operational task through plain English. It runs uip against your logged-in session, scoped to the folder context you provide.

⚠️
Read vs. Write: The agent runs read operations (list, get, logs) autonomously. For write operations (create, update, delete), it must confirm with you first. Best practice: "Clean up unused packages, but show me the list first."

Full Capabilities

AreaWhat the Agent Can DoExample Ask
Folders & AccessList, create, nest, move, delete folders; assign users and roles"Add <alice@x.com> to Finance as Automation User"
Users, Roles & PermissionsInspect role grants, create/edit/delete roles, list permissions, see who has access to what"Which folders does <bob@x.com> have access to, and with what roles?"
LicensesInspect allocations across users and robots"Add 2 more Unattended runtime licenses to prod-worker-01"
Machines & RuntimesList machines, inspect runtimes available in a folder"What runtimes are available in Production?"
SessionsList active robot sessions, see which machine each is on, disconnect stuck sessions"Show all running unattended sessions in Production and how long they've been active"
Processes & PackagesUpload packages, list versions, create or update process releases"Publish version 2.3.0 of InvoiceBot to Finance"
SolutionsPublish, upload, deploy, activate; list deployed versions across tenants and folders"Deploy v1.4 of the invoice processing solution to Accounting in the staging tenant"
JobsStart, stop, monitor, fetch execution logs and attachments"List every faulted job from yesterday with error messages"
AssetsList, get, create, update, delete assets across folders"Copy all assets from folder A to folder B"
Credential StoresList, create, update, delete vault integrations; rotate referenced credentials"Which assets in Production still reference the old CyberArk store?"
Queues & Queue ItemsInspect queue depth, push items, bulk-update statuses"Which queues have more than 100 NewItems right now?"
Storage Buckets & FilesCreate buckets, upload, download, list, delete files"Download files from reports/ modified in last 7 days"
LibrariesPublish, list versions, manage releases"Show all versions of Email.Utilities library"
TriggersCreate time, queue, and event triggers; manage calendars"Show last 10 runs of triggers for HR onboarding solution"
WebhooksCreate, list, test webhooks"List all active webhooks and events they listen to"
Integration ServiceList, test, refresh OAuth tokens; check connection health"Test every Salesforce connection in Production"
Audit LogsFilter by user, component, action, date range"Who deleted the Finance folder last month?"

Common Multi-Step Patterns

🔥

Triage Faulted Jobs

"Triage every faulted job in Production from last 24 hours. Group by error message. Show top 5 root causes." — Agent lists jobs, fetches logs, groups by error signature.

🚀

Promote Between Tenants

"Promote invoice processing from dev to staging. Deploy under Accounting folder." — Agent pulls from dev, switches session, deploys into staging (confirms each switch).

👤

Assign User + Machine

"Assign <charlie@x.com> as Automation User to the Accounting folder, and also assign the VM Machine." — Agent assigns the user with the role, then assigns the VM Machine so the user can run unattended jobs against it.

🔐

Audit External Role Grants

"Audit all roles for external users. Report which folders they can edit." — Agent lists users, filters domain, expands role grants per folder.

7. Troubleshooting with Coding Agents

Describe a failure in plain English — from any terminal including the UiPath Studio integrated terminal — and the Troubleshoot skill triages it, ranks causes, tests hypotheses against live platform data, and returns a root cause with a concrete fix.

How to Start a Diagnostic Session

# Explicit skill invocation
/uipath-troubleshoot

# Or just describe the problem naturally:
My job abc-123 failed this morning, what happened?
Why is InvoiceBot not picking up queue items?
The approval process faulted on step 3, diagnose it.

Four-Stage Diagnostic Flow

1
Triage

Classify the failure scope. Asks a clarifying question if ambiguous.

2
Hypotheses

Ranked list of possible causes from the domain playbook.

3
Test

Fetches logs, traces, queue items, audit events via uip to confirm/eliminate.

4
Resolution

Root cause + event chain + concrete fix. Asks approval before changing anything.

Diagnosable Domains

DomainWhat You Can DiagnoseExample Ask
RPASelectorNotFoundException, TimeoutException, queue item failures, selector mismatches, Healing Agent data analysis"Why did InvoiceBot fail this morning?"
OrchestratorStuck/faulted jobs, permission and role issues, deployment and publishing errors, queue configuration problems"Why is my robot not picking up items from Finance?"
MaestroFaulted process instances, step-level failures, incidents, multi-domain failure chains (agents + RPA)"Why did the approval process fault on step 3?"

Tips for Effective Sessions

TipWhy It Helps
Be specific — include job IDs, process names, exact errorsSkips the clarifying round; agent can go straight to evidence
Say "that's not it" or "dig deeper" if first answer missesThe skill adjusts and follows a different hypothesis chain
Redirect if needed: "The issue is not permissions; job ran fine yesterday"Prevents the skill from following the wrong thread
Share XAML, project.json, config files when askedSharply improves diagnosis quality for code-level failures
Match expectations to your permissionsThe skill uses your uip session — if you can't view a log, neither can it

8. Best Practices

Outcome-First Prompting

✅ Do This (Outcome-First)❌ Not This (Command-First)
"Deploy this Solution to production""Run uip solution pack then uip solution publish then..."
"Triage faulted jobs from last 24h in Production""Run uip or jobs list --state Faulted --folder Production"
"Create a new UiPath Agent project and validate it""Run uip agent init, then uip agent validate"
"What runtimes are available in Finance folder?""Run uip or machines list --folder Finance"

When the Agent Gets Stuck

If the agent cannot determine the correct command, use this recovery pattern:

# Run the help command and paste output into the agent
uip <command> --help

The built-in help is authoritative. Agents recover immediately once given the help output.

Session Identity

# Personal work — agent acts as you
uip login

# CI/CD or service identity (cleaner audit trail)
uip login --client-id <id> --client-secret <secret> --tenant <name>

MCP — Alternative Path

For MCP-aware clients not on the supported-agents list, the CLI ships an MCP server exposing a run_command tool:

uip mcp

This is not the primary path — skills are. Use MCP only for specialized integrations that already speak MCP.

9. Command Reference

Authentication

uip login # Interactive OAuth (browser)
uip login --client-id <id> --client-secret <s> --tenant <t> # External App
uip login status # Check session
uip logout # Revoke session

Skills

uip skills install --agent claude
uip skills install --agent cursor
uip skills install --agent cursor --local
uip skills install --agent copilot
uip skills install --agent gemini
uip skills install --agent codex
uip skills install --agent opencode
uip skills install # Interactive
uip skills update --agent <name>
uip skills uninstall --agent <name>

Solution Lifecycle

uip solution new
uip solution pack
uip solution publish
uip solution upload
uip solution deploy run
uip solution bundle

Agent Lifecycle

uip agent init
uip agent config
uip agent validate
uip agent pack
uip agent publish
uip agent deploy
uip agent run
uip agent eval

Orchestrator Operations (uip or)

uip or jobs list
uip or jobs list --state Faulted --folder <name>
uip or jobs start --process <name>
uip or jobs stop --id <job-id>
uip or folders list
uip or folders create --name <name>
uip or users list
uip or roles list
uip or licenses list
uip or machines list
uip or sessions list
uip or packages list
uip or packages upload --file <pkg>
uip or processes list
uip or audit-logs list
uip or credential-stores list

Resource Management (uip resource)

uip resource assets list
uip resource assets create
uip resource assets update
uip resource assets delete --name <name>
uip resource queues list
uip resource queue-items list
uip resource buckets list
uip resource bucket-files list
uip resource libraries list
uip resource triggers list
uip resource webhooks list

Global Options

uip <command> --help # Always authoritative
uip <command> --output json
uip <command> --output yaml
uip <command> --output table # default
uip <command> --tenant <name>
uip <command> --folder <name>

10. Common Issues & Fixes

SymptomLikely CauseFix
Agent doesn't recognize UiPath tasksSkills not loadedFor Claude Code: restart the app. For others: verify skill files exist at expected path. Re-run uip skills install --agent <name>
--local flag causes ValidationError with Claude CodeClaude Code is global-onlyRemove the --local flag.
uip: command not foundnpm global bin not in PATHRun npm bin -g to find the path. Add it to ~/.bashrc or system PATH.
401 Unauthorized on uip commandsSession expiredRun uip login to re-authenticate. Verify tenant name and region.
Agent proposes wrong command chainSkills outdatedRun uip skills update --agent <name> and uip tools update
Agent stuck on wrong flagSkills don't cover specific flagRun uip <command> --help and paste the output into the conversation.
Skill files missing after installInstall path mismatchRe-run install — it prints the destination path. Verify that path exists on disk.
Permission denied during deploymentInsufficient Orchestrator permissionsCheck: uip or roles list --folder <name>. Ensure Write + Deploy access on target folder.

11. Security & Compliance

⚠️
Disclaimer: The Coding Agent does not validate the compliance of generated code, operational commands it executes, or troubleshooting actions it recommends. You are responsible for reviewing all outputs before applying them to production.

Responsible Use Checklist

  • Authenticate with a dedicated minimal-permission token — not your full personal access
  • Use External App sessions for CI/CD and scheduled agent tasks
  • Configure agent guardrails to limit scope and prevent unexpected behavior
  • Run in attended mode for any production operational tasks
  • Verify personal data processing complies with your organization's data policies
  • Maintain human oversight at all approval checkpoints
  • Do NOT use production credentials for testing
  • Review audit logs after each agent-driven deployment: uip or audit-logs list

12. Resources & Documentation

Official Documentation

Support

🆘

Support Portal

support.uipath.com

💬

Community Forum

forum.uipath.com

🎓

Academy

academy.uipath.com