What's a Coding Agent?

Learn how coding agents work and what makes them useful for UiPath automation work.

A coding agent is an AI assistant that reads, writes, and reasons about code — not just autocompletes it. Instead of suggesting the next line, it understands a goal described in plain English, explores the codebase, makes decisions, and generates entire working components. You stay in control: you review what it builds, ask it to adjust, and iterate until it's right.

Several coding agents are available today, each with their own strengths:

• Claude Code — Anthropic's CLI-based agent, runs in your terminal, deeply integrated with the file system. This is the one used in this exercise.
• Codex CLI — OpenAI's command-line coding agent, similar terminal-first approach
• Cursor — an IDE built around AI, tight editor integration with inline generation and multi-file context
• Gemini CLI — Google's terminal agent, strong on reasoning and long context
• GitHub Copilot Agent — embedded in VS Code and JetBrains, well suited for in-editor workflows

What makes coding agents especially powerful for UiPath work is the skill layer: by loading domain-specific knowledge about UiPath activities, APIs, and project conventions, the agent can generate production-ready automation code — not generic boilerplate. That's exactly what the UiPath skills in this exercise do.

The Mental Model

Five concepts show up in every coding agent:

Concept	What it is
Tools	Built-in actions the agent can take: read a file, run a bash command, search the web, edit code, etc.
Skills	Domain knowledge bundles the agent loads when it sees a matching task. Like a playbook for one thing.
Commands	Manual shortcuts you type: /commit, /test, /deploy. Triggered by you, not the agent.
Hooks	Run automatically at lifecycle events: before an edit, after a commit, on session start. Used for guardrails.
Subagents	Specialised workers spawned by the main agent. Used to parallelise or isolate complex tasks.

Plugins

A plugin is a package that bundles any of these together — like a Marketplace app for your coding agent. The UiPath skills used in this exercise are plugins.

Tools, in practice

Claude Code's built-in tools are always available — no setup required:

Tool	What it does
Read	Read any file in your project
Edit / Write	Modify or create files
Bash	Execute shell commands
Glob / Grep	Search across the codebase
WebSearch / WebFetch	Look things up online
Task	Spawn a subagent

Extension Tools (MCP)

MCP (Model Context Protocol) is the standard way to give your coding agent new tools without changing the agent itself.

• GitHub MCP — read issues, open PRs
• Atlassian MCP — Jira, Confluence
• Slack MCP — read channels, post messages
• Custom UiPath MCP — call Orchestrator, Maestro

Skills, in practice

A skill is a folder — sitting in .claude/skills/ (or equivalent), with a SKILL.md inside.

Property	What it means
Discovered automatically	The agent reads each SKILL.md's description. When your prompt matches, it loads the full skill on demand.
More than just text	Skills can bundle reference docs, code patterns, helper scripts, even compiled binaries the agent can run.
Scoped permissions	A skill declares which tools it's allowed to use. Bash access can be tightly controlled.

What's in a SKILL.md file

Every skill is built around a single SKILL.md file. Here's what one looks like:

---
name: uipath-coded-agents
description: Scaffold, build, run, evaluate, and deploy UiPath coded agents...
allowed-tools: [Read, Edit, Bash(uip:*), Bash(uipath:*)]
---

# How to build a UiPath coded agent

When the user asks for a new coded agent:

1. Run `uipath new --framework langgraph`
2. Create agent.py following the pattern in references/
3. Add an evals/ folder with at least 5 test cases
4. Run `uipath eval` before deploying

## References
- references/sdk-methods.md
- references/langgraph-patterns.md

The file has four parts:

Part	What it does
Frontmatter	Declares the skill's name, description, and which tools it's allowed to use.
Discovery prompt	The description field is what the agent matches against your request — this is how the skill gets loaded automatically.
Instructions	The body of the file: a step-by-step prompt the agent follows when the skill is active.
References	Paths to files in the skill folder the agent can read for deeper context — API docs, code patterns, example projects.

Where UiPath fits in this picture

UiPath ships a plugin for Claude Code. It bundles skills, hooks, and references — giving the agent domain knowledge about UiPath products out of the box.

npm -g install @uipath/cli
uip skills install

The plugin contains three things:

What	Details
10+ skills	Domain knowledge for every UiPath product: uipath-agents, uipath-coded-apps, uipath-rpa, uipath-maestro-flow, uipath-platform, uipath-test, and more.
1 subagent	uipath-project-discovery-agent — auto-runs on first contact with a UiPath project and builds context for the agent.
Hooks & references	Session-start nudges, activity docs, and allowlists for safe read-only commands.

The UiPath skill catalog

The plugin includes skills for every part of the UiPath platform. Three of them drive the labs in this exercise:

Skill	Description
uipath-agents	Scaffold, build, run, evaluate, and deploy coded agents (LangGraph, LlamaIndex, OpenAI Agents)
uipath-rpa	Generate and edit RPA workflows (XAML) for Studio Desktop with a discovery-first approach
uipath-coded-apps	Push/pull TypeScript + React apps to Studio Web, pack & publish to Orchestrator
uipath-solution	Create and deploy solutions
uipath-maestro-flow	Create, validate, and debug UiPath Flow projects using the .flow JSON format
uipath-platform	Auth, Orchestrator management, solution lifecycle, Integration Service, CLI tools
uipath-test	Generate, run, and maintain automated test suites for UiPath projects

Skills used in this exercise

The three skills in bold — uipath-agents, uipath-rpa, and uipath-coded-apps — are the ones you'll use to build the loan underwriting pipeline.