Orchestrating with Maestro Flow

Use the uipath-maestro-flow skill to generate a UiPath Flow process that wires the coded agent and the loan officer dashboard together and routes applications by risk band.

1. Open Your Project Folder in Your Coding Agent

Create an empty folder called {YourName}LoanUnderwritingProcess and open it in your coding agent (e.g. Claude Code, Codex):

mkdir {YourName}LoanUnderwritingProcess

2. Describe the Process

Give your coding agent a plain-English description of the full flow:

Build a UiPath Flow process using the uipath-maestro-flow skill called
{YourName}LoanUnderwritingProcess. It should:

1. Receive six inputs: loan_id, applicant_first_name, applicant_last_name,
   loan_amount, loan_term, income
2. Call the agent {YourName}LoanUnderwritingAgent (in the CodingAgentsILT
   folder), passing loan_id and storing the returned values as process
   variables: credit_score, risk_band, assessment_summary,
   monthly_repayment, debt_to_income_ratio, loan_to_market_ratio
3. Auto-approve the application and end the process if risk_band is "Low"
4. For Medium or High risk, open the app {yourname}loanofficerdashboard (in the
   CodingAgentsILT folder), passing all twelve inputs: LoanId,
   ApplicantFirstName, ApplicantLastName, LoanAmount, LoanTerm, YearlyIncome,
   CreditScore, RiskBand, AssessmentSummary, MonthlyRepayment,
   DebtToIncomeRatio, LoanToMarketRatio — and capturing officer_decision
5. End with Approved or Rejected based on officer_decision
6. Output final_decision (string — "Approved" or "Rejected") in all cases,
   whether the application was auto-approved or decided by the loan officer

Your coding agent reads the uipath-maestro-flow skill context and generates a complete project:

• {YourName}LoanUnderwritingProcess.flow — the flow definition with all nodes, edges, and variable mappings
• project.uiproj — project manifest

3. Review the Generated Flow

Open {YourName}LoanUnderwritingProcess.flow in your text editor or ask your coding agent to describe the structure. You should see:

• A Start node with six process input variables: loan_id, applicant_first_name, applicant_last_name, loan_amount, loan_term, income
• An Agent node calling {YourName}LoanUnderwritingAgent with loan_id as input and six output variables: credit_score, risk_band, assessment_summary, monthly_repayment, debt_to_income_ratio, loan_to_market_ratio
• A Condition node branching on risk_band == "Low"
• A Human Task node wired to {yourname}loanofficerdashboard with all twelve variables mapped as inputs (LoanId, ApplicantFirstName, ApplicantLastName, LoanAmount, LoanTerm, YearlyIncome, CreditScore, RiskBand, AssessmentSummary, MonthlyRepayment, DebtToIncomeRatio, LoanToMarketRatio) and officer_decision as output
• A second Condition node branching on officer_decision == "Approve"
• Three End nodes — Auto-Approved, Approved, Rejected

If any mapping or condition needs adjusting, ask your coding agent:

Update the risk band condition so the default path goes to the human review
node and the explicit condition branch handles the "Low" auto-approve case

Visualise the flow in Studio Web

You can import the generated .flow file into Studio Web to see a visual diagram of the process and test it interactively — without deploying to Orchestrator first. Open Studio Web, create a new Flow project, and import {YourName}LoanUnderwritingProcess.flow. The canvas will render all nodes, conditions, and connections exactly as the coding agent generated them.

4. Test the Generated Flow

Ask your coding agent to fetch a verified loan record from UiBank and generate valid test input for the flow:

Open and inspect the UiBank Swagger documentation:

https://uibank-api.uipath.com/explorer/swagger.json

Using only endpoints documented by the API, identify the read-only endpoint for
retrieving loan quotes. Make a GET request and select one loan quote that
currently exists in UiBank.

Use the verified loan record to generate test input for the
{YourName}LoanUnderwritingProcess Flow.

Requirements:
- Use only GET endpoints documented in the Swagger specification.
- Do not create, update, or delete any data.
- Do not invent or reuse an unverified loan_id.
- Verify that the selected loan record exists.
- Populate loan_amount, loan_term, and income from the same verified record.
- loan_term represents years and must be one of: 1, 3, 5, or 10.
- Map the record's yearly income to income.
- Use applicant names from the record when available. If the API does not provide
  applicant names, use the workshop-safe values John and Doe.
- Return exactly one valid JSON object.
- Do not include Markdown, explanations, comments, or a PowerShell command.

Required output format:

{
  "loan_id": "<VERIFIED_LOAN_ID>",
  "applicant_first_name": "<FIRST_NAME>",
  "applicant_last_name": "<LAST_NAME>",
  "loan_amount": <LOAN_AMOUNT>,
  "loan_term": <1_OR_3_OR_5_OR_10>,
  "income": <YEARLY_INCOME>
}

Your coding agent should return something like:

{
  "loan_id": "6a4203a1f4865600481c657b",
  "applicant_first_name": "John",
  "applicant_last_name": "Doe",
  "loan_amount": 15000,
  "loan_term": 3,
  "income": 48000
}

5. Validate and Pack

ManuallyUsing Your Coding Agent

Navigate to the parent folder of the generated .flow file:

cd "parent-folder-path/to/{YourName}LoanUnderwritingProcess"

Replace the path with the actual location on your machine.

Validate the flow before packing:

uip maestro flow validate {YourName}LoanUnderwritingProcess.flow

If you get a command not found error, the maestro tool may not be installed or on your PATH:

# 1. Install maestro-tool (only needed once)
npm install -g @uipath/maestro-tool --prefix "$HOME/.npm-global"

# 2. Add it to PATH (needed each new terminal, or add to ~/.zshrc permanently)
export PATH="$HOME/.npm-global/bin:$PATH"

# To avoid step 2 every time, run this once:
echo 'export PATH="$HOME/.npm-global/bin:$PATH"' >> ~/.zshrc

# 3. Validate
uip maestro flow validate {YourName}LoanUnderwritingProcess.flow

Fix any warnings, then pack:

uip maestro flow pack "<path-to-your-flow-folder>/{YourName}LoanUnderwritingProcess" /tmp/dist --version 1.0.0 --output json

For example:

uip maestro flow pack "/path/to/{YourName}LoanUnderwritingProcess/{YourName}LoanUnderwritingProcess" /tmp/dist --version 1.0.0 --output json

Upload the package to Orchestrator and create a process in the CodingAgentsILT folder:

uip or packages upload /tmp/dist/{YourName}LoanUnderwritingProcess.flow.Flow.1.0.0.nupkg --output json

For example:

uip or packages upload /tmp/dist/MadalinaLoanUnderwritingProcess.flow.Flow.1.0.0.nupkg --output json

If you get a command not found error, the maestro tool may not be installed or on your PATH:

# 1. Install maestro-tool (only needed once)
npm install -g @uipath/maestro-tool --prefix "$HOME/.npm-global"

# 2. Add it to PATH (needed each new terminal, or add to ~/.zshrc permanently)
export PATH="$HOME/.npm-global/bin:$PATH"

# To avoid step 2 every time, run this once:
echo 'export PATH="$HOME/.npm-global/bin:$PATH"' >> ~/.zshrc

# 3. Retry the upload
uip or packages upload /tmp/dist/{YourName}LoanUnderwritingProcess.flow.Flow.1.0.0.nupkg --output json

Then create the process:

uip or processes create --name "{YourName}LoanUnderwritingProcess" --package-key "{YourName}LoanUnderwritingProcess.flow.Flow" --package-version "1.0.0" --folder-key "c30345cd-5543-46a9-b42b-0354e60b4f15" --output json

For example:

uip or processes create --name "MadalinaLoanUnderwritingProcess" --package-key "MadalinaLoanUnderwritingProcess.flow.Flow" --package-version "1.0.0" --folder-key "c30345cd-5543-46a9-b42b-0354e60b4f15" --output json

Give this prompt to your coding agent — it validates, publishes, and deploys the flow for you:

Publish and deploy the {YourName}LoanUnderwritingProcess Maestro flow I just built.

1. Find the generated .flow file and validate it, fixing any warnings it reports.
2. Publish it to Orchestrator, into the CodingAgentsILT folder
   (folder-key c30345cd-5543-46a9-b42b-0354e60b4f15), as version 1.0.0.
3. Deploy it by creating a process from the published package in that same folder.
4. Report back the process key.

6. Test the Full Flow

Trigger the process from Orchestrator three times — once per applicant below. Each loan_id is a real, verified record in UiBank (the agent re-fetches and scores off this record, so the amount/term/income below match it exactly):

{"loan_id": "6a8dabd21feca3004834197a", "applicant_first_name": "Alice", "applicant_last_name": "Nguyen", "loan_amount": 25001, "loan_term": 3, "income": 80000}

{"loan_id": "6a8c0bb11feca300483416cb", "applicant_first_name": "Ben", "applicant_last_name": "Carter", "loan_amount": 120000, "loan_term": 10, "income": 30000}

{"loan_id": "6a8db0e11feca30048341981", "applicant_first_name": "Clara", "applicant_last_name": "Osei", "loan_amount": 12000, "loan_term": 10, "income": 1200}

These IDs point to live records in the shared UiBank sandbox — verified to exist at the time this lesson was written, and chosen so the debt-to-income ratio clearly separates each band: Alice ~10% (Low), Ben ~40% (Medium), Clara ~100% (High). If UiBank's data is ever reset, look up 3 fresh records with a similarly clear spread using the swagger lookup from step 4 — since scoring is LLM-driven, always re-run and confirm the resulting risk_band before relying on a new record.

Applicant	Expected path
Alice — low loan, solid income	Low risk → auto-approved, no human task
Ben — large loan, high term	Medium risk → human task opens in Action Center
Clara — young age, tight income	High risk → human task opens in Action Center

For Ben and Clara, open Action Center while the process is paused. The loan officer dashboard should appear with the credit score and risk band pre-filled. Submit a decision and watch the flow route to the matching end state.

Watch the execution trace

In Orchestrator, click into a running flow instance to see the live diagram with the active node highlighted. This is the fastest way to confirm routing is correct at each condition.