Submission type
UiPath for Coding Agents
Name
RAVINDRA REDDY MADIREDDY
Industry category in which use case would best fit in (Select up to 2 industries)
Banking
Banking and financial services
Finance
Complexity level
Advanced
Summary (abstract)
Prompt2Provision is an agentic automation framework for disconnected enterprise applications using UiPath, Codex, Python orchestration, metadata-driven application profiles, reusable workflow generation, Orchestrator execution, and automated repair loops.
The solution addresses disconnected applications that do not expose APIs or standard connectors. It supports both JML operations and aggregation operations. Instead of manually building one-off automations for every application and operation, the framework analyzes a request, creates a governed transaction, checks whether a reusable workflow exists, builds one only when missing, validates it, approves it, registers it, executes it, and can repair/redeploy it when runtime issues occur.
The core principle is simple: generate reusable workflows once per application operation, then reuse them for future transactions with different runtime data.
• JML operations: create user, delete/disable user, modify access, assign or revoke roles/groups.
• Aggregation operations: account aggregation, group aggregation, entitlement discovery, disabled-user extraction, orphan-account identification, and reconciliation support.
• Enterprise execution: UiPath Studio/Assistant/Orchestrator execution with transaction data read from an external path exposed through an Orchestrator Asset.
• Governance: validator checks, human approval, workflow registry, deployment results, and repair history.
Detailed problem statement
Disconnected applications remain one of the most difficult areas in identity governance and enterprise automation. Many organizations use IGA platforms such as SailPoint for Joiner-Mover-Leaver processes, but many internal, legacy, vendor, and department-owned systems still do not expose APIs or standard connectors.
These applications are commonly operated through web portals, admin screens, export pages, CSV downloads, manual spreadsheets, or ticket-based human actions. That creates two major operational problems.
Problem 1 – Provisioning and Deprovisioning
• Access teams manually create users, disable users, modify roles, assign groups, and remove access.
• Each application has different fields, navigation paths, entitlement names, approval rules, selectors, and UI behavior.
• Traditional RPA can automate a single flow, but every new application and operation still requires custom workflow development, selector design, testing, packaging, deployment, and maintenance.
Problem 2 – Aggregation and Reconciliation
• Governance teams need accurate account, group, entitlement, disabled-user, and orphan-account data for audits and access certifications.
• Disconnected applications often require manual report exports, screen scraping, CSV downloads, and spreadsheet reconciliation.
• This creates delayed visibility, audit risk, inconsistent access reviews, and repeated manual effort across applications.
The challenge is therefore broader than one JML transaction. The real challenge is building a scalable and governed disconnected application lifecycle framework that supports provisioning, deprovisioning, access modification, account aggregation, group aggregation, entitlement discovery, and reconciliation support.
Detailed solution
Prompt2Provision separates build-time workflow generation from runtime workflow execution. It uses multiple agents with clear responsibilities. Runtime transactions are driven by governed JSON data, while workflows are generated, validated, approved, registered, deployed, and reused at the operation level.
Agent Responsibilities
Agent / Component Role Key Output
Agent 1 – Requirement Analysis Detects application and operation, loads app profile, extracts dynamic fields and entitlements, validates request, creates transaction JSON. automation_contract.json, pending_queue_item.json, external queue data file
Agent 4 – Router Checks workflow registry and decides BUILD_WORKFLOW, EXECUTE_WORKFLOW, or STOP/CLARIFICATION. routing_decision.json, Codex builder prompt
Agent 2 – Workflow Builder (Codex) Generates reusable UiPath XAML and artifacts when workflow is missing. Uses app profile and UI map. Can deploy and help repair selector issues. XAML workflows, README.md, test_data.json, workflow_arguments.json
Agent 3 – Validator Checks generated workflow artifacts, confirms no hardcoded runtime values, and enforces governance rules. validation_result.json
Human Approval Approves or rejects generated workflow before registration. approval_status.json
Agent 5 – Registry Update Registers approved workflow and version/status metadata for future reuse. workflow_registry.json
Agent 6 – Executor Bot Resolves workflow path, prepares execution metadata, and signals Orchestrator readiness. execution_result.json
Agent 7 – Orchestrator Run & Repair Starts Orchestrator job, monitors execution, detects failures, triggers Codex repair, packages, redeploys, and reruns if needed. orchestrator_job_result.json, repair_history.json, packages
Current End-to-End Pipeline
- Request Intake: A user submits a JML or aggregation request such as create_user, delete_user, aggregate_accounts, or aggregate_groups.
- Requirement Analysis: Agent 1 detects the target application and operation, loads the matching app profile, extracts required fields dynamically, validates entitlements, and creates a governed transaction JSON.
- External Runtime Data Copy: Agent 1 writes the transaction locally and copies it to an external runtime path such as C:\Prompt2ProvisionData\pending_queue_item.json. UiPath reads this path from the Orchestrator Asset Prompt2Provision_QueueDataPath.
- Routing: Agent 4 checks workflow_registry.json to determine whether an approved reusable workflow already exists.
- Build Path: If the workflow is missing, Agent 2 generates reusable UiPath artifacts using the queue transaction, application profile, and UI map.
- Validation: Agent 3 validates the generated artifacts, including no hardcoded runtime values and proper reusable structure.
- Human Approval: The pipeline asks for approval before registering a generated workflow.
- Registry Update: Agent 5 registers approved workflows with version, status, and metadata.
- Execution Path: Agent 6 resolves the workflow path and marks the transaction as ready for Orchestrator execution.
- Orchestrator Run & Repair: Agent 7 starts the Orchestrator job. If execution fails due to UI/selector/synchronization issues, it triggers Codex repair, repackages, redeploys, and reruns the job.
Workflow Reuse Strategy
The framework avoids generating workflows for every request. Instead, it generates one reusable workflow for each application-operation pair.
• broadriver.create_user → CreateUser.xaml
• broadriver.delete_user → DeleteUser.xaml
• broadriver.aggregate_accounts → AggregateAccounts.xaml
• broadriver.aggregate_groups → AggregateGroups.xaml
Future requests reuse the same approved workflow with different transaction data.
Deployment and Runtime Design
The latest design includes Orchestrator-oriented execution while preserving the local development pipeline. When a workflow already exists, the pipeline does not rebuild it. Agent 7 starts the existing process job directly in the configured Orchestrator folder. If the job fails and the error is classified as selector/UI/synchronization related, the repair loop updates the workflow, packages it, deploys the updated version, and reruns the job.
• Queue data is copied to an external local path: C:\Prompt2ProvisionData\pending_queue_item.json.
• UiPath reads the external file path from the Orchestrator Asset: Prompt2Provision_QueueDataPath.
• This avoids packaging runtime transaction data inside the UiPath project.
• Agent 7 handles Orchestrator run, package upload, process update, monitoring, and repair loop integration.
Autonomous Repair Loop
Prompt2Provision includes a self-healing loop for runtime UI failures. This is designed for selector changes, missing UI elements, timeouts, changed buttons, or synchronization issues. - Orchestrator job runs and returns success or failure.
- If failure occurs, Agent 7 captures logs and job output.
- The failure classifier checks whether the error is UI/selector/synchronization related.
- If repairable, Agent 7 creates a Codex repair prompt with the failed workflow, app profile, UI map, and error output.
- Codex updates selectors, waits, Check App State, or workflow logic without hardcoding runtime values.
- The project is repackaged, redeployed, and rerun.
- Repair history is saved for auditability.
Narrated video link (sample: https://bit.ly/4pvuNEL)
Expected impact of this automation
Prompt2Provision significantly reduces disconnected application automation effort by converting application metadata into reusable UiPath workflows.
Expected benefits include:
• Faster disconnected application onboarding
• Reduced manual RPA development effort
• Reusable operation-level workflows
• Faster provisioning/deprovisioning
• Improved governance visibility
• Reduced selector maintenance effort
• Faster aggregation onboarding
• Reduced operational overhead
• Standardized disconnected-app automation framework
• Human-in-the-loop governance for high-risk workflows
• Autonomous selector repair and redeployment
The framework creates a scalable pattern for disconnected application lifecycle automation.
UiPath products used (select up to 4 items)
UiPath Assistant
UiPath Orchestrator
UiPath Robots
UiPath Studio
Automation Applications
JML Operations • User onboarding • User offboarding/deactivation • Access modification • Role/group assignment • Access revocation Aggregation and Reconciliation Operations • Account aggregation • Group aggregation • Entitlement discovery • Disabled-user extraction • Orphan-account detection • Reconciliation reporting Automation Engineering and Governance • Reusable workflow generation • Dynamic workflow routing • Workflow validation • Human approval • Registry management • Package/deploy/run lifecycle • Autonomous runtime repair
Integration with external technologies
UiPath Technologies • UiPath Studio • UiPath Assistant • UiPath UIAutomation modern activities • UiPath coded agents / coding-agent-assisted development • UiPath Orchestrator processes, jobs, packages, and assets • UiPath CLI / uip skills for package, publish, update process, and start job actions Coding and Metadata Technologies • Codex for workflow generation and repair • Python for orchestration agents • JSON application profiles • JSON UI maps • JSON transaction contracts • Workflow registry metadata • Repair history and execution result artifacts
