Example: social sentiment agent

This example walks through the full agent development lifecycle using a low-code agent that analyzes sentiment in social media comments. It covers every stage: scaffold, configure, add a tool, evaluate, publish, and wire as a sub-agent inside a social media manager agent.

Pré-requisitos

• The uipath-agents skill installed for your coding agent. See Install and set up.
• Uma sessão conectada (uip login).

Step 1: Scaffold the agent

Open your coding agent in an empty folder and ask:

Create a new sentiment analysis agent for social media inbound comments. The agent should accept a comment as input and return a sentiment classification (positive, neutral, or negative) with a confidence score.

The coding agent runs uip agent init and creates the full project structure, including agent.json, project.json, entry-points.json, and a default evaluation set.

Step 2: Configure prompts and schema

Review the generated agent.json and refine it conversationally:

Update the system prompt to classify sentiment on a 5-point scale (very negative, negative, neutral, positive, very positive) and add a brief reasoning field to the output schema.

The coding agent edits agent.json, runs uip agent refresh to regenerate derived artifacts, then uip agent validate to confirm the schemas are consistent.

Step 3: Add a tool

Connect the agent to an Integration Service connector:

Add a Slack connector tool so the agent can post the sentiment result to a Slack channel.

The coding agent creates a resource.json under resources/slack-post/, then runs uip agent refresh and uip agent validate to sync the tool into the agent entry points.

Step 4: Push to Studio Web and run locally

Push the agent to Studio Web so I can review the configuration.

The coding agent runs uip solution upload, which uploads the project. From Studio Web, run Debug to test the agent against a sample comment before building evaluation sets.

Step 5: Evaluate

Create an evaluation set called regression-tests with five test cases covering positive, negative, and neutral comments. Run it and show me the results.

The coding agent creates the test cases in evals/eval-sets/, runs the evaluation set, and returns the per-test-case scores. If a test case falls below threshold:

The third test case scored below 0.7 — update the system prompt to make the classification criteria more explicit and re-run.

Iterate until the evaluation set passes consistently.

Step 6: Publish

Publish the sentiment agent to my personal workspace.

The coding agent runs uip solution pack followed by uip solution publish. The agent is now a process in your Orchestrator personal workspace.

Step 7: Wire as a sub-agent

To use the sentiment agent as a tool inside a social media manager agent:

Create a social media manager agent that uses the sentiment analysis agent as a tool to handle a batch of inbound comments and generate a summary report.

The coding agent scaffolds a new manager agent project and registers the sentiment agent as a sub-agent tool using --type agent. Follow the same refresh → validate → upload → deploy steps above.

Result: A two-agent solution where the manager agent delegates sentiment classification to the sentiment sub-agent and compiles the results into a report.

Veja também

• Build agents with Coding Agents — overview and 8-step developer journey.
• Build low-code agents with Coding Agents — full reference for the low-code agent build loop.
• Build coded agents with Coding Agents — Python-based coded agents from a coding agent terminal.