Create Habilidades de IA

Enhance your automations with the power of IA generativa by creating reusable Habilidades de IA in AI Agent Studio. Habilidades de IA enable Pro Developers configure prompts, parameters, file inputs, and model connections to produce contextual and consistent AI outputs that can be reused across multiple solutions.

Habilidades de IA are created by selecting a Conexões de modelo, configuring prompt logic, and making the skill available to Citizen Developers and Pro Developers for automation building.

Vendors Types Model Supported file types
Amazon Bedrock

Standard

 Amazon Nova Lite, Amazon Nova Pro png, jpeg, gif, webp
Anthropic Claude 3.5 Sonnet, Anthropic Claude 3.5 Sonnet v2, Anthropic Claude 3.7 Sonnet, Anthropic Claude 4 Sonnet, Anthropic Claude Sonnet 4.5 png, jpg, jpeg, gif, webp, csv, xls, xlsx, html, txt, md, doc, pdf, docx
Amazon Nova Micro -

Fine tuned

Fundamentado em base de conhecimento

Amazon Nova Lite, Amazon Nova Pro png, jpeg, gif, webp
Automation Anywhere

Standard

 Claude Sonnet 4, GPT-4o -
Grounded by Enterprise knowledge - -
Azure OpenAI

Standard

 GPT 3.5 Turbo, GPT-4, GPT-5, GPT-5 Mini -
GPT-4.1, GPT-4o, GPT-4o mini png, jpg, jpeg, gif, webp

Fine tuned

 GPT 3.5 Turbo, GPT-4, GPT-4 Turbo Preview -
GPT-4.1, GPT-4o, GPT-4o mini png, jpg, jpeg, gif, webp

Fundamentado em pesquisa de IA

GPT 3.5 Turbo, GPT-4 -
GPT-4.1, GPT-4o png, jpg, jpeg, gif, webp
Google Vertex AI Standard Anthropic Claude 3.5 Sonnet, Anthropic Claude 3.7 Sonnet, Anthropic Claude Sonnet 4 png, jpg, jpeg, gif, webp
Gemini 2.5 Flash, Gemini 2.5 Pro pdf, png, jpg, jpeg
Fine tuned Gemini 2.5 Flash, Gemini 2.5 Pro pdf, png, jpg, jpeg
Fundamentado em fonte de dados Gemini 2.5 Flash, Gemini 2.5 Pro pdf, png, jpg, jpeg
OpenAI Standard GPT 3.5 Turbo, GPT-4, GPT-4 Turbo Preview, GPT-5, GPT-5 Mini -
GPT-4.1, GPT-4o, GPT-4o mini png, jpg, jpeg, gif, webp
Fine tuned GPT 3.5 Turbo, GPT-4, GPT-4 Turbo Preview -
GPT-4.1, GPT-4o, GPT-4o mini png, jpg, jpeg, gif, webp
Note: Some models support multimodal inputs, enabling the Habilidade de IA to process images and documents in addition to text. These capabilities are available only for supported generative AI models, and vary based on the model vendor.

This topic walks you through the steps of creating Habilidades de IA.

Prerequisites

A Pro Developer requires these roles and permissions to create and test an Habilidade de IA.
  • Role: AAE_Basic, Pro Developer Custom role
  • Permission: Bot Creator
  • Setting: Administration > Settings > AI Data Management > Data logging settings must be enabled. The Automation Admin can enable this setting for the Pro Developer

See Funções e permissões

Other requirements:

As part of testing the Conexão de modelo, you would have to run the bot on your desktop. Hence ensure the Agente de bot is configured to your user. If you have to switch connection to a different Control Room, see: Alternar o registro do dispositivo entre instâncias da Control Room.

Procedure

  1. Create a new Habilidade de IA.
    1. Log in to the Control Room.
    2. Navigate to Automation > Create or click the ‘+’ icon and select AI Skill.
    3. Enter a name and description and then click Create & edit. Let us create an example with the name Patient Discharge Agent.
  2. Model connection: Choose a Conexão de modelo. At the top of the Habilidade de IA editor, click Choose... under Conexão de modelo to open the list of Conexões de modelo assigned to you.
    Once selected:

    • The Model connection name, vendor, and Description appear in the header.

    • Parameter fields such as Frequency Penalty, Presence Penalty, N, Max Tokens, Temperature, and Top P become editable.

    These defaults are tailored to the selected model but can be adjusted as needed.

  3. System prompt (Optional): Use this to define high-level context or behavioral guidelines. If the selected Conexão de modelo supports the Prompt do sistema, you will see a Prompt do sistema section. For more information on Prompt do sistema, see Prompt do sistema. For example:You are a Family Physician creating a comprehensive patient discharge plan.
  4. User prompt: Enter instructions the model. You reference string inputs using $variableName$ syntax. The user prompt supports multi-line text entry and the expanding text area as shown in the screenshot. For example: Create a verbose patient discharge plan... Use the following $patientdetails$ to create the heading…
  5. Prompt Input: (right side)
    String inputs:
    • Click Add String or + to add an input string. Enter a Key and Value.
    • Shows variables created for text-based inputs.
    • Each input includes a name, value, and actions (copy, delete).
    • Example: patientdetails.
    File inputs:
    • Click Add File or + to add a file string. Enter a Key and Value.

    • Choose a file from the Control Room repository or upload a new one. Supported file types include:

      • Images: .jpeg, .jpg, .png, .gif, .webp

      • Documents: .txt, .pdf, .docx, .csv

    Habilidade de IA parameters
    Note: Model-specific restrictions on file size, supported types, and maximum file count are automatically enforced.
  6. File prompt (optional): The File prompt section is displayed only when the selected model supports multimodal input. If you have added files, you will be able to select the files from the drop-down. It controls which file inputs (from the right-side file panel) are included in the final request sent to the model.
    Note: Files attached to an Habilidade de IA are not processed by Proteções de IA for data masking.

    Guardrails evaluate and mask text prompts only.

    File contents are passed directly to the model exactly as provided.

    Note: Multimodal input support (images and documents) is available only for select generative AI models. Supported models include Amazon Bedrock (Nova and Claude Sonnet variants), Azure OpenAI (GPT-4.1, GPT-4o, GPT-4o-mini), OpenAI (GPT-4.1 and GPT-4o), and Google Vertex AI (Gemini 2.5 and Claude Sonnet models). The maximum number of files and allowable file sizes depend on the selected model connection. Automation 360 enforces the vendor’s limitations automatically.
  7. Get reponse: Use Get reponse to preview the result using the selected Conexão de modelo. You can test your Habilidade de IA with the Proteções de IA. Use the Test with AI Guardrail option to simulate the impact of selected guardrails on your prompts and model responses. If a guardrail is chosen, the prompt will be scanned for toxicity and sensitive data. If no guardrails are selected, no guardrail evaluations will be performed on the prompts.
    Note: To use Test with AI Guardrail, the Proteções de IA must be enabled in the global admin settings and you will need View AI Guardrails and View Settings permissions. You can review the audit trail within Governança de IA for a detailed record of each test.

    When you are testing with a guardrail you will see the Toxicity for both Prompt do usuário and Response. To view the results of data masking, click view data masking execution. It will display the tokenized masked values of any sensitive data identified in the prompt.

    When an Habilidade de IA sends a prompt that violates a blocking rule defined in an active guardrail, the automation using that skill will stop execution with an error message and no response from LLM. The error message will typically indicate that the prompt has been blocked by a guardrail. For more information on setting up Toxicity rule, see Criar e gerenciar Proteções de IA.

  8. Data Logging: A toggle appears at the top-right of the screen: Data logging disabled. .

    When enabled: The toggle switches toData logging enabled. The system logs the prompt, response, and the model parameters and file metadata to AI > Governance > AI prompt log

    When disabled: Only minimal metadata is captured.

    Developers can decide how much information is logged based on the sensitivity of the content.

    Note: Governança de IA does not store the contents of uploaded files or its metadata.

Example:

Click on the ellipsis next to an Habilidade de IA to view and access actions available to the Pro Developer.

Habilidade de IA actions

Actions Function
Move Habilidade de IA Click on this action to select the location where you want to move the Habilidade de IA to, as per your folder permissions.
Check in the Habilidade de IA Lets you check in an Habilidade de IA to a different folder. The Status of the Habilidade de IA is updated and it displays the folder you checked in the Habilidade de IA. Example: if you checked in the Habilidade de IA to the Public folder, then the Status label displays Public.
View Habilidade de IA Opens the Habilidade de IA displaying details.
Edit Habilidade de IA Lets you edit the Habilidade de IA.
Note: The Pro Developer has this permission.
Copy Habilidade de IA Lets you copy an existing Habilidade de IA to create a new one with the same details. You have to provide a different name though, as there cannot be duplicates of an Habilidades de IA.
Note: The Pro Developer has this permission.
View references Lets you view references to other bots that are directly dependent on the current one where this Habilidade de IA has been used.
Delete Habilidade de IA Lets you delete an Habilidade de IA.
Note: The Pro Developer has this permission.

Next steps

Your next step would be to check-in the Habilidades de IA to make it available to Citizen Developers from the Habilidades de IA pacote.

Why would you check-in an Habilidade de IA?

After creating an Habilidade de IA, check it into the Public folder. This will let the Pro Developer, and Citizen Developer use it from the Habilidades de IA pacote in the production environment.

For reference, see:

A Task Bot, with one or multiple embedded Habilidades de IA can be added to a larger automation that would run a complete workflow scenario.

Note: When you create or test an Habilidade de IA in the AI Skills screen, the success or failure details along with the model responses can be viewed in Administration > Audit log and Administration > AI governance navigation. See Governança de IA

As the next step in your sequence of tasks, go to Usar Habilidades de IA em um Task Bot.