AWS quick documentation change
Summary
Expanded documentation for custom agent configuration, including detailed mode descriptions, prompt writing best practices, action definitions, Bedrock model integration instructions, and testing procedures
Security assessment
The changes primarily focus on feature documentation improvements and operational guidance. While there are references to IAM roles and model security prerequisites, these are standard AWS configuration requirements rather than addressing specific vulnerabilities or adding new security controls. The Custom Mode section mentions billing implications but not security implications.
Diff
diff --git a/quick/latest/userguide/actions-agents.md b/quick/latest/userguide/actions-agents.md index 0d57d7274..6d1c8c881 100644 --- a//quick/latest/userguide/actions-agents.md +++ b//quick/latest/userguide/actions-agents.md @@ -171 +171 @@ Custom agent is an intelligent action that processes natural language inputs to -Title + * **Title** : Name of the step/custom agent @@ -172,0 +173 @@ Title + * **Mode** : A mode defines how the agent operates based on your use case. The three available modes are: Fast, Pro, and Custom. Fast is best for simple tasks like summarization, classification, and high-volume automations, and Pro is ideal for complex tasks that involve reasoning and orchestration of multiple tools or actions. Fast and Pro are fully managed modes that require no pre-setup needed in advance. In Custom Mode, you'll need a Bedrock runtime connector and can select the model you want to use (Explained below). This is ideal when you already have a prompt fine-tuned for a particular Bedrock model, specifically need a particular Bedrock model for the Agent, or want to include your own custom or fine-tuned model hosted on Bedrock. In Custom Mode, since you bring your own model from Bedrock via an integration, model inference is billed separately to the account associated with that Bedrock integration. @@ -174 +175 @@ Title -Name of the step/custom agent + * **Instructions** : In this field you write the prompt for the agent in natural language. Best practices while writing the prompt: @@ -176 +177 @@ Name of the step/custom agent -Mode + * Be clear and explicit about what you want. @@ -177,0 +179 @@ Mode + * Structure the prompt. Start with mentioning the 'Task' or 'Role' first and then 'Instructions' to achieve the task with numbered steps @@ -179 +181 @@ Mode -A mode defines how the Agent operates based on your use case. The three available modes are: Fast, Pro, and Custom. Fast is best for simple tasks like summarization, classification, and high-volume automations, and Pro is ideal for complex tasks that involve reasoning and orchestration of multiple tools or actions. Fast and Pro are fully managed modes that require no pre-setup needed in advance. In Custom Mode, you'll need an Bedrock Converse connector and can select the model you want to use. This is ideal if you already have a prompt fine-tuned for a particular Bedrock model, specifically need a particular Bedrock model for the Agent, or want to include your own custom or fine-tuned model hosted on Bedrock. In Custom Mode, since you bring your own model from Bedrock via a connector, model inference is billed separately to the account associated with that Bedrock connector. + * To improve tool-call accuracy and guide the Agent, clearly specify in the prompt which tool to use at each step, if applicable. @@ -181 +183 @@ A mode defines how the Agent operates based on your use case. The three availabl -Instructions + * Specify length requirements (e.g., less that 100 words) or output format (e.g., date in MM/DD/YY format) clearly @@ -182,0 +185 @@ Instructions + * Wrap the text in triple quotes (""") to write multiline prompts. For example: @@ -184 +187,2 @@ Instructions -In this field you write the prompt for the agent in natural language. Best practices while writing the prompt: + """You are content summarization agent. + Summarize the last two paragraphs of the provided text, focusing only on the main conclusion.""" @@ -186 +190 @@ In this field you write the prompt for the agent in natural language. Best pract - * Be clear and explicit about what you want. + * **Actions** : Action is a tool that enables the AI agent to interact with external systems or perform specific tasks. This is optional. You can run the custom agent without any actions. Below are the different actions which can be used in the custom agent @@ -188 +192 @@ In this field you write the prompt for the agent in natural language. Best pract - * Structure the prompt. Start with mentioning the 'Task' or 'Role' first and then 'Instructions' to achieve the task with numbered steps + * **General Actions** @@ -190 +194 @@ In this field you write the prompt for the agent in natural language. Best pract - * To improve tool-call accuracy and guide the Agent, clearly specify in the prompt which tool to use at each step, if applicable. + * **Create user task** \- If enabled, this tool allows the Agent to trigger a Human-in-the-Loop (HITL) task whenever it gets stuck and needs assistance during execution. The Agent pauses and waits for human input. The HITL task is visible in the task center. For best results, the author can specify in the prompt exactly when the Agent should invoke HITL. This is selected by default. The automation runs until the task is finished. @@ -192 +196 @@ In this field you write the prompt for the agent in natural language. Best pract - * Specify length requirements (e.g., less that 100 words) or output format (e.g., date in MM/DD/YY format) clearly + * **Code** \- The Code action generates and executes python code within a restricted python environment, same as code actions, to solve tasks involving calculations, data manipulation, and file processing. Unlike code generators, it actively creates and runs scripts to accomplish objectives, working with Excel, PDF files, various data formats and available integrations @@ -193,0 +198 @@ In this field you write the prompt for the agent in natural language. Best pract + * **Key Capabilities:** @@ -194,0 +200 @@ In this field you write the prompt for the agent in natural language. Best pract + * **File Operations** : Process multi-tab Excel files, extract content, perform date calculations, apply conditional formatting, and upload results to S3 @@ -195,0 +202 @@ In this field you write the prompt for the agent in natural language. Best pract + * **Data Transformation** : Convert between JSON and table formats, transpose data, rename columns, and join tables @@ -197 +204 @@ In this field you write the prompt for the agent in natural language. Best pract -Wrap the text in triple quotes (""") to write multiline prompts. For example: + * **Advanced Computations** : Generate numerical sequences and perform automated validation @@ -198,0 +206 @@ Wrap the text in triple quotes (""") to write multiline prompts. For example: + * **Integrations** : If you have added specific integration actions — such as Salesforce, MS Exchange, or Bedrock—to your automation group, their corresponding actions appear here to be use in the custom agent. The author can then select the relevant actions to use as tools for the agent. @@ -200,2 +208 @@ Wrap the text in triple quotes (""") to write multiline prompts. For example: - """You are content summarization agent. - Summarize the last two paragraphs of the provided text, focusing only on the main conclusion.""" +List of integrations which can be used as tools/actions in the custom agent @@ -203 +210 @@ Wrap the text in triple quotes (""") to write multiline prompts. For example: -Tools (Optional) + * Amazon S3 @@ -204,0 +212 @@ Tools (Optional) + * Amazon Bedrock Data automation @@ -206 +214 @@ Tools (Optional) -A tool enables the AI agent to interact with external systems or perform specific tasks + * Amazon Comprehend @@ -208 +216 @@ A tool enables the AI agent to interact with external systems or perform specifi -**General tools** + * Amazon Textract @@ -210 +218 @@ A tool enables the AI agent to interact with external systems or perform specifi -**Create user task** + * Custom REST API @@ -212 +220 @@ A tool enables the AI agent to interact with external systems or perform specifi -If enabled, this tool allows the Agent to trigger a Human-in-the-Loop (HITL) task whenever it gets stuck and needs assistance during execution. The Agent will pause and wait for human input, then resume once the required information is provided. The HITL task will be visible in the task center. For best results, the author can specify in the prompt exactly when the Agent should invoke HITL. + * Custom MCP connector @@ -214 +222 @@ If enabled, this tool allows the Agent to trigger a Human-in-the-Loop (HITL) tas -**Integrations** + * Microsoft Outlook @@ -216 +224 @@ If enabled, this tool allows the Agent to trigger a Human-in-the-Loop (HITL) tas -If you've added specific connectors—such as Salesforce, MS Exchange, or Bedrock—to your automation group, their corresponding actions will appear here. The author can then select the relevant actions to use as tools for the Agent. For optimal performance, it's recommended to limit the Agent to 3–5 tools. + * Salesforce @@ -218 +226,39 @@ If you've added specific connectors—such as Salesforce, MS Exchange, or Bedroc -Structured Output (optional) + * **Structured Output (optional)** + +Configure your AI agent to return structured JSON output that downstream steps can process. This feature is ideal for text summarization, report generation, data transformation, and extracting statistics from unstructured content. This is an optional field. If you do not define structured output, the agent returns output in natural language by default. Use structured output when your output has a defined structure, such as a list, data table, or JSON. + +###### Note + +The structured output configuration for Custom agents follows the same format as UI agents. Refer to the UI agent structured output section for detailed configuration instructions. + + * **Agent response** : Name of the variable to assign the output of the agent. The response follow your structured output format in a JSON schema if defined, otherwise is a free-form text. + + + + +### Using Custom Models in Custom Agent (Bring your own bedrock model) + +Integrate your desired or custom fine-tuned models hosted in AWS Bedrock with Quick Suite automation workflows. + +Before you begin, ensure you have the following: + + * A fine-tuned model deployed and accessible in AWS Bedrock + + * Quick Suite Admin access for creating connectors + + * An IAM role with Bedrock permissions for invoking models + + * Your model ID (for example, `us.anthropic.claude-3-5-sonnet-20241022-v2:0`) + + + + +**Step 1:** Create a Bedrock Runtime Action integration by following the detailed instructions in [AWS service action connectors](./builtin-services-integration.html) + +**Step 2:** Set Up Your Automation Group + +Create an automation group and connect the integration: + + * **Create an automation group** \- Follow the detailed instructions in [Setup tasks](./getting-started-quick-automate.html#automate-setup-tasks) + + * **Configure integrations** \- Follow the detailed instructions in [Setup tasks](./getting-started-quick-automate.html#automate-setup-tasks) @@ -219,0 +266 @@ Structured Output (optional) + * Once configured, the connector appears in your available assets list @@ -221 +267,0 @@ Structured Output (optional) -Configure your AI agent to return structured JSON output that can be easily processed in subsequent steps. This feature is ideal for text summarization, report generation, data transformation, and extracting statistics from unstructured content. This is an optional field. If you dont define structured output, the agent will by default return output in natural language. @@ -223 +269,26 @@ Configure your AI agent to return structured JSON output that can be easily proc -Agent response: Name of the variable to assign the output of this operation + + +**Step 3:** Configure a Custom Agent + +Add and configure a custom agent to use your fine-tuned model: + + * Within your automation workflow, add a custom agent + + * Configure the following agent settings: + + * **Agent Title** : Enter a descriptive name for your agent + + * **Instructions** : Enter custom prompts tailored to your use case + + * **Mode** : Select Custom + + * **Connector** : Choose your Bedrock Runtime connector (required when Custom mode is selected) + + * **Custom Model** : Enter your model ID (for example, `us.anthropic.claude-3-5-sonnet-20241022-v2:0`) - required when Custom mode is selected + + + + +**Next Steps** + +Once configured, your custom agent uses the fine-tuned model to process requests according to the instructions you provided. You can now incorporate this agent into your Quick Automate workflows. @@ -227 +298 @@ Agent response: Name of the variable to assign the output of this operation -The structured output configuration for Custom agents follows the same format as UI agents. Refer to the UI agent structured output section for detailed configuration instructions. +Ensure your model ID is correctly formatted and matches the model deployed in your AWS Bedrock account. You can find your model ID in the AWS Bedrock console under your provisioned models. @@ -231 +302,45 @@ The structured output configuration for Custom agents follows the same format as -Users can test the agent independently of the full automation to validate behavior, debug prompts, and iterate faster. +Custom agent testing enables you to test individual agents independently from the complete automation workflow. This capability helps you validate agent behavior, debug prompts, and iterate more efficiently without executing the entire workflow. + +#### Prerequisites + + * An automation workflow with at least one configured custom agent + + * Appropriate permissions to run automations in your workspace + + + + +#### Start a test + + * In the workflow canvas, hover over the agent card you want to test + + * Choose the **Unit test** button that appears at the top of the card + + * In the variable collection window that opens, review the automatically detected variables from your agent's prompt + + * The prompt preview displays all detected variables with highlighting + + * Enter a value for each variable + + * Values must use valid expression syntax + + * If a value contains invalid syntax, an error message appears and prevents test execution + + + + +#### Monitor test execution + +During test execution, you can monitor progress in the audit panel on the right side of the screen. The test skips all preceding workflow steps and executes only the selected agent. You get the same logging experience as a full workflow run. + +#### Review test results + +After the test completes, review the following information in the Test panel: + + * Metrics Card (Monitor Tab at the top of the Test panel) + + * Total execution time + + * Number of tools used