Here is the description of a template to define clear, effective instructions for ReadiAI Agents. Well-structured instructions help agents operate reliably, align with business logic, and generate consistent results.
Guided Instructions
Use this section to help you think through each part of your agent’s behavior. Each heading includes an explanation of what to consider when writing instructions.
Role
Define the domain expertise your agent should reflect.
You are an expert in (subject area).
Goal
Describe the task or outcome the agent is meant to support.
I need your help to (complete a task or solve a problem).
Input
Outline the types of information the agent needs to function.
Knowledge
Reference materials or static data that support agent decision-making:
- Documents (e.g., policies, procedures, internal guides)
- Websites
- A subset of data from selected objects (e.g., active risk libraries, open incidents)
Context
Real-time or record-specific information used during the interaction:
- Current records
- Relationships or linked entities
Decision Criteria
Clarify how the agent should reason or evaluate choices.
The agent should make decisions using the following logic: * Use details from the current record (e.g., impact, urgency, affected service) to assess priority * Match the issue type to known solutions or related past incidents * Recommend next steps based on severity and available resolution options
Steps
Map out the interaction flow between the agent and the end user.
Agent: 1. Read the current incident record 2. Check for similar past incidents or known issues 3. Suggest a priority level and recommended next steps End User: 1. Review the agent’s suggestion 2. Confirm or adjust the priority and next action
Guardrails
List constraints the agent must follow.
Do’s and Don’ts:
Do: * Use only verified, up-to-date information * Confirm required fields before creating records Don't * Do not assume values if data is missing * Do not access fields that haven't been explicitly requested
Output Format
Specify how the agent should present information.
Use Markdown to define the response structure:
Copy-Paste Template (for agent instructions)
Below is a ready-to-use format you can paste into the agent’s instruction field.
## Role You are an expert in (subject area) ## Goal I need your help to... ## Input ### Knowledge What domain knowledge or reference material does the agent need? (e.g. documents, policies, wikis, structured objects) Reference the tool names here where needed. ### Context What specific, real-time record data is required? (e.g. current task, user input, related records) Reference the tool names here where needed. ## Decision Criteria Define how the agent should reason and evaluate: - When is something considered relevant? - What logic must be used to assess completeness, accuracy, or compliance? - Should the agent explain or justify its recommendations? ## Steps List the step-by-step flow, specifying both agent and end user actions. Reference the tool names here where needed. **Agent:** 1. Step the agent performs 2. ... **End User:** 1. Step the user performs 2. ... ## Guardrails ### Do’s - What the agent must always do ### Don’ts - Actions the agent must never take ## Output Format Define the format for agent responses using Markdown. Include headers, bold text, bullet points, or code formatting as needed.