Skills are the building blocks that define what a VoiceAgent can do during a call. They control how the agent handles tasks such as collecting information, answering questions, transferring calls, or triggering follow-up actions.
In this article, you will learn what VoiceAgent skills are, how they work, which skills are available, and how to configure them to match your specific use case.
User level:
Admin
Understanding VoiceAgent Skills
VoiceAgent Skills define what a VoiceAgent can do during a call. They control how the agent handles specific tasks, what information it collects, how it responds to callers, and when it should involve a human agent or another system.
CloudTalk offers predefined skills that cover the most common VoiceAgent actions:
Take a Message - allows the VoiceAgent to collect a caller’s message and capture the details needed for follow-up
Extract Information - allows the VoiceAgent to capture structured information from the conversation and store it for reporting, call logs, or external use
Answer Questions – allows the VoiceAgent to respond to caller questions based on the configured knowledge base
Transfer to Human – allows the VoiceAgent to transfer the call to a live agent or group when specific conditions are met
In addition to these predefined skills, you can also create Custom Skills to support more specific or advanced use cases. Custom Skills allow you to define behavior beyond the standard skill set and tailor the VoiceAgent to your business needs.
Take a Message
The Take a Message skill tells the VoiceAgent that it can collect a message from the caller during the conversation.
When this skill is enabled, the VoiceAgent can capture a caller’s message and pass it forward as part of the call data. As a result, the extracted output automatically includes the relevant message fields, such as:
caller_name
message_from_caller
Within the Take a Message skill configuration, you can define what the VoiceAgent should say after the message has been collected.
⚠️ On its own, this skill does not automatically make the VoiceAgent offer message-taking in every conversation. The VoiceAgent will typically use it when it cannot move the conversation forward, for example when a transfer is not possible or the caller’s request cannot be resolved directly.
If you want the VoiceAgent to explicitly tell callers that they can leave a message, add that instruction to the agent’s Scenarios.
Example:
Using this skill together with a clear scenario helps ensure callers know that leaving a message is an available option.
Extract information
The Extract Information skill defines what data should be captured from the call and saved after the conversation ends. This data will be displayed in the call detail, passed to agents after a transfer and sent to external systems.
In this skill, you define a set of properties that the AI should extract from the conversation. For each property, you configure:
Property name - the label used for the extracted value
Type - the format in which the value should be saved (Text/ Number/Yes or No/ List)
Instruction for AI - guidance for the AI describing what value should be captured
Once configured, extracted properties are automatically reflected in the call detail in the CloudTalk Dashboard and are also exported through native integrations.
How does information extraction work?
Once a call with the VoiceAgent ends, the full call transcript is processed by AI using the instructions configured in the Extract Information skill. The AI then returns a JSON object containing the exact fields defined in the skill configuration.
In addition to the properties you configure, CloudTalk automatically includes the Call Result value (Connected, Not Connected or Voicemail).
For template-based VoiceAgents (such as AI Receptionist), additional fields may be included automatically depending on the selected use case.
For outbound calls, the exported object also includes values passed in the output variables of the trigger API request.
Send via webhook
In the Extract Information skill configuration, you can enable Send via webhook setting. When enabled, system sends the extracted data to an external endpoint of your choice in addition to the standard data export available through native integrations. This endpoint acts as the connection point between your VoiceAgent and the workflows, integrations, or internal systems you want to use.
Answer Questions
The Answer Questions skill instructs the VoiceAgent to respond to customer questions based on the information available in VoiceAgent’s knowledge base.
When this skill is enabled, the AI uses the configured knowledge sources to provide relevant answers during the call. This is useful for handling common inquiries without requiring a live agent.
In the skill configuration, you define what the VoiceAgent should do if it does not have enough information to answer the customer’s question. Available options include:
Be honest - the VoiceAgent clearly states that it does not have the information
Guide to other topics - the VoiceAgent redirects the conversation to another relevant topic
Hand over to a person - the VoiceAgent transfers the call to a live agent or group
If you select Hand over to a person, you must choose the specific Agent or Group the call should be transferred to. In most cases, this would be a general support group or another team responsible for handling unanswered questions.
Transfer to Human
The Transfer to Human skill defines the rules under which a VoiceAgent should transfer a call to a live agent or group.
Each transfer rule consists of:
a condition that defines when the call should be transferred, for example based on what the caller says during the conversation
a Transfer to setting that defines who should receive the call
Calls can be transferred to either an Agent or a Group. You can configure multiple transfer rules for different situations, including a combination of agent-based and group-based routing.
⚠️ One of the most common challenges with the Transfer to Human skill is that the VoiceAgent may not route calls as expected. In most cases, this happens when the transfer conditions overlap or are too similar.
Make sure each condition is clearly defined and distinct so the VoiceAgent can reliably decide where the caller should be transferred.
In the Transfer Settings section, you can enable or disable Ask caller for confirmation before transfer. When enabled, the VoiceAgent will ask the caller for explicit confirmation before transferring them to a live agent.
In the Context shared with human agent section, you can define what information the AI should pass to the receiving team before they answer the call. You can enable or disable sharing of Caller name and AI-generated summary. You can also click +Add custom field to define additional properties that should be shared. For each custom field, specify:
Property name – the label of the value to be shared
Instruction – guidance for the AI describing what information should be filled into that field
All shared information is displayed to the receiving agent in Context transfer card within their desktop calling app, helping them understand the context of the call before answering.
When the call is incoming to the agent, a Context Transfer card appears in the agent’s app.
By clicking on it, the agent can view the full set of details passed by the VoiceAgent.
Custom Skill
Custom Skills allow you to define additional behavior for your VoiceAgent beyond the predefined skill set. To add a Custom Skill, select Custom from the list of available skills and click Add. You will then be asked to provide:
Skill Name - the name under which the skill will appear in the list of applied skills
Skill Prompt - the instruction that tells the AI what to do when this skill is used
The Skill Prompt should clearly describe:
when the skill should be applied
what the goal of the skill is
how the VoiceAgent should behave when performing it
You can add as many Custom Skills as needed. These skills will be applied during the call as part of the VoiceAgent’s behavior.
💡 Pro tip: Within the skill prompt, you can also use dynamic variables such as system_prompt variables from outbound call trigger or contact context variables available for inbound calls.
If you need any further assistance please do not hesitate to reach out to our Support team, we are always happy to help!