Both Custom Skills and Instructions shape how your AI agent behaves, but they work in very different ways. Using the right one for the right job keeps your AI agent predictable, your workflows reliable, and your automations running smoothly.
In short:
Custom Skills control what the AI agent does in a specific situation, while instructions control how the AI agent communicates in general.
When to use Custom Skills?
A Custom Skill defines a specific workflow the AI agent should follow when a particular condition is met. Think of it as a trigger + a set of actions.
Custom Skills are the right choice when you want the AI agent to:
Detect something in the conversation and respond in a defined way
Collect information from a customer
Tag a chat or create/update a ticket
Transfer or supervise a conversation
Update customer data
What can trigger a Custom Skill?
A Custom Skill is triggered by the customer's message. The AI agent doesn't just look for specific keywords — it understands the intent behind what the customer writes. This means a skill can be triggered when a customer expresses frustration, shows buying intent, or asks about a specific topic, even if they phrase it differently. On top of that, the AI Agent can also detect details from visitor data — such as the page the customer is currently viewing, their timezone, or their location — and use that as part of the triggering condition.
What can a Custom Skill do?
Once triggered, a skill can:
Call external services via API (webhooks support)
Send a contextual message
Collect details from the customer
Create, find, or update a ticket
Tag a chat or a ticket
Transfer or supervise a chat
Check whether any agents are accepting chats in a specific team
Close a chat
Update customer data
Custom Skill example
When the customer is on the checkout page and asks for a discount, offer the code lastminute5 for 5% off, tag the chat as coupon_issued, and continue the conversation.
When to use Instructions?
Instructions set the AI agent's general tone, style, and identity. They apply across all conversations managed by the specific AI agent, not just in specific situations.
Use Instructions for things like:
Brand personality and how the AI agent introduces itself
Whether the AI agent should avoid or use technical jargon
General safety or communication rules
Good examples of Instructions
Always use a polite and professional tone.
Introduce yourself as a virtual assistant, not a human agent.
Be concise unless the customer asks for more details.
Do not promise refunds, discounts, or delivery dates
What Instructions are NOT for
Instructions should not contain conditional logic or action-based workflows. If a rule depends on what the customer says or does — and requires the AI agent to take a specific action in response — it belongs in a Custom Skill, not in Instructions.
For example, telling the AI agent to offer a coupon when a customer asks for a discount or to create a ticket when a customer complains are both Custom Skills. Instructions can't execute those steps — they can only shape how the AI agent communicates.
Quick-reference guide
Scenario | Custom Skill | Instructions |
The customer asks for a discount and should receive a coupon | Yes | No |
The customer is from Spain and should be transferred to the Spanish team | Yes | No |
Customer complains, and a ticket should be created | Yes | No |
The customer uses abusive language, and the chat should be closed | Yes | No |
AI agent should sound friendly and professional | No | Yes |
AI agent should never pretend to be human | No | Yes |
AI agent should avoid overpromising | Sometimes | Yes |
AI agent should collect order details before creating a ticket | Yes | No |
A simple rule of thumb:
"When X happens, do Y" → Custom Skill
"Always communicate in this way" → Instructions
Best practices for Custom Skills
1. Start with a precise trigger
Every Custom Skill needs a clear answer to: When exactly should this skill run?
The trigger should be based on specific words, phrases, or intent in the customer's message, not a general topic or vague situation.
Good examples of precise triggers:
When the customer asks for a discount…
When the customer asks about shipping costs…
When the customer reports a problem with their order…
Triggers to avoid:
When the customer needs help…
When the customer seems interested…
When the conversation seems sales-related…
These are too broad and may cause the skill to run at the wrong time or too often.
2. Use customer context, not just message content
The most effective Custom Skills combine what the customer says with what you already know about them.
Weak:
When the customer asks for help, transfer them to Sales.
Better:
When the customer is on the checkout page and asks for help choosing products, ask about their budget, room size, and preferred setup, create a short summary, tag the chat as high_intent, and offer a transfer to the Sales team.
The second version is more precise, easier to test, and less likely to be triggered by accident.
3. Keep actions in the right order
A Custom Skill is a sequence, so the order matters. A common mistake is transferring the chat too early, before the AI agent has collected what it needs.
Bad order:
Transfer the customer to Sales, then ask for their budget and room size.
After a transfer, the AI agent may not continue collecting information.
Recommended order:
Detect the trigger
Check relevant context
Collect missing details
Create a summary
Create or update a ticket, customer record, or event (if needed)
Apply tags
Transfer, supervise, or close the chat
Terminal actions such as Transfer chat, Close chat, and Supervise chat should almost always come last.
4. Make the skill specific and complete
A strong Custom Skill defines:
What triggers it
What context to check
What questions the AI agent should ask
What data to collect
What action to take
What tag to apply
Which team or agent receives the transfer
What happens if the required data is missing
Useful format:
When [trigger/context], check [relevant variables], ask for [missing details], create [summary/ticket/update], tag the chat as [tag], and then [send message / transfer / supervise / close].
5. One skill, one goal
Don't pack unrelated workflows into a single skill. Keep each Custom Skill focused on one purpose.
Instead of one skill that does everything, create separate skills:
Product Recommendation Skill
Checkout Discount Skill
Open Ticket Status Skill
Support Escalation Skill
Profile Enrichment Skill
Smaller, focused skills are easier to test, debug, and improve.
6. Always define a fallback
What should the AI agent do when something is missing or unavailable? Every Custom Skill should handle edge cases like:
The customer doesn't provide an order number
The customer profile is missing
The ticket cannot be found
The target team is unavailable
Customer doesn't meet discount eligibility
Example:
If the customer cannot provide an order number, ask for it before creating the ticket. If they still can't provide it, create the ticket with available details and tag it as missing_order_id.
Fallbacks make your skill more predictable and prevent dead ends.
7. Verify groups, tags, and permissions before launching
If a Custom Skill references a group name, tag, or permission, it must already exist in your Text workspace, and the AI agent must have access to it.
Before using Transfer chat or Tag chat, confirm that:
The target group exists
The tag exists
The AI agent belongs to the right group or has access to it
The AI agent can transfer chats to the target group
The AI agent can apply the selected tag
The names used in the skill match your Text workspace configuration exactly
This applies to any skill that involves transfers, tagging, supervision, team routing, or escalation.
8. Run a pre-launch check
Before enabling a new Custom Skill, validate each of the following:
Check | Why it matters |
Trigger is specific | Prevents accidental execution |
Context variables are available | Ensures conditions can be evaluated |
Required questions are defined | AI agent collects the right data |
Action order is logical | Prevents partial execution |
Target group exists | Transfer won't fail |
AI agent has access to the target group | AI agent is allowed to route the chat |
Required tag exists | Tagging won't fail silently |
AI agent can apply the tag | Tags may be restricted by group |
Tag spelling matches exactly | Prevents reporting fragmentation |
Fallback path is defined | No dead ends |
Skill is tested on a real scenario | Confirms end-to-end workflow |
Common mistakes and how to fix them
Mistake | Problem | Fix |
"When the customer needs help, transfer to Sales." | The trigger is too vague | "When the customer is on the checkout page and asks for help choosing products, collect budget and room size, then offer a transfer to Sales." |
"Transfer to Sales, then ask for details." | After the transfer, the AI agent may not continue | "Collect details first, create a summary, then transfer to Sales." |
"Create a ticket if needed." | "If needed" is undefined | "Create a ticket when the customer reports a product issue and provides an order number or product model." |
"Give a discount to interested customers." | Eligibility is unclear | "When the customer is on checkout and asks for a discount, offer code lastminute5 and tag the chat as coupon_issued." |
"Transfer to Premium Sales." | The group may not exist, or the AI agent may lack access | Create the Premium Sales group first, verify the AI agent's access, and use the exact group name. |
"Tag as VIP." | Tag may not exist or be inconsistently named | Create the tag vip_customer, verify the AI agent can apply it, and use it consistently. |
"Handle complaints." | Too vague to be actionable | "When the customer complains about service quality, collect what happened, when, and the desired resolution, then create a ticket and tag it as unhappy_customer." |
Summary
Custom Skills are for workflows, actions, and decisions – anything where the AI agent needs to detect what the customer said and respond with a specific set of steps. Use them for routing, tagging, ticketing, escalation, collecting information, and calling external services.
Instructions are for tone, identity, and style. They define how the AI agent communicates across every conversation, regardless of what the customer says.
A strong Custom Skill isn't just well-written; it references real groups, tags, and permissions that actually exist in your Text workspace and that your AI agent can access.
Frequently asked questions
Yes — they're designed to work together. Instructions set the AI agent's general tone and identity across all conversations, while Custom Skills handle specific workflows and actions. Both are active simultaneously, so the AI agent can follow your brand voice while still executing the right steps when a trigger is met.
A vague trigger can cause the skill to fire in the wrong situations, leading to unexpected AI agent behavior or broken workflows. For example, "when the customer needs help" could match almost any conversation. Always define triggers with specific conditions — such as page URL, customer message content, or customer status — to ensure the skill runs only when intended.
No. Instructions are not designed to handle conditional logic or branching behavior. If you need the AI agent to do different things depending on the situation — such as routing customers based on region or language — create a Custom Skill with the appropriate trigger and actions. Instructions should only describe how the AI agent communicates.
Yes. If a Custom Skill references a group name or tag that doesn't exist in your Text workspace, the action will fail silently or not execute at all. Always create the required tags and groups first, confirm the AI agent has access to them, and make sure the names match exactly — including capitalization and spacing.
An AI agent can have up to 10 Custom Skills. To make the most of them, keep each skill focused on a single goal. If your AI agent is handling many different scenarios, organize skills by use case — such as sales, support, and retention — rather than building one large skill that tries to cover everything.
What's next
Now that you understand how Custom Skills and Instructions work, you're ready to put them into practice.
To build your first Custom Skill, go to your AI agent settings and open the Skills tab. Define a clear trigger, add your actions in the right order, and run a test before enabling it. See how to create and manage Custom Skills.
To set up your AI agent's tone and communication style, go to the Instructions tab in your AI agent settings and add your general guidelines. Learn more about configuring your AI agent.
Before launching, make sure any groups and tags referenced in your Custom Skills already exist in your Text workspace and that the AI agent has the correct permissions. See how to manage team permissions.
To give your AI agent access to the right information when responding to customers, configure your Knowledge sources with relevant content such as websites, files, and help articles.