> ## Documentation Index
> Fetch the complete documentation index at: https://docs.conduit.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Teach

> Teach V3 agents with Knowledge, Skills, Persona, Guardrails, Tools, and Behavior settings.

Teaching an agent means putting each instruction in the right place. V3 separates facts, reusable process guidance, tone, restrictions, actions, and deployment behavior so changes are easier to reason about.

## Teaching Surfaces

| Surface | Put this there |
| - | - |
| **Knowledge** | Facts the agent may need to retrieve. |
| **Skills** | Reusable process instructions, tool-use instructions, and conditional playbooks. |
| **Persona** | The agent's role, tone, and default communication style. |
| **Guardrails** | Hard boundaries the agent should not cross. |
| **Tools** | Executable capabilities and model-facing descriptions of when to use them. |
| **Behavior** | Escalation rules, working hours, delay, auto mark-as-done, structured output, and hooks. |

## How to Teach

Use this flow when adding or changing agent behavior:

1. Add or update factual material in **Agents > Knowledge**.
2. Add repeatable process guidance in **Agents > Skills**.
3. Assign the Skill to the right agent from that agent's **Skills** tab.
4. Update the agent's **Persona** if the overall role or tone changed.
5. Update **Tools** if the agent needs an action or live lookup.
6. Update **Behavior** if the change affects escalation, timing, language, or automation.
7. Test before publishing.

Keep Knowledge and Skills composable. Knowledge pages can reference related Knowledge pages so facts stay focused and reusable. Skills can reference Knowledge pages for facts and other Skills for reusable sub-processes, so one playbook can build on another without copying the same instructions into multiple places.

## Teaching From Real Conversations

Repeated escalations are the best signal that something is missing. When you inspect a failed or escalated case, decide what kind of missing input caused it:

* Missing fact: add Knowledge.
* Missing process: add or edit a Skill.
* Wrong tool call: update tool access or the tool description.
* Wrong tone: update Persona or a tone Skill.
* Unsafe behavior: add a Guardrail or Behavior rule.
* Bad deployment timing: update working hours, delay, or triggers.

## Maintenance

Keep Skills and Knowledge narrow. One page or Skill should cover one clear topic. If a policy, tool, or workflow changes, update the related Skill and then run Sandbox tests for common and edge-case messages.

<CardGroup cols={2}>
  <Card title="Knowledge Base" icon="book-open" href="/build-your-agent/knowledge-base/overview">
    Add facts, folders, documents, links, and synced sources.
  </Card>

  <Card title="Skills" icon="cube" href="/build-your-agent/skills">
    Write reusable process and tool-use instructions.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.