diff --git a/docs.json b/docs.json
index 1ea5fb17..9e34baa0 100644
--- a/docs.json
+++ b/docs.json
@@ -154,7 +154,8 @@
                   "guides/ai-agents/best-practices",
                   "guides/ai-agents/self-improvement",
                   "guides/ai-agents/mcp-servers",
-                  "guides/ai-agents/autopilot"
+                  "guides/ai-agents/autopilot",
+                  "guides/ai-agents/built-in-skills"
                 ]
               },
               "references/integrations/lightdash-mcp",
diff --git a/guides/ai-agents/built-in-skills.mdx b/guides/ai-agents/built-in-skills.mdx
new file mode 100644
index 00000000..ef39e141
--- /dev/null
+++ b/guides/ai-agents/built-in-skills.mdx
@@ -0,0 +1,59 @@
+---
+title: "Built-in skills"
+sidebarTitle: "Built-in skills"
+description: "Specialized instruction sets that Lightdash AI agents load on demand to handle focused tasks like editing dashboards in the UI."
+---
+
+Lightdash AI agents ship with **built-in skills** — specialized instruction sets the agent loads on demand when a request matches a skill's purpose. Skills keep the base system prompt small and let the agent pull in deep, task-specific guidance only when it needs it.
+
+You don't install or configure built-in skills. They're packaged with Lightdash and available to every AI agent automatically.
+
+## How skills work
+
+When you message an AI agent, it sees a short list of available skills (name and one-line description). If your request matches a skill, the agent:
+
+1. Calls `loadSkill` to fetch the full instructions for that skill.
+2. Optionally calls `loadSkillResource` to pull a specific reference document the skill points to.
+3. Follows the skill's instructions to complete your task.
+
+You'll see both calls appear in the chat as tool-call chips, so you can tell which skill the agent picked up and which reference it consulted.
+
+## Available skills
+
+### `developing-in-lightdash`
+
+Use this skill to read or edit the dashboard you currently have open in the UI. The agent inspects the live dashboard editor state, builds the smallest possible change, and applies it in place — no download, edit, upload loop required.
+
+Trigger it with requests like:
+
+- *"Add a filter for the last 30 days to this dashboard."*
+- *"Move the revenue tile to the top row."*
+- *"What tiles are on this dashboard right now?"*
+
+When the skill is loaded, the agent uses the `frontendAction` tool with `dashboard.read` to inspect state and `dashboard.edit` with an [RFC 6902 JSON Patch](https://datatracker.ietf.org/doc/html/rfc6902) to apply changes. The skill also exposes a `dashboard-reference` resource that documents tile types, layout rules, and filter configuration — the agent loads it on demand when it needs the full schema.
+
+<Tip>
+The `developing-in-lightdash` skill operates on the **in-memory dashboard editor state**. Your changes aren't saved until you click **Save** in the dashboard editor, so you can review the agent's edits before committing them.
+</Tip>
+
+## When to use built-in skills vs. other workflows
+
+| Task | Use |
+|------|-----|
+| Edit the dashboard you have open right now | Built-in `developing-in-lightdash` skill (just ask the agent) |
+| Bulk edit many dashboards or charts at once | [Editing dashboards with agents](/guides/developer/editing-dashboards-with-agents) (download / edit / upload) |
+| Build or maintain your dbt semantic layer with an external coding agent | [Agent skills for coding assistants](/guides/developer/agent-skills) |
+
+## FAQs
+
+**Do I need to enable built-in skills?**
+
+No. They're available to every Lightdash AI agent out of the box.
+
+**Can I add my own built-in skills?**
+
+Not today. Built-in skills are packaged with Lightdash. If you need agent-specific guidance, use [agent instructions](/guides/ai-agents/getting-started) or [agent memory](/guides/ai-agents/agent-memory).
+
+**Why do I see `loadSkill` and `loadSkillResource` in the chat?**
+
+Those are the tool calls the agent makes to fetch a skill's instructions and reference material. Seeing them means the agent recognized your request as a good fit for a specialized workflow.