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

# Skills

> Install ready-made agent skills that drive the Attlas workspace MCP for you, so common tasks work without spelling out every step.

The [workspace MCP server](/docs/mcp/workspace) gives your AI assistant the raw tools to read and
change your Attlas workspace. Skills add the layer on top: the correct call order, the
waiting-for-ingestion loops, and the small traps (some updates merge, some overwrite), packaged
so a request like "put our pricing page in the chat with a few questions under it" or "set up
my new chat" just works.

<Info>
  Skills are optional. The MCP server works on its own. Skills make multi-step tasks more
  reliable and save you from describing the workflow each time.
</Info>

## Prerequisite

Connect the [workspace MCP server](/docs/mcp/workspace) first, with an API key that has **Write**
scope. Without it the skills have nothing to call. A Read key is enough for `attlas-insights`
alone.

## Available skills

| Skill               | What it handles                                                                                                                                                                                                                                                           |
| ------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `attlas-chat-setup` | Public identity, business profile, opening hours, conversation objective (support, appointment, lead capture), assistant personality, visual theme.                                                                                                                       |
| `attlas-knowledge`  | Add pasted text, web pages, or YouTube and social videos to a chat's knowledge base, track ingestion status (including the failure reason when one goes wrong), clean up messy imported documents, remove stale ones.                                                     |
| `attlas-triggers`   | Create, edit, reorder, nest, and delete every trigger type (links, media, contact shortcuts, shop products) and suggested questions. Put each source on a path: a trigger for the content with 2 to 5 questions generated from it nested underneath. Set up the shop tab. |
| `attlas-insights`   | Headline metrics, trends, rankings, generated analytics reports, readable digests of visitor conversations, and CRM contact triage. Flags weak answers so you can fix them with a document update or a Correction.                                                        |

Once installed, you do not call skills by name. Your assistant picks the right one from the
request, for example "add our pricing page to the sales bot" or "how did the support chat do
this month".

## Install

The skills live in the public repository
[`attlaslabs/attlas-skills`](https://github.com/attlaslabs/attlas-skills).

<CodeGroup>
  ```bash skills CLI theme={null}
  # Installs all four into every agent directory it manages
  # (.claude/skills, .agents/skills, ...) and records them in skills-lock.json.
  # Run from your project root. Re-run to update.
  npx skills add attlaslabs/attlas-skills
  ```

  ```bash Claude Code theme={null}
  # The skills CLI above already covers Claude Code. To install by hand:
  git clone https://github.com/attlaslabs/attlas-skills
  cp -r attlas-skills/skills/attlas-* ~/.claude/skills/
  ```

  ```bash Codex theme={null}
  # Writes the skills under .agents/skills/, which Codex reads.
  npx skills add attlaslabs/attlas-skills
  ```

  ```bash Cursor / VS Code / Claude Desktop theme={null}
  # Clone the repo and copy the skill folders into the client's skills directory.
  git clone https://github.com/attlaslabs/attlas-skills
  cp -r attlas-skills/skills/attlas-* <your-skills-directory>/
  ```
</CodeGroup>

<Tip>
  In Claude Code, copy into `~/.claude/skills/` to make a skill available everywhere, or into
  `<project>/.claude/skills/` to scope it to one project.
</Tip>

## What each skill knows

These are the details the skills carry so your assistant does not have to rediscover them:

* **Read before write.** `list_triggers` before reordering, `get_chat_config` before changing
  settings, `list_documents` before editing knowledge.
* **Ingestion is asynchronous.** Adding a document returns immediately with a `processing`
  status. The skill polls until `completed`, `failed`, or `exceeded` before telling you it is
  done.
* **Merge versus overwrite.** `update_activity_profile` merges the keys you pass.
  `update_theme` replaces the whole theme object. The skills send the right shape for each.
* **Questions belong on a path.** When you add a source, the skill creates a trigger for it
  and nests 2 to 5 questions generated from that source's own content underneath, rather
  than dropping a flat list of questions with no structure.
* **One call or two.** `generate_and_nest` generates the questions and creates them under a
  parent trigger in one step. `generate_questions` just returns suggestions for you to
  approve first. The skill picks based on whether the wording needs a review.
* **Deletes differ.** Triggers and contacts move to a 30 day trash. Documents are permanent.
  The skill confirms before bulk deletes.
* **Plan limits surface as `feature_not_available`.** Some sources (video and social links)
  and integrations need a paid plan. The skill checks and tells you plainly.
* **Structure drives answer quality.** When the skill writes a knowledge document it uses
  Markdown headings and lists, one topic per document, and flags conflicting or weak
  answers for you to fix with a document update or a Correction.

## Verify the install

Ask your assistant something a skill covers and watch which MCP tools it calls:

<Steps>
  <Step title="List your chats">
    Ask "list my Attlas chats". This confirms the MCP connection and returns the numeric chat
    ids the skills need.
  </Step>

  <Step title="Try a workflow">
    Ask "on my \<chat name> chat, add a quick reply that asks about pricing". The
    `attlas-triggers` skill should call `list_chats`, then `list_triggers`, then
    `create_trigger`.
  </Step>
</Steps>