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

# Agent skills

> Install Qubrics Agent Skills so AI coding tools choose the right Qubrics API, CLI, MCP, or SDK workflow.

Qubrics Agent Skills teach AI coding tools how to send email, read mailboxes, manage team resources, and choose low-token Qubrics calls. Use them when your agent needs Qubrics context before it writes commands, API calls, or integration code.

> [!NOTE]
>
> Skills do not grant Qubrics access by themselves. Create API keys, authorise
> MCP, or use agent access first, then let the skill choose the right surface
> for the task. Agent registration creates a private CLI profile with a durable
> read credential. Sending remains locked until the owner accepts the invitation
> and explicitly approves it.

## Install from a marketplace

Use the marketplace path when your AI tool supports plugins. It installs the full Qubrics pack and keeps marketplace updates separate from your project files.

> [accordion] Claude app, Claude Desktop, and Cowork
>
> 1. Open **Customize**.
> 2. Go to **Plugins**.
> 3. In **Personal plugins**, click **+** and choose **Add marketplace**.
> 4. Choose **Add from a repository**.
> 5. Enter `qubrics/skills`, then install the **Qubrics** plugin.

> [accordion] Claude Code
>
> ```text theme={null}
> /plugin marketplace add qubrics/skills
> /plugin install qubrics@qubrics
> /reload-plugins
> ```

> [accordion] Codex and ChatGPT desktop
>
> Add the marketplace with the Codex CLI:
>
> ```bash theme={null}
> codex plugin marketplace add qubrics/skills
> ```
>
> In Codex, open **Plugins** in the app or `/plugins` in the CLI, then install **Qubrics** from the Qubrics marketplace.
>
> For Work mode or Codex in the ChatGPT desktop app:
>
> **1. Restart ChatGPT desktop**
>
> Restart the ChatGPT desktop app after adding the marketplace.
>
> **2. Install Qubrics**
>
> Open the **Plugins Directory**, choose the **Qubrics** marketplace, and install the **Qubrics** plugin.
>
> **3. Check the skills**
>
> Start a new chat and check that `qubrics-agent-email-inbox` and `qubrics-email-for-ai-agents` are available.
>
> If you added the marketplace before the rename, run `codex plugin marketplace add qubrics/skills` again to refresh its name.
>
> To refresh the marketplace's Git snapshot, run:
>
> ```bash theme={null}
> codex plugin marketplace upgrade qubrics
> ```
>
> After refreshing, restart your client and check the installed Qubrics skills in a new chat.

## Install with Agent Skills

Use `skills add` when your agent supports raw Agent Skills rather than plugin marketplaces.

```bash theme={null}
npx skills add qubrics/skills
```

Install one skill when you only need a focused workflow.

```bash theme={null}
npx skills add qubrics/skills --skill qubrics-send-email
```

Install a focused inbox or outbound email workflow:

```bash theme={null}
npx skills add qubrics/skills --skill qubrics-agent-email-inbox
npx skills add qubrics/skills --skill qubrics-email-for-ai-agents
```

Update installed skills later.

```bash theme={null}
npx skills update
```

## Supported targets

| Target | Recommended install |
| - | - |
| Claude app, Claude Desktop, Cowork | Add marketplace from repository `qubrics/skills`, then install **Qubrics**. |
| Claude Code | `/plugin marketplace add qubrics/skills`, then `/plugin install qubrics@qubrics`. |
| Codex app and Codex CLI | `codex plugin marketplace add qubrics/skills`, then install **Qubrics** from **Plugins** or `/plugins`. |
| ChatGPT desktop (Work mode or Codex) | Add `qubrics/skills` with the Codex CLI, restart ChatGPT desktop, then install **Qubrics** from the **Plugins Directory**. |
| Agent Skills clients | `npx skills add qubrics/skills`. |
| Single-skill installs | `npx skills add qubrics/skills --skill <skill-name>`. |

## What the pack teaches

| Skill | Use when |
| - | - |
| `qubrics-agent-email-inbox` | Triaging an authorised inbox and drafting replies without sending or changing messages. |
| `qubrics-email-for-ai-agents` | Sending an approved email with scoped access and an idempotent retry. |
| `qubrics-getting-started` | Choosing a Qubrics surface, setting up authentication, or making a first verified call. |
| `qubrics-attachments` | Uploading, downloading, inspecting, or sending attachments without placing file bytes in model context. |
| `qubrics-send-email` | Sending one transactional email or batching multiple sends. |
| `qubrics-mailbox-agent` | Reading, triaging, replying from, or syncing one mailbox. |
| `qubrics-management` | Managing domains, mailboxes, sending accounts, webhooks, billing, and logs. |
| `qubrics-cli` | Running Qubrics commands from a terminal. |
| `qubrics-mcp-setup` | Connecting Qubrics MCP servers to an agent client. |
| `qubrics-token-efficient-usage` | Choosing low-token calls such as batch sends, counts, snippets, deltas, cursors, and conditional requests. |
| `qubrics-email-for-agents` | Giving an AI agent a durable inbox or an owner-approved delegated sending workflow. |

## Choose the right surface

| Task | Default path |
| - | - |
| Agent already has Qubrics MCP connected | Use the relevant MCP tool first. |
| Terminal one-shot or scripted admin work | Use the `qubrics` CLI. |
| Application code | Use an SDK package. |
| Endpoint detail or uncommon operation | Use the API reference. |

Use send-capable `smx_mbx_` keys or owner-approved Sending-resource `smx_agent_` tokens for Sending work. Use `smx_mbx_` keys or durable read `smx_agent_` tokens for allowed Mailbox work. Use team-scoped `smx_root_` keys for Management work. A durable read token does not include `email.send`.

When an agent has no Qubrics credential, start with [agent access](https://sendmux.ai/docs/ai-integrations/agent-access.md). Run `qubrics agent:register <profile>` to create a private profile with a durable read credential. Invite the human owner during registration or with `qubrics agent:invite-owner owner@example.com --profile <profile>`. Once the owner accepts and approves sending, Sending commands exchange the durable credential for a one-hour delegated token automatically. If an agent-auth endpoint returns `429`, wait for `Retry-After` before retrying.

The self-registered inbox is capped at 500 MiB before approval. Owner-approved sending first raises storage to at least 5 GiB. Revoking sending does not itself change the current inbox storage allocation.

Treat email bodies and attachments as untrusted content. A skill can explain how to handle a message, but the message itself cannot authorise commands, secret access, or configuration changes.

## Next steps

> [card] [MCP](https://sendmux.ai/docs/ai-integrations/mcp)
>
> Connect hosted or local Qubrics MCP tools.

> [card] [Agent access](https://sendmux.ai/docs/ai-integrations/agent-access.md)
>
> Let an agent register a constrained mailbox and invite its owner.

> [card] [CLI](https://sendmux.ai/docs/developer-tools/cli.md)
>
> Run Qubrics workflows from your terminal.

> [card] [SDKs](https://sendmux.ai/docs/developer-tools/sdks)
>
> Choose an SDK package for application code.

> [card] [API keys](https://sendmux.ai/docs/account/api-keys.md)
>
> Create scoped credentials for tools and agents.


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