Skills and tools
Tools
Actions your agents can take, from sending an email to running your own script against your own data.
A tool is something the AI can do. The Tools page lists every tool in the workspace, grouped by the app or bundle it belongs to, with its approval mode.
Three kinds of tool
| Kind | What it is | Where it comes from |
|---|---|---|
| Connected app action | An action in an app you authorised: send an email, create a deal, post a message. | Add connection on the Tools page. See Connected apps. |
| Custom tool | An action you configure yourself: a database query, an HTTP request, an FTP download, list or upload, OCR on a PDF or scan, or a JavaScript snippet. | Custom tool on the Tools page. |
| Script tool | A bundle of code and data with a description: a lookup script plus the spreadsheets it reads, for example. | Built by the automation builder in chat, or imported with an automation. |
Custom and script tools are listed together under Native on the Tools page. Built-in abilities such as searching knowledge or asking a person are always present, need no setup, and are not listed there.
The description matters
Each tool has a name and a description. The agent reads the description to decide whether to call the tool, so be specific: what it takes, what it returns, what side effects it has. "Look up purchase and sale prices for a product at a port; returns per-packaging prices and a ready-to-paste block" is a good description; "price tool" is not.
Approval modes
Every connected app has an approval mode, set in the dropdown on its group header on the Tools page; each custom or script tool has its own toggle.
| Mode | Behaviour |
|---|---|
| Writes freely | The agent uses the connection without asking. The default. |
| Asks before writing | Reads run freely; sends, posts and edits wait for approval. |
| Asks first | Every call waits for your approval. |
| Off | The agent cannot use the connection. |
Approvals show as a card in the chat or channel, and on Slack or WhatsApp when the automation has an escalation door there.
Scoping per agent and per automation
The workspace decides what exists. Each agent chooses all tools or a set, and each automation chooses, per connection, which actions it may use. An action allowed nowhere is never called.
Script tools in detail
A script tool's own page holds its basics, inputs, parameters and expected output. Its files (the entry script, reference documents and data files) are listed on the automation that uses it, in the Files and scripts card on the Steps tab. Text files open in a viewer; binary assets are listed only. The automation builder can update these files when you ask it to, and an MCP client with a workspace key can read and write them too. See MCP server and API.
Disabling a tool
A custom or script tool can be switched off on its page with the Active / Disabled toggle. A disabled tool shows an Inactive pill in the list and is not offered to agents.