Execute the Code Extension Workflow by Using Agentic Tools and Salesforce Skills

Author, debug, and deploy Data 360 code extensions by using natural-language prompts in agentic AI tools such as Claude Code and Gemini CLI.

Prompts work best when you follow the natural workflow sequence — each step builds on context from the previous one. The exact wording of your prompt is flexible. The agentic tools use the installed Salesforce Skills to interpret your intent and carry out the action.

• Make sure that you have access to a Data 360 sandbox and the Data Cloud Architect permission set. If you don’t have a sandbox, create one. See Create a Data 360 Sandbox.
• Enable code extension in Data 360 Setup by using Feature Manager. See Enable Data 360 Features in Feature Manager.
• Set up Salesforce CLI for code extension. See Set Up Salesforce CLI for Code Extension.
• Install and set up your agentic AI tool, such as, Claude Code or Gemini CLI. Then, open your Salesforce DX project.
• Authenticate to your target Salesforce org by using Salesforce CLI. See Authorize an Org Using the Web Server Flow.

1. Install the Salesforce skills for code extension.

   When you run the command, it automatically detects the agentic tools that you have installed and prompts you to select which tools to install the skills to. For example, if you have Claude Code, Codex, and Gemini CLI installed, select the tools that you want to use and continue. To skip the prompt and install to a specific tool directly, add the --agent option, for example, --agent gemini-cli or --agent claude-code.

   This command installs the following skills, along with their helper scripts, into your selected agentic tools.

   Skill	Description
   developing-datacloud-code-extension	Guides end-to-end code extension development and deployment.
   getting-datacloud-schema	Retrieves DLO and DMO schema metadata so that you can validate object and field names before testing and deploying code extension.

   The getting-datacloud-schema skill includes helper scripts that fetch live DLO and DMO schemas before writing or validating entrypoint.py. This reduces hallucinated object names, prevents field mismatch errors during run and deploy, and keeps your prompts intent-focused instead of manually listing schema details.

   Each agentic tool stores and loads skills from its own location, so the install path can differ by tool. If the skills don’t install or your tool doesn’t detect them, see your tool’s official documentation for how it sets up and loads skills.

2. Before you start prompting, note the following expected behaviors:

   • The tool can pause and ask you to review and approve certain commands before running them — for example, terminal commands or file operations that aren’t pre-approved. When prompted, review the proposed action and approve or reject it.
   • The tool can ask for your org alias during the workflow. For example, running local tests or deploying. Sometimes, it determines the org automatically by running sf org list. If prompted, provide your org alias to continue.

3. In chat, describe each workflow action. For example:

   1. Initialize the package: create new code extension script package named code_extension_pkg_1

   2. Author the script: Refer @/code_extension_pkg_1/payload/entrypoint.py I want you to modify that file to read Employee DLO and write to Employee_Hierarchy DLO. Make sure to compute values for all the fields of Employee_Hierarchy DLO.

      The tool uses the getting-datacloud-schema skill to retrieve schema context and the developing-datacloud-code-extension skill to update payload/entrypoint.py based on your prompt.

   3. Test locally: test it locally

   4. Deploy: deploy it

• Invoke Code Extension by Creating a Batch Data Transform

• Write and Validate Custom Scripts
• Deploy a Custom Script to Data 360 Sandbox by Using CLI
• Quick Start