Let AI Run Code & Reusable Scripts for You
You can have an AI assistant write and execute real code — Python, JavaScript, or Bash — in an isolated cloud sandbox, without installing anything. Once a piece of logic works, it's saved as a reusable workspace script with a defined input schema, so you (or your agents) can re-run it later with new inputs instead of re-explaining the task. Encrypted workspace secrets can be attached to sandbox runs so scripts can call external APIs without ever exposing credentials in chat.
Code & Scripts
Run code in a sandbox and manage reusable workspace scripts.
Execute Python, JavaScript, or Bash in an isolated sandbox, and save, run, and maintain reusable workspace scripts with structured inputs and secrets.
What you can do
Run code in an isolated sandbox
Save and re-run reusable workspace scripts
Attach encrypted secrets to sandbox runs
Try saying
“Run this Python script”
“Execute some code to transform this data”
“Save this as a reusable script”
When the assistant uses this
Use when the user wants to execute code, run a sandbox computation, or create, run, update, or delete reusable workspace scripts.
How does Code & Scripts work?
Loads guidance for running Python, JavaScript, or Bash in an isolated sandbox, validating logic before persisting it, saving reusable scripts with structured input schemas, re-running saved scripts, and attaching encrypted workspace secrets to sandbox runs. The code-execution and script tools are unlocked only when this Skill is loaded.
What phrases trigger this Skill?
“run a script”
“run code”
“execute code”
“run python”
“save a script”
“sandbox”
Frequently asked questions
Where does the code actually run?
In an isolated cloud sandbox (an E2B virtual machine), not on your computer and not inside the AI model. Each run is contained: you can use a fresh ephemeral VM per run, or a persistent sandbox when state needs to survive between runs.
What's the difference between running code and a saved script?
Running code is a one-off: the assistant validates logic in the sandbox and nothing is persisted. A saved script is durable: working code is stored in your workspace with a name, description, and a structured input schema, so it can be re-run on demand — by you, by an assistant, or by an automated agent — with different inputs each time.
How do scripts use API keys without leaking them?
Credentials live as encrypted workspace secrets. The assistant can list which secrets exist and attach one to a sandbox by reference, but it can never read the plaintext value — the secret is injected into the sandbox environment at run time, not into the conversation.
Can scripts work with my workspace files and data?
Yes. Files and tables from your workspace are passed into runs as references, and the script reads them inside the sandbox. Large files are never pasted into the code itself, which keeps scripts clean and re-runnable on new data.
What kinds of tasks is this good for?
Anything deterministic that chat alone can't guarantee: data transformation and cleanup, file conversion, calculations, calling external APIs, and recurring jobs you want to standardize. The pattern is validate once in the sandbox, save it, then re-run it forever.