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

# Running a skill's scripts

> The sandbox a skill's own code runs in, what's preinstalled, and the approvals needed before Abbie can execute it.

Many published skills — including the Excel, Word, PowerPoint and PDF ones — ship Python scripts alongside their instructions, rather than relying on Abbie to write the equivalent code from scratch each time. Abbie can run these scripts, inside a sandbox built to limit what they can touch.

## The sandbox

A skill's scripts run with:

* **No network access.** A script cannot reach the internet, your systems, or anything else outside the sandbox.
* **No access to your uploaded documents.** A skill's scripts only ever see the skill's own files, plus whatever Abbie explicitly passes in.

The only thing a script can write is its own output, which is what comes back to Abbie.

### Preinstalled libraries

Only these libraries are available to a script: `openpyxl`, `pandas`, `numpy`, `python-docx`, `python-pptx`, `pypdf`, `pdfplumber`, `reportlab`, `Pillow`, `lxml`, `beautifulsoup4`, `markitdown`.

<Warning>
  A skill's own dependency list can't be installed. A skill that ships a `requirements.txt` still installs fine, but nothing in that file is ever fetched — only the libraries above are available, whatever the skill asks for.

  Where the file sits changes what you see. A `requirements.txt` at the top of a skill is taken as its dependency list, so its scripts are refused up front rather than run and failed halfway. One tucked away beside a single script, or shipped as documentation, doesn't block anything — those scripts run, and only fail if they actually reach for something missing. Either way the message names the `requirements.txt` file, so you can tell this apart from a broken skill.
</Warning>

## Two approvals before anything runs

Running a skill's scripts at all takes two separate approvals:

1. An account admin turns script execution on for the account.
2. Each individual skill is approved on its own, on top of that.

Approving a skill shows you the exact script files you're approving, not just the skill's name. Revoking approval is one click, and a skill that ships no scripts of its own offers no approval control at all, because there's nothing to approve.

## LibreOffice isn't available

Some document skills use LibreOffice for a final validation pass after producing a file. That step isn't available here, so it's skipped — the document itself is still produced, just without that last check.