Script actions

Autobot can act on your scripts directly from the chat — reading what you already have, writing new ones, and (in execution mode) running them. It can list your scripts, read an existing script to explain or build on it, create or update scripts in Python, PowerShell, or shell — saving them straight into your Script library, owned by you — and run a script against a target server.

Action	What it does	Mode
List scripts	Surfaces the scripts in your library so suggestions match what you already have.	Research
Read script	Pulls a script's contents to explain it, debug it, or use it as a starting point.	Research
Create script	Writes a new script and saves it into your library.	Generation
Update script	Edits an existing script in place when you ask for changes.	Generation
Run script	Executes a script on a target server and reports the result.	Execution

Reading and writing are available to everyone. Running requires execution mode and your own LLM key — see Running a script from chat below.

Two ways in

You can reach script generation from two places:

1. From the chat — just ask. "Write me a Python script that pings a list of servers and prints any that are unreachable."
2. From the Script Editor — the Autobot button in the bottom-right corner of the editor opens a pre-filled Autobot thread focused on script generation. This is the natural entry point when you're already writing scripts and want a hand.

Both routes land you in the same chat surface; the editor button simply scopes the conversation toward scripts.

What a good prompt looks like

Autobot writes better scripts when you tell it about the environment and the intent, not just the mechanics:

_"A shell script for Ubuntu that rotates /var/log/myapp/_.log files older than 7 days, gzips them, and writes a one-line summary to stdout."*

"A PowerShell script that checks the status of the Spooler service and restarts it if it's stopped. Print before/after state."

Mention the language if you have a preference, the target OS, and any conventions you follow (header comments, exit codes, logging style). If you've set custom instructions, Autobot already knows your defaults and will follow them.

Saving into your library

When Autobot creates a script, you'll see an inline tool badge in the chat — e.g. "Creating script: ping_servers.py" — and the script is saved into your Script library as it happens.

After that:

• Open the Scripts page (or the Script Editor) and the new script is there, owned by you.
• It behaves exactly like a hand-written script: you can edit it, run it against a server from the editor terminal, or reference it from a workflow Action node.

Review before you run

Autobot generates a first draft. Read the script body before executing it — especially anything destructive (deletes, service restarts, rm). Whether you run it from the editor terminal or ask Autobot to run it, the review is on you.

Iterating on a script

You don't have to get the prompt perfect the first time. Keep refining in the same thread:

"Now add a --timeout argument that defaults to 2 seconds."

"Make it write the unreachable hosts to a file instead of stdout."

Each turn updates the same conversation context, so Autobot remembers the script it just wrote and edits it in place rather than starting over.

Running a script from chat

In execution mode, you can ask Autobot to run a script — not just draft it. It executes the script on a target server and shows you the result inline.

A script run needs four things wired up: the script, the server, the vault, and the credential. Name them precisely and Autobot resolves them for you:

"Run disk-check.py on prod-db-1 using the db-admin credential."

The run is asynchronous: Autobot launches it and hands you a status card that polls for the outcome (pending → running → completed/failed), with the logs available once it finishes. Unlike the editor terminal, chat-initiated scripts don't stream token-by-token — you get a result card instead.

Execution mode is gated

Running a script requires execution mode, which is only available with your own LLM key connected. In Research or Generation mode, Autobot drafts and saves the script — but won't run it. You can always run from the editor terminal instead.

The full mechanics — modes, the Run Panel, investigating failures, and the security model — live on the Execution copilot page.

What's next

• Execution copilot — running scripts and workflows, and investigating failures.
• Workflow actions — chain scripts together into a runnable graph.
• Script Editor guide — where generated scripts live, and how to edit and run them.
• Action Nodes — how a workflow node runs a script.