Workflows

A single message can be more than one command. If you start several lines with / or @, Cockpit reads the whole message as an ordered workflow — each command line is a step, and Cockpit assembles them into one numbered plan before the AI starts. It's the fastest way to say "clarify this, then fix it, then have a sub-agent review the fix" without sending three separate messages.

This builds directly on the slash menu and Skills — the same commands, now chainable.

One message, several steps

Each line that starts with a known command becomes a step:

verb is any built-in command (/qa, /fx, /cr, …) or any installed skill of the same name. A verb starts with a letter and may contain letters, digits, and hyphens — so /new-branch and /qa are both single commands.

Here's a three-step message:

Here is the failing test output: payment webhook 500s on retries.
/fx
figure out why the idempotency key isn't being honored
@cr
audit the fix for race conditions and missing rollbacks

Cockpit turns that into a single ordered prompt — roughly:

Here is the failing test output: payment webhook 500s on retries.

Complete the following steps in order:

Step 1 (run in the main session): 
Please read this skill file:
~/.cockpit/skills/fx/SKILL.md
Question: figure out why the idempotency key isn't being honored

Step 2 (run in a subagent): 
Please read this skill file:
~/.cockpit/skills/cr/SKILL.md
Question: audit the fix for race conditions and missing rollbacks

You write four lines; the AI receives a structured plan it works through in order.

How a message is split into steps

The rules are line-based and predictable:

• A command line is any line whose first non-space character is / or @ followed by a known verb. Lines that start with a slash but aren't a real command (/usr/local/bin, @mention) are left as ordinary text.
• A step's body is everything after the verb on that line, plus every following line, up to the next command line. Blank lines and multiple paragraphs are kept — so a step can carry as much context as you want.
• Preamble is any text before the first command line. It's passed through as-is at the top of the plan — a good place to paste an error log or describe the goal once for the whole workflow.

Main session vs sub-agent — / vs @

• /verb keeps the work in the current chat. Use it for steps you want to watch and steer turn by turn.
• @verb hands the step to a sub-agent. Use it for self-contained work — a review pass, an exploration, a focused investigation — that you want done and summarized without cluttering the main thread.

A common shape is "do the work in the main session, then send a sub-agent to check it":

/go
implement the retry backoff described in the ticket
@cr
review what was just written for correctness and style

Built-ins and your own skills, mixed

A workflow can freely mix built-in commands and your installed skills — they resolve through the same "read this SKILL.md" path. If a skill you installed shares a name with a built-in, your skill wins: a /cr you authored shadows the built-in /cr, so your edits always take effect.

Autocomplete follows your cursor

The command menu no longer triggers only at the very start of the box. Type / or @ at the start of any line — including the second, third, or fourth — and the autocomplete dropdown appears for that line, filtered as you type. Tab or Enter inserts the selected command. That's what makes stacking commands line by line comfortable.

When it stays a single command

If your message is just one /verb with no preamble and no @, nothing changes — you get the original compact behavior: the command's skill plus your trailing text, sent as one normal turn. The numbered step list only appears when there's genuinely a workflow to run: two or more commands, any @ sub-agent step, or leading preamble text.

Next

• Skills — the built-in commands and how to install your own
• Message Input — everything else the message box does
• Sessions — running multiple chat tabs