8 · Chain jobs into pipelines

What you'll do

Set up a pipeline — a named series of steps that run in order — so finishing one piece of work automatically leads into the next, instead of you starting each one by hand.

When you'd want this

Some work isn't a single job. "Build the feature, then write tests for it, then update the docs" is three steps where each only makes sense after the one before. A pipeline lets you describe that order once and let it run start to finish.

Build a pipeline

Go to the Pipelines screen. Your saved pipelines are listed on the left, and the builder canvas is on the right.

1. Click New Pipeline and give it a clear name — feature-with-tests-and-docs, say.
2. Pick the project it runs on by default.
3. Click Add Step for each piece of work. Each step is one AI job.
4. Drag the grab handle (⠿) on a step to move it up or down — that's the order they'll run in.
5. Click Save.

Tune each step

A step doesn't have to use your default settings — you can shape each one for the job it does:

Setting	What it's for
Title	A short label so you can tell steps apart — "Run linter", "Generate tests".
Prompt template	Start the step from one of your saved prompts (see page 12).
AI / model	Use a fast, cheap AI for a "format the code" step and a stronger one for "implement the feature".
Safety check	Whether this step builds and tests the project before merging. On by default; leave it on unless you have a reason not to.

How the steps connect

By default each step waits for the one before it. Step 2 doesn't start until step 1 has passed its check and merged — so a later step always builds on finished, working code, never a half-done change.

Click the Dependencies tab to see the whole chain at a glance, colored by state:

Color	Means
Green	Finished and merged
Blue	Running now
Gray	Waiting for the step before it
Red	Failed its check or was blocked — everything after it stops and waits

Live step-by-step progress appears inside a running job, not on a standalone screen — open a task that's running a pipeline to watch one step work while the next waits.

Pass information between steps

You can leave blanks in a step's prompt — write them in curly braces, like {feature_name} — and fill them in when the pipeline runs.

• You're asked up front. When you start the pipeline, it asks you for a value for each blank.
• Steps hand off automatically. Use a built-in blank like {last_changed_files} and a later step is told what an earlier step changed — so "now write tests for it" knows what "it" is.
• You can still change course. Edit a value in the running task's Status panel before a step begins, the same way you'd steer any job (page 5).

You should now see

• A pipeline you built and reordered on the canvas.
• Steps with their own prompts, AIs, and check settings.
• The Dependencies tab showing which step is waiting on which.
• Blanks filling in from one step to the next as the pipeline runs.

If something's not right

Problem	What to do
I edited a pipeline but a running one didn't change	Edits apply to the next run; a run already going keeps the steps it started with.
A step never starts	The step before it failed or is still running — check the earlier step and the dependencies view.
The dependencies view is empty	Nothing depends on anything else yet — a single job has no connections to show.
A step is stuck on Blocked	The step before it failed its safety check. Fix that step's problem, or approve it by hand, and the pipeline picks back up.

Next

→ 9 · The live terminal — talk to an AI directly when you want a hands-on session.