# Re-runs and follow-ups

> Running and debugging · https://spoletta.com/docs/reruns

Two ways to carry on from a run that has ended: re-run it from one of its processes, as a new run on the board as it is now, or send an agent step a follow-up message in the same session and checkout.

*Diagram: Re-running from Fix: what came before is copied as the run left it, and Fix and what follows run anew.*

## Re-run from a process

Once a run has ended, you can re-run it from any process that ran in it (though not from a step it copied from an earlier run): press **Re-run from here** on the process's card while the run is shown, or on its step in the run panel. A dialog, *Re-run from Fix*, opens:

-   Each object the process took is in an editable box, as it was. Change what you want; only changed values are sent. An object the run had nothing of is marked so.
-   *Copied as the run left them* lists what came before the process: those steps are copied into the new run as they were, not run again.
-   *Generated anew* lists what the process and everything after it will generate.

Press **Re-run**. The new run runs on **the board as it is now**, so you can fix a prompt, a script or a link, then re-run from the step that went wrong without repeating the slow or costly steps before it. In the runs list it shows under the original as *Re-run of #R from Fix*.

> **Its own branch:** A re-run gets a git branch of its own, starting from the commit the original run's branch was at when the process started. The original run's branch is left as it was. Re-running inside a flow process's run goes only as far as that board: its output doesn't go on past it.

## Follow-ups to an agent

An agent step that has finished keeps its session and its checkout. Below it in the run panel is **Message the agent**: write to it as you would to a colleague, “push it”, “rename that function”, “add a test for the empty case”, and press **Send** or `Cmd`+`Enter`.

1.  **You ask**
    
    Your message shows as *You asked*, then *The agent is working on it…*, with **Stop**.
    
2.  **The agent carries on**
    
    It continues the same conversation, in the same workspace and on the same branch.
    
3.  **It replies**
    
    *The agent replied*, with what it said, or *Error* if it failed.
    

> **Watch out:** Follow-ups wait until nothing in the run is going: *Wait until what is going in this run is over.* And a follow-up doesn't change what the step generated. Its output stays as it was handed in; to change that, re-run from the step.

## Which to use

| You want to | Use |
| --- | --- |
| Fix the flow and try a step again | Re-run from here |
| Try a step with a different input | Re-run from here, editing the input |
| Have the agent finish off work in its checkout | Message the agent |
| Ask the agent why it did something | Message the agent |