Escalated work needs a return path back onto the line

The smallest automation in the factory at software-factory.dev does exactly one thing. It is worth a closer look because it is so small.

One question every two hours

It is called the Needs-Human Requeue, and it runs every two hours on the cron line 0 */2 * * *, with a manual trigger as a backup. Each run it lists every open issue parked with a needs-human label and checks one thing on each.

Has a real person finally answered?

When an automation cannot finish a piece of work, it tags the issue needs-human and the issue leaves the line. It sits in a pile. At some point a human reads it and writes a comment with the missing context. And then, in most setups, nothing happens, because the next step has no owner.

This automation is the next step. When it finds a human comment, it removes the needs-human label and posts a short note that the issue will be picked up on the next triage run. That is the entire job.

Counting only the answers that count

The detail worth copying is how carefully it decides what counts as a human answer. It reads the issue timeline to find the exact moment the needs-human label was added, and only counts comments posted after it, so an old comment from before the escalation never triggers a false requeue. And it tells a real person from a bot by the author.

Teams build the way out and skip the way back

Here is the pattern I want teams building their own factory to see. Almost everyone builds the escalation path, the way work leaves the line and reaches a human, because that path is obvious the first time an agent gets stuck. Almost no one builds the return path. So escalated work rots in a pile that never gets checked again.

A handoff has two directions. Machine to human gets built because the pain is loud. Human to machine gets skipped because the pain is silent, a slow leak instead of a crash. This automation earns its place by owning the quiet direction completely.

The return path is not a nice-to-have. It needs a name and an owner, like everything else on the line.

You can read the whole automation in the Needs-Human Requeue definition on GitHub (opens in a new tab).

In your setup, what actually happens to a ticket after a human answers the question that got it escalated?

Further reading

• Human accountability →
• Agentic Software Factory: Advanced Team Training →