Dark mode
How automation works

How automation works

Every automation in Helpin can be read as one sentence: when something happens, run an agent on something. This page explains that sentence and what happens between an event arriving and an agent finishing its work.

One sentence per flow

A flow answers five questions. The flow editor lays them out top to bottom so the flow reads like a sentence.

Part

Question

Example

When

What starts the flow?

A GitHub pull request merges

If

Which events count?

Repository is acme/web, base branch is main

Then

What should happen?

Start an agent run

Using

Which agent does the work?

Your review agent

On

What does the agent work on?

A specific repository

Some flows don't use an agent. A flow that starts when a task enters a workflow state can instead move the task to another state or merge its working branch.

From event to outcome

Each time something happens in your workspace, Helpin follows the same chain:

  1. An event happens. For example, a task moves to In review, a release publishes, or a schedule comes due.

  2. Flows are checked. Helpin compares the event with each active flow's trigger and filters. If a flow has a semantic condition, Helpin checks that too.

  3. The result is recorded in Activity. An agent flow either starts a run, is skipped or fails before it can start.

  4. An agent run starts. The agent works on the target you chose. It might pause to ask for your approval or input.

  5. The run finishes. You can open the run to read the transcript, see what the agent changed and check for errors.

Why an event doesn't always start a run

Not every event produces agent work, and that's expected:

  • The event doesn't match any active flow, for example because a branch filter excludes it.

  • A semantic condition doesn't clearly match, so Helpin skips the action.

  • The agent is already working on the same target, so a duplicate run is skipped.

  • The flow can't launch, for example because its agent was removed or can't work on that target.

Activity shows skipped and failed attempts, so you can find out why a flow didn't run. See Activity statuses.

Manual runs follow the same path

When you start an agent yourself, for example from a task or from Agents, the run appears in Activity alongside flow-triggered runs, with Manual as its source.

Was this article helpful?