Feeding it work
Harmonic runs the tickets you tell it to. This page covers how tickets get onto the board and the one decision that controls what happens next: whether a ticket is a human’s to do or an agent’s.
Tickets come from your tracker
Section titled “Tickets come from your tracker”Once a Workspace’s tracker is connected, Harmonic polls it and mirrors every open issue onto the board, one card per issue, kept in sync on each poll. The tracker stays the source of truth: edit an issue there and the card follows. You don’t maintain anything twice.
You can also add a one-off task by hand for work that isn’t worth a ticket; those get a review gate you control before anything merges (see Reviewing & merging). To work with an agent interactively instead of queuing it, use a Conversation.
The label decides who does it
Section titled “The label decides who does it”Every mirrored ticket is either for an agent (Harmonic runs it unattended) or for a human (it waits for you). Harmonic reads that from the ticket’s labels:
| A ticket labelled… | …is | and Harmonic |
|---|---|---|
ready-for-agent, research | for an agent | starts running it as soon as there’s a slot |
ready-for-human, grilling, prototype, or a plain decision | for a human | shows it on the board but never runs it |
When the labels are ambiguous, Harmonic leans toward running it rather than letting it sit, then hands it back if it turns out to need you.
The labels aren’t Harmonic’s invention, they’re the ones the Matt Pocock’s Skills already apply as they turn a spec into tickets. A ticket those Skills marked ready for an agent runs with no extra step from you.
Flip it yourself
Section titled “Flip it yourself”You don’t have to go through the tracker to change a ticket’s mind. On the board you can hand a waiting ticket to an agent, or pull one back to work on yourself, and Harmonic respects the change on the next poll without fighting the tracker over it.
Controlling how much runs at once
Section titled “Controlling how much runs at once”Tickets that are ready for an agent don’t all start at once. The Auto-Runner starts them in order, and you have three dials:
- Priority (high / normal / low) sets which ready ticket goes next; ties break oldest-first.
- Concurrency caps how many agents a Workspace runs at the same time, under a machine-wide ceiling so one Workspace can’t swamp the host.
- The master switch pauses or resumes all automatic running at once, across every Workspace, when you want hands on the wheel.
When a ticket can’t finish on its own
Section titled “When a ticket can’t finish on its own”Two things take a running ticket off the automatic path:
- It needs a decision. If the agent hits a permission prompt or a question it can’t answer, Harmonic stops, flags the ticket “escalated to human,” and leaves it for you. It won’t auto-run again until you take over.
- The work didn’t hold up. If the agent’s own review rejects the change, or it finishes without actually resolving the ticket, Harmonic tries a fresh run up to a set limit, then escalates rather than retrying forever.
Either way the ticket comes back to you clearly marked, never silently dropped. See Watching & steering the fleet for picking those up, and Reviewing & merging for what happens to the code when a ticket does pass.