Runs and cost

A run is one execution of a step, and what it cost: the state it ended in, the output as it came, the exit code of a command, the spend and duration of an agent turn, the lane it was sent from, the folder it ran in and the session it ran as.

A card shows its runs. That is how "the agent is doing something" becomes a sentence with numbers in it.

Where they are

On a card, the Runs section lists its own, newest first, with Stop beside one that is still going. For the whole board, Runs in the board toolbar opens the list of every run, filtered by Lane, State, From and To.

The empty states are three different sentences, because they are three different situations: no lane runs anything yet, nothing has run yet, or nothing matches the filter you set.

The states

lost is the honest cost of having no daemon: a run in flight belongs to the window that started it, and at the next launch those rows are closed rather than left saying they are still working. Your terminals are not affected — they are tmux sessions, and they outlive the window.

Cost

Every agent step declares a spending cap, and what the turn actually spent is written on the run and added up on the card. A command step costs nothing of its own; so does starting a session, and what you then spend driving that session by hand is not read back — the card counts its runs and is quiet about the rest.

Usage has the totals by day, model and project, and, project-scoped, this project's cards by what their runs cost.

Secrets devpit handed out — a profile's environment, the account token, the hook secret — are taken out of a run's output before it is stored. That is not detection: a secret the command read from a file or fetched itself is one devpit never saw, and it passes through.

What it proves

A run also records the circumstances it happened in, so a green row can answer green on which code. Expanding a run opens that panel — see Reading a check.