The run lifecycle
Every state a run can be in, what the live session shows, and what cancel, start over, and rework each do.
A run is one execution of a workflow for one task. A task keeps an ordered history of runs — at most one active at a time — so nothing about earlier attempts is ever lost.
States
| State | Meaning |
|---|---|
pending | Created, waiting to start — usually queued behind the concurrency cap; starts by itself when a slot frees. |
running | A phase is executing in its sandbox. |
awaiting_answers | The agent asked mid-run questions; the session is held open for your answer. |
awaiting_preview | A preview step is serving the built result for you to try; the sandbox is held open. |
awaiting_review | Parked at a gate. Nothing is running and nothing is held — a run can wait here for days. |
completed · failed · canceled |
Terminal. A failed phase first parks at its gate so you can retry or send it back with notes. |
The Runs list groups by exactly these: Running, Queued, Pending answers, Preview ready, Awaiting review, Failed, Done, Canceled.
The live session
Open any run and you watch the work as it happens, on every surface — editor, dashboard, desktop, phone. The transcript shows the agent's narration and thinking, every tool call (file edits as diff cards, commands with output), the exact prompt each step sent, check results (failures open, passes fold away), gate and question panels, published artifacts, and per-step time, token, and cost chips. Opening a run late replays the full history, then tails live.
Three different "do it again"s
- Request changes (rework) — re-runs the current phase in the same run, with your note as feedback, on the same branch. This is the everyday loop.
- Start over — mints a fresh run from the first phase
on a fresh branch (suffixed
-r2,-r3…). The old run, its branch, and its log stay in the task's history. Starting over while a run is active cancels it first. - Cancel — terminal, available from every state. Stops any in-flight work and closes the open gate. Cancel never re-runs anything.
Cost & usage
Each phase reports tokens split into disjoint buckets — input, output,
cache reads, cache writes — so run totals are honest sums, comparable
across agents. Claude reports exact dollars; Codex doesn't report cost,
so Peractor estimates from a rate table and marks any total that includes
an estimate with ~$.
Queued, not rejected. Past your plan's
concurrency cap, new runs (including reworks and retries) enter as
pending and start automatically — you never lose a click to
a busy moment. Only the monthly quota hard-stops new runs.