Runs & actions

Every run in view. Every decision yours.

Every run your agents work stays in view from its first node to its end, and every approval, question and fault it raises waits for you in one inbox. ctxflows, the context layer and the workflows for agent-built software, records all of it, whether the engine ran it or your own session did.

Runs

A run is one execution of a workflow, and its view shows where it stands: every node not reached, running, waiting on you, done or faulted, with the agent that ran it.

Redo a node, send work back or cancel the run. A run waiting on you holds no agent session open, so it spends nothing, and every change it makes goes into the journal.

Approvals

An approval is a node where the run waits for a person's decision: human in the loop, exactly where your workflow puts it.

It shows the plan, the review's verdict, the check results from your machine and the branch, and your answer, such as Approve or Send back, decides the path the run takes. A person decides what is built and whether it was met.

Actions waiting on you

Actions are everything waiting on a person, in one inbox: whatever your runs, your workflows or your roadmap need from you.

Approvals, questions, faults, paused workflows and milestones to confirm wait there. Answer in the web app, through the built-in assistant or from your own agent's session.

The morning review queue

The engine works through your ready items overnight, and in the morning Actions holds what needs you.

  1. Overnight, the engine works

    It takes the ready stories and bugs from your board in order, each in its own working copy, and runs each through plan, build and review.

  2. Approvals wait with what you need

    Each finished change waits at its approval with the plan, the review's verdict, the test results and the branch.

  3. You approve one and send one back

    Approve, and the item is done. Send one back, and it returns to the board to be built again.

  4. You answer a question

    An agent asked whether an export should include archived items. You reply, and it carries on from where it stopped.

  5. You redo a fault

    One run reached its time limit. Raise the limit and redo the node, or cancel the run and leave the item where it stands.

Nothing waited for you to be awake, and nothing went past you: every answer is in the journal, with who gave it.

Questions about runs and actions

Can I keep a human in the loop before anything is merged?

Yes. Put an approval before the node that delivers the change, and the run waits there until a person answers. Anyone in the project can answer, and the answer is recorded.

Can an agent approve its own work?

No. An agent the engine started never approves its own work; the approval waits for a person. In your own session your agent acts for you, and an answer it gives on your instruction is recorded as given through your agent.

Where do I answer approvals and questions?

In Actions in the web app, by asking the built-in assistant, or from your own coding agent's session. The run goes on the moment one of them answers.

Does a waiting run cost anything?

No. A run waiting on you holds no agent session open and spends no tokens. The engine works on other items meanwhile.

Can I see the code change in an approval?

The approval shows the outputs, the check results and the branch the change is on. The code itself stays in your repository, so you open the diff where you already review code.

What happens when a run fails?

It waits on you as a fault, showing what went wrong, the node it stopped at and what it had written by then. Redo the node, raise the limit it reached, or cancel the run.

Keep every run in view

Let your agents work, and decide what ships when it's ready for you.

Beta accessSee pricing