Run Tasks · Trama

An approved Proposal leaves its Tasks in To do. The agent builds them when you ask.

Find the Tasks

The Tasks wait in the Active column, each with View on diagram, View requirements and Run. Run all is at the bottom of the column.

The Board with three Tasks in the Active column, Returns after shipping, Declined payment email and Cart total at the bottom, each with a Run button, and the Run all button at the bottom.

Run one Task

Press Run on a Task card: the card gets an animated border, its dot says Agent working, and a conversation for the Task opens in omp Chat. The other Run buttons are disabled with the reason A run is in progress.

The Board while Task T1 runs: its card has a blue border and an orange dot, the other Tasks have disabled Run buttons, and the T1 conversation is open in omp Chat on the right with the agent's first lines.

The agent moves the Task to Done when finished. A fix Task whose code already does what the Model says is closed without changes, with the badge unchanged.

Or run them all

Press Run all: the agent takes the oldest runnable Task, builds it, and asks for the next until none is left, skipping blocked Tasks. The button stays in place, disabled, with a bar filling it and the text Running….

The Board during Run all: the first Task has a blue border, the button at the bottom of the Active column reads Running, and an All tasks conversation is open in omp Chat on the right.

Its tooltip gives a range of the time left, such as ~15–35 min left; past the estimate, the text is Taking longer than expected. Only one Run goes at a time: while it lasts, Run, Run all, Clear, Resolve and Delete are disabled.

Parallel execution

With Parallel Execution on (the default), a Run of several Tasks may start together the Tasks that concern different things, up to Max Parallel Agents sub-agents (20 by default); the agent gives a reason for each choice. Tasks that concern the same thing always run one after the other. Where the agent has no sub-agents, the Tasks run one at a time. If you assigned different models to different Proposals with the gear on their cards, the Run splits by model.

Drifts during a Run

If the code the agent reads or writes does not follow the Model or the Requirements, it tells you through a Drift. A Drift filed during a Run leaves its Task In progress, blocks the Tasks it concerns, and does not stop the Run: the Run goes on with the others. See Resolve Drifts.

With Advisor Scanner on (the default), an advisor checks the code against the Model during every Run, using the Scan model: it tells the agent what to fix in the code of the Task being written, and files a Drift for code that already existed. With the setting off, the Run scans the Tasks it is about to build first, and again at the end.

At the end

The Run ends with a final verification of tests, lint and build, then the agent writes the History entries. See The Board and the History.

Taking a Task back

The agent holds a Task while it works on it, and you cannot move it. To take it back, run trama task move <id> doing in a terminal: the Task stays In progress and is yours. The command line reference describes every command.