What did your agents do this week?

Agents work while you are somewhere else, which is the point of them - and it makes Friday afternoon strange. Plenty happened this week, and almost none of it happened in front of you. Before you plan next week, it is worth five minutes to find out what your fleet did, and where your own instructions made it harder than it needed to be.

Open History and set it to a week

In the Cockpit, choose History, then Week. The line beside the buttons is the first real answer: how many sessions ran, across how many repositories. It is usually higher than people expect, which is reason enough to look.

Work is grouped under the repository it happened in, and then by day. That grouping answers "where did my week go" rather than "what was session 101". It covers every repository and machine in your fleet, including the sessions running right now, so it is a fair picture rather than a tidy one.

What each entry tells you

• The line it leads with. A session on a mission shows the mission and its role on it; otherwise the first instruction it was given; and only if it has neither, its name and repository. Day to day that means the instruction - which is far more useful than a name, because a name is what you hoped a session would do and the instruction is what you asked.
• Turns. 7 turns counts the agent's completed turns: each stretch of work that ended with it stopping for input.
• Needed you. needed you 3x counts how many times the session started waiting on a person.
• What it left behind. Open an entry and, where the record has them, you get a summary, what was built, what was left unverified, and the branches, pull requests and commits. A session that started other sessions shows them nested underneath it, open by default, so you can see what one piece of work set off.

The number worth reading is needed-you

It is a raw tally, not a diagnosis. A real question, a permission prompt and plain finishing all count the same way, and the number never says why the agent stopped. Treat it as a pointer.

But it is a good pointer. Find the entry with the worst needed-you count and read the line it leads with. If it shows a mission rather than an instruction, take the worst entry that does show one. Often the instruction left something open - which file, which approach, what done looks like - and the agent came back to ask, more than once. Rewrite that instruction in your head until the opening is closed.

That is the whole review. Not a report, one improvement. Next week the same kind of job needs you less, and the week after you find the next one.

Make the improvement stick

Do this on a Friday with the week window open. A single day tells you almost nothing; over a week, the pattern in your own prompts becomes obvious. And when the fix is a rule you would give every session, do not keep it in your head - put it in a workflow, so it is followed rather than retyped. See workflows.

The step-by-step version of this review is What did I get done this week?.