Repository navigation
Seats of the same role that learn together: pod-scoped skills grown from repeated procedures #2919
cavalcanticampos
started this conversation in
Ideas
Replies: 1 comment
|
Welcome, @cavalcanticampos, and thanks for the idea and the numbers behind it. As you describe it, a second agent in the same role learns alone today. In your fleet, a daily job that turns a procedure repeated in 3 or more tasks into a skill every worker loads saw one skill loaded 67 times in a week, while 16% of sessions read the written-down decisions. It's received. For a feature like this we confirm the fit first, before anyone builds it, and your question, core or a documented pattern, is part of that. If it fits, would you want to build it yourself? Either answer is fine. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
What I'm trying to do: scale a role by running more seats of it, without each copy learning alone.
What stops me today: a duplicated seat is a separate folder. What
dev-2learns (a query that answers a recurring question, how to read a given log, a repo's quirk) stays indev-2. The culture file is shared, but it's written by people, not grown from the work. So adding seats adds throughput and dilutes what the role knows.What has worked for us (a fleet outside OpenRig, each task a fresh session, so we had this problem from day one): a daily job reads the task reports of the last 7 days, finds a procedure that repeated in 3 or more tasks, and writes it as a skill in a folder every worker loads. At most two new or improved skills per run, only when there's a new report, and each one is shown to a person who can delete it. Last week one of those skills was loaded 67 times. We also measured how often workers actually read the written-down decisions: 16% of sessions. Generated skills get picked up much more than notes, because the harness loads them on its own.
What that could look like in OpenRig:
Is pod-scoped skill sharing something you'd want in core, or better as a documented pattern on top?
All reactions