Repository navigation
|
What are the most effective ways to review a number of sessions to improve my setup? I look at it in layers For the execution (rig) layers, I'm curious what are the possibilities to assess handover, transitions, seat setup. My usual questions are:
Then, I move to the agent layer where I have multi-client execution with openring (thanks!) and I'm also wondering how you have being approaching it. How to read the sessions and draw insights from them considering the (beautiful) new world? Anyone already connecting the analysis in a particular way? |
Replies: 1 comment 2 replies
|
Thanks for the kind words, @smota. The rig layer. These commands cover most of your three questions:
On seat design, no command decides it. What we look at is whether each seat's queue carries real work, and whether two seats keep doing the same job. The agent layer. What works for us: each coding agent keeps its own session record on disk (Claude Code under We'd be glad to hear how others connect the analysis, too. |
Thanks for the kind words, @smota.
The rig layer. These commands cover most of your three questions:
rig usage top --window 2dranks seats by the tokens they used over a window, andrig usage series --seat <session>shows one seat over time (<session>is a seat's session name, such asdev-impl@my-rig). It shows where compute goes rather than idle time directly; a seat that's low there and has no open work inpod-loadis one to look at.rig view show pod-loadcounts the open work each seat holds,rig view show pickuplists claimed work and marks each item as moving or stalled, with the evidence, andrig view show heldlists blocked work.rig q…