Rename and pin: the two shortcuts that made me use Claude agents more
I use Claude agents a lot more than I did a month ago. Nothing about the agents changed. What changed is that I can finally find my sessions.
The problem was the index
When you run a lot of sessions, the agents index is where you live. Every session shows up with a name Claude generated from the first thing you asked.
That works for about a day. Then a session that started as "fix the payout test" turns into planning the next three tickets, and the name still says "fix the payout test". Multiply that by fifteen sessions and the index stops telling me anything. Half my problem with agents was just figuring out which session was which.
Get the next note like this by email.
about one a week · no spam · one-click out · or rss
Two shortcuts fixed it
In Claude agents, Shift+# shows the list of commands you can run. I'd skipped past it. When I finally read it, two of them changed how I work:
- Ctrl+R renames a session. I give every session a name I'll recognize later, and I rename it when the work inside it shifts.
- Ctrl+T pins a session to the top. The sessions I come back to all day stay at the top, no matter how many others are running below them.
That's it. Neither one is clever. They're just the difference between a list I can use and a list I scroll through hoping.
One pinned session per project
Here's the flow they made possible.
At work I usually have four or five projects going at once. Each project is a parent ticket in Linear, and every piece of work is a sub-issue directly under it. Only one level deep: I don't let sub-issues have their own sub-issues, because that gets out of hand fast.
For every parent ticket, I start an orchestration session in Claude. Its only job is to stay up to date on that project. I rename it after the project and pin it to the top.
So the top of my index is four or five pinned sessions, one per project, with names I picked. Everything else, the sessions actually writing code, sits below them.
When I want to move a project forward, I open its pinned session and ask:
What's the status of this project? Can we kick off any new work in separate agent sessions using the factory skill?
It checks Linear and the open pull requests, tells me where things stand, and starts new background sessions for anything that's ready. Each of those runs one ticket through the factory. The orchestration session never writes code itself. It just knows the project.
This has worked really well so far. The project's context lives in one place I can always find, and I don't have to rebuild it in my head every time I switch between projects.
Where agentctl still wins
I haven't dropped agentctl. The last mile of a pull request still happens there.
Once a PR is open, the work is small and fiddly: a failing test, a merge conflict, a CodeRabbit comment to answer. agentctl shows me exactly where each PR is, and kicking off an agent to fix one of those things is one step. The orchestration sessions are good at "what's next for this project". agentctl is still better at "what does this PR need right now".
So it's two tools for two altitudes. Pinned sessions for projects, agentctl for pull requests.
The lesson
I kept thinking the next improvement in my agent workflow would be a smarter agent or a better prompt. This one was naming things and putting the important ones at the top. When you're running a lot of sessions, being able to find them matters as much as what's in them.
Liked this? Get the next note by email.
about one a week · no spam · one-click out · or rss