Last updated 4 October 2026
Naming agent sessions: why the agent should name itself
A tab called websta.ai tells you which repo. It does not tell you that this session is halfway through migrating the bookings table, and the other websta.ai tab is writing release notes.
Titles from the work itself
adhdTerminal asks Claude for a two-to-four-word name whenever a tile's output changes, at most every thirty seconds. It gets the last prompt and about 1800 characters of recent output, nothing else. The result reads like Fix webhook retries or Draft pricing copy.
Because the name follows the work, a session that drifts to a new task gets a new name. You see the drift on the wall instead of discovering it later.
When to name by hand
Double-click a title to set your own. A manual title is never overwritten. Use it for sessions where the job is fixed and you want the label to stay put, such as a long-running deploy watcher.
Privacy
The naming request goes through Claude Code under your own account, like everything else in the tile. Turn the toggle off and nothing is sent.
Read next
- How to run five Claude Code sessions at once without losing oneParallel agent work fails at the attention layer, not the compute layer. A practical setup for keeping every running session visible.
- Claude Code resume: picking a session back up after a restartWhat --resume and --continue actually do, where the history lives, and how to make resuming automatic.
- The restart tax: why tabs are the wrong unit for agent workA tab you cannot see is a job you will restart. The hidden cost of hiding running work behind a title bar.
- Full Disk Access and Claude Code on macOS: what the permission doesWhy a terminal app asks for Full Disk Access, what Operation not permitted means, and how to grant it once.