Problem
The current resume session picker groups sessions by local vs remote. When working across multiple projects, finding "the session I had in this repo" requires scanning all local sessions regardless of which directory I'm currently in.
Proposed solution
Add a primary grouping by current working directory. Sessions started in the current pwd (or any subdirectory of it) should appear first in the picker as their own visually distinct group. All other local sessions follow, ordered by recency.
Conceptually, the picker would render something like:
This directory (/path/to/current/project):
- session-abc | "Implementing the parser" | 2 hours ago
- session-def | "Debugging the test suite" | yesterday
Other local sessions:
- session-ghi | "Refactoring auth module" | last week
- session-jkl | ... | ...
Remote sessions:
- ...
Why this matters
When jumping between projects, the most relevant session is almost always one started in the current directory. Today this requires scanning through unrelated sessions to find the right one, which gets worse as the local session count grows.
Reference
Claude Code handles this well: it surfaces sessions from the current working directory first, making "resume what I was doing in this repo" a one-step action.
Additional notes
- The "match current directory" check could include parent matches (if
pwd is inside a session's working directory) and child matches (sessions started in subdirectories of pwd).
- The existing local/remote split is still useful, this would layer on top of it, not replace it.
Problem
The current resume session picker groups sessions by local vs remote. When working across multiple projects, finding "the session I had in this repo" requires scanning all local sessions regardless of which directory I'm currently in.
Proposed solution
Add a primary grouping by current working directory. Sessions started in the current
pwd(or any subdirectory of it) should appear first in the picker as their own visually distinct group. All other local sessions follow, ordered by recency.Conceptually, the picker would render something like:
Why this matters
When jumping between projects, the most relevant session is almost always one started in the current directory. Today this requires scanning through unrelated sessions to find the right one, which gets worse as the local session count grows.
Reference
Claude Code handles this well: it surfaces sessions from the current working directory first, making "resume what I was doing in this repo" a one-step action.
Additional notes
pwdis inside a session's working directory) and child matches (sessions started in subdirectories ofpwd).