Feature request: configurable agent view — project scoping, group sessions by repo (fold .claude/worktrees), settings surface like statusLine

This is exactly the gap I keep running into when running 5-10 parallel sessions. The interleaved-rows problem is the worst part: you open agent view and immediately have to play "which cwd is this session in" before you can do anything useful. The lack of worktree grouping especially hurts because you can have 3 sessions all working on the same repo's worktrees and they look completely unrelated in the list.

The agents.scope: "project" | "all" config key would cover most of my daily annoyance. Right now I have a wrapper alias that does claude agents --cwd $(git rev-parse --show-toplevel) from whatever directory I'm in, but that breaks the moment I want to pull in a related session from a linked repo.

One more thing that would help alongside grouping: a short truncated cwd path shown as a subtitle on each row (just the last two path segments is enough). Even without full grouping, that single change would let you scan the list without opening each session.

Upvoted. Hope this gets some traction.

The session-scoping gap becomes a real friction point once you're running more than two or three projects at once. Right now the agents view is essentially a global process list -- useful, but the signal-to-noise ratio degrades fast when unrelated projects mix in.

The worktree grouping request (folding .claude/worktrees/<name> under the parent repo) is the one I'd prioritize. Git worktrees are already a first-class Claude Code pattern for parallel-branch work, and the session rows for worktrees look identical to main-branch sessions unless you manually inspect the cwd. Grouping would make the agent hierarchy readable at a glance.

The agents.* settings key proposal aligns well with how statusLine already works -- that pattern (user-owned JSON config that persists across launches) is the right model to follow. A agents.scope or agents.group_by key would handle most of the use cases here without requiring a more complex UI.

One addition worth considering: a per-session label or tag (set at dispatch time) that's shown in the view. Even a short prefix would let users impose their own grouping without waiting for native project scoping.

The --cwd flag works but it's a workaround for a missing feature rather than a real solution. The data model already has everything needed - cwd per session is in claude agents --json right now. The missing piece is just the presentation layer filtering and grouping it.

The worktree grouping point is the one that gets me the most. When you run parallel work across .claude/worktrees/<branch-name>/, each worktree session appears as a peer row with no visual connection to the parent repo. You end up reading session names to figure out what belongs together, which defeats the purpose of a session list.

A few things that would actually solve this in priority order:

1. Group sessions by repo root (walk cwd up to find .git, use that as the group key). Worktrees under <repo>/.claude/worktrees/ collapse under the same repo header.
2. Persist scope preference per-cwd in settings.json so you don't have to pass --cwd every launch.
3. A "project" column in the row (just the last path segment of the repo root) would be enough in the interim even without grouping.

The claude agents --json output already has everything needed for (1) and (3). This feels like a display change rather than a data change.