feat(switch): picker shortcuts — copy branch, open PR, refresh, remap remove to alt-x

[max-sixty]

Adds four keyboard shortcuts to the interactive wt switch picker:

• alt-y — copy the selected branch name to the system clipboard
• alt-o — open the selected row's PR/MR in the browser
• alt-r — refresh the worktree list (pick up worktrees created elsewhere)
• alt-x — remove the selected worktree/branch (remapped from alt-r)

Approach

alt-y and alt-o are native skim Action::Custom callbacks that read the selected row off App.item_list and run the OS action on a background thread — no reload, so the cursor doesn't reset and --prs rows aren't dropped. A side table keyed by each row's output() token maps to the row's branch and URL, which avoids skim's unreliable cross-thread downcasting.

alt-r refresh re-enters the collect pipeline via a new PipelineFactory, extracted so startup and refresh build the pipeline identically; worktrees and branches created outside the session then appear without reopening the picker.

The one breaking change is the alt-r → alt-x remap for remove. It fails safe — an old-habit alt-r now refreshes rather than removes.

Key files

• src/commands/picker/mod.rs — PipelineFactory, install_shortcut_keybindings, the refresh branch in invoke, and the bind block
• src/commands/picker/os.rs (new) — copy_to_clipboard / open_url via arboard + open
• src/commands/picker/items.rs — the row-shortcut side table and the controls line
• src/cli/mod.rs — the keybindings table and help text

New cross-platform dependencies: arboard (clipboard, with its image feature dropped) and open (browser launch), both gated behind the cli feature.

Testing

Unit tests cover the keybinding wiring, the row → branch/URL resolution, and the refresh/remove dispatch; PTY tests cover the alt-x sticky-cursor reposition and the unmerged-branch-keep path. The OS side effects themselves — the clipboard write and the browser launch — are not asserted (both are headless-hostile in CI), so they want a manual check on a desktop; they work on macOS, with the one documented limitation being X11 without a clipboard manager.

This was written by Claude Code on behalf of max

[worktrunk-bot]

One edge case on the new alt-y copy, inline. Otherwise this looks solid — the alt-r→alt-x remove remap leaves the destructive primitive (SafeDelete via delete_branch_if_safe) untouched and moves remove to a less-reflexive key, so an old-habit alt-r now refreshes rather than removes, which is the safe direction.

codecov/patch is failing on this commit — per CLAUDE.md, patch coverage gates merge. Dismissing until the coverage gap is addressed or merging-with-gap is explicitly approved. See follow-up comment for the breakdown.

[worktrunk-bot]

codecov/patch is failing on this commit, so I'm dismissing my approval — per CLAUDE.md, patch coverage gates merge even though GitHub marks it non-required, and an empty-body approval over a red patch check reads as a rubber-stamp. codecov/project passes; this is purely the patch (new-lines) gate.

The gap is concentrated in picker/os.rs and picker/mod.rs, almost entirely in the interactive/OS-effect surface you already flagged in the PR description as headless-hostile:

File	Misses	What's uncovered
picker/os.rs	9	all of copy_to_clipboard / open_url
picker/mod.rs	43	the alt-y / alt-o Action::Custom closures, spawn_shortcut, and the refresh → factory.spawn() arm
picker/prs.rs	1	the GitLab arm of spawn_worktree_comments_fetch

Most of this is genuinely irreducible — os::copy_to_clipboard / os::open_url and the spawn_shortcut thread wrapper can't run in CI, exactly as the description says. But the bulk of the mod.rs miss is the row → branch/URL resolution inside the two closures, which is pure logic that doesn't need a live App:

table.lock().unwrap().get(m.item.output().as_ref()).and_then(|d| d.branch.clone())

Lifting that lookup out of each closure into a small free fn (e.g. resolve_shortcut_branch(&table, token) -> Option<String> and the url.resolve() equivalent) would let a unit test cover the part that actually matters — the side ShortcutTable keyed by output() token, and the detached-worktree no-op — leaving only the thin spawn_shortcut(os::…) call uncovered. That's likely enough to clear the patch threshold, and it directly tests the lookup the shortcut_table_branch_is_none_for_detached_worktree test only checks one half of today.

I can push that refactor + tests if you'd like — say the word and I'll add it to the branch. Alternatively, if you'd rather merge with the OS surface deliberately untested, that's your call to make explicitly (the codecov gate exists to force exactly that decision); I just won't leave a silent approval standing over the red check.

Uncovered lines (from the Codecov compare API)

• os.rs: copy_to_clipboard + open_url bodies
• mod.rs: the alt-y closure (.and_then(|d| d.branch.clone()) → spawn_shortcut("picker-copy", …)), the alt-o closure (url.resolve() → spawn_shortcut("picker-open", …)), spawn_shortcut itself, and match self.factory.spawn() { Ok(SpawnedPipeline { rx, .. }) => … } in the refresh branch
• prs.rs: Some(CiPlatform::GitLab) => RefKind::Mr

[max-sixty]

Implemented your suggested refactor in d20aadb — pulled the row → branch/URL lookup out of the alt-y / alt-o closures into resolve_shortcut_branch / resolve_shortcut_url and unit-tested them (present / detached-no-branch / no-URL / missing-token). Only the irreducible OS surface (os::copy_to_clipboard / os::open_url and the spawn_shortcut thread wrapper) stays uncovered, since headless CI can't drive a clipboard or browser. Re-running codecov now.

This was written by Claude Code on behalf of max

[worktrunk-bot]

The refactor did its job — resolve_shortcut_branch / resolve_shortcut_url are now fully covered (they've dropped out of the patch miss list), so the row → branch/URL lookup the alt-y / alt-o closures delegate to is unit-tested across the present / detached / missing-token cases.

codecov/patch is still red, though, so I'm dismissing my approval again — same reason as before (CLAUDE.md treats patch coverage as a merge gate despite the non-required mark, and an empty-body approval over a red check reads as a rubber-stamp). What's left (42 missed patch lines) is now confined to the surface headless CI genuinely can't drive, exactly as the PR description called out:

File	Misses	What's uncovered
picker/os.rs	9	copy_to_clipboard / open_url — clipboard + browser
picker/mod.rs	30	the alt-o closure body, spawn_shortcut thread wrapper, and the refresh → factory.spawn() arm — all need a live skim App / pipeline
picker/progressive_handler.rs	1	the comments_fetched.get() → return guard in maybe_spawn_comments (async slot state)
picker/prs.rs	1	the GitLab arm of spawn_worktree_comments_fetch — needs a live PreviewOrchestrator + glab

So this is now the deliberate "merge with the OS/interactive surface untested" call the codecov gate exists to force — your decision to make explicitly. I just won't leave a silent approval standing over the red check. If you'd rather I take a swing at any of the residual (e.g. threading the ci_platform branch in spawn_worktree_comments_fetch through a unit test for the GitLab arm), say the word.

Codecov compare API, pullid=3233

Overall patch: 906 / 948 hits (≈95.6%). Misses by file from api.codecov.io/.../compare/?pullid=3233: os.rs 9, mod.rs 30, progressive_handler.rs 1, prs.rs 1, items.rs 1.

codecov/patch failing on d20aadb — patch coverage gates merge per CLAUDE.md. Residual gap is the irreducible OS/interactive surface (clipboard/browser, keybind closures, GitLab orchestrator arm); merging over it is the author's explicit call.