fix(worktree): create branches with --no-track and auto-setup remote

[brennanb2025]

Summary

• New worktree branches are now created with git worktree add --no-track, so an unpublished branch isn't reported as "behind by N" against its base ref by git status or by tools/agents that read upstream state.
• After creating the worktree, set push.autoSetupRemote=true (locally, only when not already set at any scope) so a plain git push from the terminal creates origin/<branch> and sets the upstream automatically — matching modern-git ergonomics without forcing -u.
• Probe with git config --get before writing, and only treat exit code 1 ("key unset") as a green light to write. Any other read failure (corrupt config, locked file, parse error) re-throws to the warn handler so we don't silently overwrite a deliberate user value.
• Config write failures are warn-only and the warning includes the worktree path; old git clients (<2.37) just fall back to the pre-2.37 "no upstream" error and git push -u once.
• SSH-relay parity: the same --no-track + probe-and-write state machine now runs on the relay side (in addWorktreeOp in src/relay/git-handler-worktree-ops.ts, called from GitHandler.addWorktree in src/relay/git-handler.ts), so SSH-mounted repos get the same git status / git push UX as local repos. The optional track field is removed from the provider interface and the SSH IPC payload (one in-tree caller updated).
• Tests cover: default unset path writes true, existing value is preserved (including blank-string values), write failure warns without throwing, a non-unset read error is treated as a real failure (no --local set), and the config block is skipped entirely when worktree add itself fails. Mirror coverage exists for both the local and relay handlers.

Verification

Unit tests: all passing.

• src/main/git/worktree.test.ts (21 tests) — local addWorktree state machine.
• src/main/git/remove-worktree.test.ts (11 tests) — removal path.
• src/relay/git-handler.test.ts (29 tests) — relay handler, including 6 new addWorktree cases mirroring the local matrix (--no-track arg ordering, write-when-unset, preserve-existing, empty-stdout = "already set", non-unset read error short-circuit, write-failure warn-only, no-config-on-worktree-add-fail).
• src/main/providers/ssh-git-provider.test.ts (20 tests) — SSH provider IPC payload (already omits track).

Combined run: 81/81 passing post-merge with main.

Live Electron verification — pre-merge (Orca dev build, repo test-repos/my-git-proj with push.autoSetupRemote initially unset at every scope):

Scenario	Expected	Result
Create worktree, autoSetupRemote unset	New branch has no upstream; git status clean (no "behind"); push.autoSetupRemote=true written locally	All three confirmed
Branch config after creation	No branch.<name>.remote / branch.<name>.merge keys (i.e. --no-track took effect)	Confirmed (git config --get-regexp empty)
Create worktree with push.autoSetupRemote=false preset locally	User value preserved (still false); branch still has no upstream	Confirmed
git status on new worktree	No "behind by N" line	Confirmed (## brennanb2025/... only)
Worktree removal	Renderer state + filesystem clean; branches deleted	Confirmed via removeWorktree(id, force=true)
Renderer console errors during flow	None	0 errors

Live Electron re-verification — post-merge (same setup, after merging origin/main into the branch on 2026-05-11):

Scenario	Expected	Result
Create worktree, autoSetupRemote unset	branch has no upstream; status clean (no "behind"); push.autoSetupRemote=true written locally	All three confirmed
Branch config after creation	no branch.<name>.remote / .merge keys	Confirmed (config --get-regexp exit 1)
Pre-set push.autoSetupRemote=false, then create worktree	preserved as false; branch still has no upstream	Confirmed
git status -b first line	## <branch> only — no [behind N]	Confirmed
removeWorktree(id, force=true)	renderer state + filesystem clean; branches deleted	Confirmed
Renderer console errors during flow	None	0 errors / 189 benign warnings

Test artifacts cleaned up after each round (worktrees removed, branches deleted, push.autoSetupRemote restored to its initial unset state).

Live remote-push verification (against personal fork brennanb2025/orca, fresh worktree at /tmp/orca-fork-autosetup-test-*):

Step	Expected	Result
git worktree add --no-track -b <branch> <wt> origin/main	branch created, no upstream	Confirmed (git status -b shows no branch.upstream / branch.ab line)
git config --get push.autoSetupRemote	exit 1 (unset)	Confirmed (exit 1)
git config --local push.autoSetupRemote true	written to shared common-dir <repo>/.git/config	Confirmed (per the design comment — no extensions.worktreeConfig involved)
Plain git push (no -u) on the new commit	creates origin/<branch> AND sets upstream	Confirmed: [new branch] + branch '...' set up to track 'origin/...'
git rev-parse --abbrev-ref @{u} after push	origin/<branch>	Confirmed

Live SSH-relay verification (driving GitHandler.addWorktree directly against a real bare-origin + working-clone setup — same code that runs on the SSH host; an in-process invocation exercises the entire relay state machine end-to-end since the SSH layer is just message transport):

Step	Expected	Result
Pre-condition: push.autoSetupRemote unset locally	exit 1 from git config --get	Confirmed
git.addWorktree with base: 'origin/main'	worktree dir created; relay invokes worktree add --no-track -b ... origin/main, then probe + --local set true	Confirmed
Branch keys after creation	no branch.<name>.remote / .merge keys (--get-regexp exits 1)	Confirmed
push.autoSetupRemote after creation	true (written by relay)	Confirmed
git status -b --porcelain first line	## <branch> only — no [behind N]	Confirmed
Pre-existing push.autoSetupRemote=false (local scope)	preserved as false; relay does not overwrite	Confirmed
worktree add itself fails (duplicate branch)	rejection bubbles up; no config probe/write occurs	Confirmed

No regressions observed.

AI-agent confusion check (Claude Code spawned in a fresh worktree off main created by Orca after this PR is in effect — the failure mode this PR is designed to fix):

Prompt	What an agent did pre-fix	What the agent did post-fix
"What's the state of this branch relative to main? Do I need to sync before making changes?"	Reads git status -b → sees [behind N] → suggests git pull / git rebase before letting you start	Runs git rev-list --left-right --count, reports "no local work, no divergence" — no spurious sync suggestion
"Run git status -b --porcelain=v2 --branch and tell me what branch.upstream reports. Is this branch tracking anything?"	branch.upstream origin/main would be present → agent confidently reports "tracking origin/main" (wrong-shaped — branch was never pushed)	No branch.upstream line → agent correctly reads "this branch is not tracking any remote branch" and self-corrects its earlier "1 behind origin/main" answer
"Will plain git push work here, or do I really need -u origin <branch>? Check the repo's git config for any relevant push settings before answering."	No push.autoSetupRemote set → agent says you need -u origin <branch>	Finds push.autoSetupRemote=true (local), correctly concludes plain git push will create the remote and set upstream automatically; calls out the git ≥2.37 requirement and notes git pull still needs an upstream until the first push

The agent reasons from observed git state rather than misreading git status as ground truth — both halves of the bug (false-positive "behind" framing and the -u ceremony) are invisible to a calm agent and explainable when probed directly.

Test plan

• pnpm test src/main/git/worktree.test.ts
• pnpm test src/main/git/remove-worktree.test.ts
• pnpm test src/relay/git-handler.test.ts
• pnpm test src/main/providers/ssh-git-provider.test.ts
• Create a worktree from Orca, then in it run git status (no "behind" line) — verified pre-merge and post-merge.
• Set git config --local push.autoSetupRemote false, create a worktree, confirm the local config is not overwritten — verified pre-merge and post-merge.
• First git push from a new worktree creates origin/<branch> and sets upstream automatically — verified end-to-end against brennanb2025/orca.
• SSH-relay path: relay addWorktree produces the same end state as local (--no-track took effect, push.autoSetupRemote=true written, git status shows no "behind", existing user value preserved, no config writes when worktree add fails) — verified end-to-end against real git.
• AI-agent confusion check: spawned Claude Code in a fresh post-fix worktree, asked it to (a) frame branch state vs main, (b) inspect branch.upstream via porcelain v2, (c) decide whether git push needs -u. No spurious sync suggestions, correct read of "no upstream", correct read of push.autoSetupRemote=true — verified.

Made with Orca 🐋