[Harry19081]

Problem

Switching branches only asked about uncommitted work after checkout failed, so compatible edits could move without a choice. The native stash/discard modal offered no Bring option, and optimistic branch updates could show the wrong branch after a partial failure. Positional stash indices and fire-and-forget editor saves also made recovery unreliable.

Solution

• Replace the native switch modal with the existing shared Modal and SelectionGrid using vertical, showRadio, and visible title/description rows for Leave and Bring.
• Route existing-branch selection, branch creation, detached checkout, ActionSystem and composer actions through one coordinator that prepares before mutating and publishes actual HEAD.
• Add a backend-owned, revalidated save/checkout/apply operation with an OS lock per common Git directory. Preserve staged, unstaged and nonignored new files using UUID metadata, immutable snapshot OIDs and recovery refs.
• Expose saved-work discovery, bounded diff previews and explicit restore in Source Control; preserve earlier snapshots and partially applied recovery copies.
• Await text/diff editor saves, check disk content, block known active writers and unsupported/unsafe Git states, and translate the new UI in all 13 supported locales. Background refreshes explicitly handle promise rejection without widening lint baselines.

Potential risks

• Additive HTTP endpoints and JSON journals introduce a new recovery workflow. No database migration is required. fs2 is added as an explicit dependency for OS locking at a version already present in Cargo.lock. App rollback leaves Git stashes and refs/orgii/branch-switch/<UUID> recovery refs usable; do not delete them during rollback.
• The lock coordinates new switch/restore operations. External Git clients and legacy mutation routes remain independent; the active-task check uses the known ORGII roster rather than an exclusive writer lease. External writers still require coordination.
• Recovery copies are retained until deliberate cleanup and consume disk. Partially applied operations are not automatically replayed. Journal recovery is tested with interrupted-phase fixtures, not abrupt process-kill testing.
• Native Windows behavior, real Tauri light/dark and constrained-viewport layout, keyboard/focus behavior, and CPU/RSS measurements are unverified. Computer control was not authorized, so native screenshots/recordings were not captured. DOM tests exercise real shared components but are not visual evidence.

Verification

Checks ran in an isolated worktree based on develop; unrelated staged work was excluded. The latest target branch was fetched and the PR remains mergeable.

• pnpm exec tsgo --noEmit --pretty false — passed the full project check. The earlier shared-checkout type errors are absent in this isolated branch.
• Frontend command below — 71 tests passed, covering the coordinator, API serialization/read sharing, real Modal/SelectionGrid behavior, banner lifecycle, editor saves, buffer recovery and existing cache behavior.
• cd src-tauri && cargo test -p git_api --lib commands::branch_switch — 25 passed using real temporary Git repositories, including Leave/Bring, index preservation, conflicts, collisions, detached/remote targets, worktree locking, stale preparation, journal recovery, pagination and symlinks.
• cd src-tauri && cargo test -p git --lib util:: — 3 passed, including timeout kill/reap and draining output larger than pipe capacity.
• Commit hooks ran without bypass: lint-staged (ESLint and Prettier), full TypeScript checking and cargo clippy --lib --message-format=short -p git -p git_api.
• git diff --check and scoped source/diff review — passed. Confirmed the PR contains only branch-switch code, localization, tests and audit notes; no new raw action controls, sensitive content or unrelated generated changes.
• Native UI, production application build and performance measurements were not run; limitations are listed above.

pnpm exec vitest run --config config/vitest.config.ts \
  src/api/http/git/branchSwitch.test.ts \
  src/components/GitDialogs/CheckoutConflictDialog/index.test.ts \
  src/components/GitDialogs/SavedChangesDialog/SavedChangesBanner.test.ts \
  src/services/git/operations/__tests__/guardedCheckout.test.ts \
  src/services/git/operations/__tests__/guardedCheckoutRecoveryFallbacks.test.ts \
  src/services/git/operations/__tests__/branchSwitchEditors.test.ts \
  src/services/git/operations/__tests__/branchSwitchWorkspace.test.ts \
  src/store/workstation/codeEditor/gitDiffEditDrafts.branchSwitch.test.ts \
  src/modules/WorkStation/CodeEditor/hooks/fileContent/__tests__/cache.test.ts

CI correction verification

• pnpm check:i18n-keys — passed: no missing keys, locale gaps, or placeholder mismatches; no new unused keys.
• NODE_OPTIONS=--max-old-space-size=6144 pnpm check:typed-lint — passed: 1,099 existing findings, zero new or increased findings. Baselines unchanged.
• node scripts/ci/run-unit-tests.mjs < /tmp/orgii-pr1836-unit-paths — passed using the NUL-delimited PR diff plus the correction paths: 653 test files, 4,766 tests passed, one skipped.
• pnpm test:typed-lint and pnpm test:i18n-keys — passed: 6 and 27 tooling tests respectively.
• pnpm exec eslint --max-warnings 0 src/components/GitDialogs/SavedChangesDialog/index.tsx src/components/GitDialogs/SavedChangesDialog/SavedChangesBanner.tsx — passed.
• Corrective commit hooks passed lint-staged and the staged-file TypeScript gate; no Rust files changed.

Audit

UI audit: 0 fix, 8 keep with reason, 0 abstract. All ten architecture layers reviewed. Automated lifecycle checks pass; the performance verdict remains blocked for unrun native measurements.

• Architecture and recovery review
• UI audit
• Lifecycle/performance audit