[beruro]

Problem

An unused-code audit found several production-gated leftovers that nothing called: the unwired orgtrack-graph / orgtrack-sync crates, registered but unused git/GitHub/perf Tauri commands, a deprecated SessionService.resume duplicate, unused agent session/tool TypeScript wrappers, a dead handoff exhaustiveness stub with stale allow(dead_code) attributes, LSP #[tauri::command] macros that were never registered, leftover HTTP session stubs, unused search/repo/mobile TypeScript wrappers, superseded usage-dashboard IPC, and unused LSP State wrappers after the macros were stripped.

Solution

Remove those unused surfaces in nine stacked commits (kept separate for review):

1. Drop unused orgtrack graph/sync crates and sync_export adapters.
2. Unregister unused git_commit, github_fetch_profile, and perf JSON/hash IPC; keep the library helpers.
3. Delete SessionService.resume and unused hosted-session factory exports; keep resumeCli and isHostedFromSearchParams.
4. Delete unused agent session/tool TypeScript wrappers; keep live RPC procedures and wrappers that still have callers.
5. Delete the unused handoff exhaustiveness stub and stale admission allow(dead_code) attributes.
6. Strip unregistered LSP #[tauri::command] macros; keep manage_lsp / LspManager as library code.
7. Delete the unused src/api/http/session/local.ts no-op stubs.
8. Delete unused search/repo/lineage/perf/mobile TypeScript wrappers; keep RPC procedures and live named helpers.
9. Unregister superseded usage-dashboard and unused GitHub/skills IPC; delete unused LSP commands/server.rs State wrappers. Keep GitHub library helpers for tests.

This supersedes #1950, #1951, #1956, #1957, #1958, and #1959. SqlEditor cleanup already landed in #1948.

Potential risks

Reviving Database query UI, orgtrack graph sync, GitHub profile/list/clone IPC, Merkle TS wrappers, or the removed TypeScript wrappers requires restoring them from git history. Store removeSession / updateSessionStatus, RPC procedures, resumeCli, manage_lsp, usage_dashboard_overview, and live git/perf/GitHub library APIs are unchanged. GroupChatContext, manage_nodes, and unused SessionService methods were left in place because they need product confirmation or a coordinated ChatPanel refactor. cargo check -p org2 was not rerun here because sidecar binaries are missing in the worktree.

Verification

Cherry-pick of the original six commits onto current develop (5984416aa, including #1948) applied cleanly. The three follow-up commits were made on the stacked branch.

• pnpm exec vitest run --config config/vitest.config.ts src/engines/ChatPanel/hooks/useWorkspaceChat/useWorkspaceChat.test.ts — 4 passed
• pnpm exec vitest run --config config/vitest.config.ts src/hooks/fileReview/__tests__/useFileReview.test.ts — 2 passed
• pnpm exec vitest run --config config/vitest.config.ts src/modules/WorkStation/CodeEditor/Panels/EditorPrimarySidebar/content/SearchContent/useSearchContent/__tests__/useSearchResults.loadMoreListeners.test.ts src/services/navigation/CodeNavigationService.test.ts src/modules/MainApp/Settings/sections/__tests__/MobileRemoteSettingsSection.relayPreset.test.ts src/api/tauri/usageDashboard/index.test.ts — 4 files, 48 passed
• cargo test -p agent_core --lib coordination::agent_org_turn_contexts — 24 passed
• cargo check -p lsp — passed after deleting commands/server.rs
• Commit hooks: lint-staged / oxlint / eslint / prettier / staged TypeScript / clippy on agent_core and lsp (org2 clippy issues in unstaged files ignored)

NOT RUN on the stacked branch: full pnpm test / pnpm typecheck / cargo test -p agent_core / cargo check -p org2 / browser, Wingman, or GitHub clone UI.

[Harry19081]

Went through all six commits against develop. Everything removed here is genuinely unused: no frontend or Rust callers for the orgtrack graph/sync crates and sync_export adapters, the 13 IPC registrations, the TS wrappers (every remaining call goes through rpc.* procedures), or SessionService.resume. The four AgentOrgTurnAdmission constructors all have non-test callers, so dropping their allow(dead_code) is safe, and TaskExecutionHandoffState::as_wire already matches every variant with no wildcard, so the exhaustiveness stub was redundant.

The main issue: in three places the PR unregisters something but keeps a body that nothing calls, and describes it as a "library helper". Inline comments below. Two more that aren't on diff lines:

• perf_utils still carries the IPC surface. All 10 unregistered functions keep #[command] and use tauri::command in perf-utils/src/hash.rs and json_fast.rs, and perf-utils/Cargo.toml:9 still says they are "re-registered from commands/handler_list.inc". Four of them (parse_json_file, parse_json_batch, compute_sha256_bytes, compute_blake3_bytes) have no caller at all. The other six are only called from perf-utils/src/tests/ and src-tauri/tests/perf_pipeline.rs. Deleting both modules also lets sha2, blake3 and simd-json go from perf-utils; nothing else in that crate uses them.
• Stale LSP doc comments now say these are Tauri commands: lsp/src/manager.rs:5 ("the Tauri lsp_start_server command"), manager.rs:572, log_buffer.rs:102, server/lifecycle.rs:187.

Description:

• The tauri-ipc verification line greps for compute_json_hash|json_canonicalize. Neither of those names is in this PR (the removed ones are compute_sha256*, compute_blake3*, *_json_*), so that search couldn't have verified anything.
• Potential risks mentions "Reviving Database query UI", but this PR doesn't touch the Database UI (that was #1948).
• cargo check -p org2 is the build that expands handler_list.inc; CI will cover it, but please confirm it's green before merge.

Follow-up, not blocking this PR: 33 more commands in handler_list.inc have no frontend caller and no Rust caller. Most were orphaned when the frontend wrappers were deleted in earlier unused-export sweeps (243061b, fbd8f0c, cabadcf):

• {cline,kimi,mimo_code,omp,pi,qoder_cli,qwen_code,trae,workbuddy}_recent_paths and external_history_auto_import_recent_paths
• usage_dashboard_{rounds,sessions,summary,trends}
• github_{check_token,clone_repo,create_branch,list_branches,list_repos} and server_check_github_visibility
• agent_session_{apply,delete,enter}_worktree
• browser_webview_{bring_to_front,send_to_back}
• builder_profile_exemplars, get_session_tool_usage_attributions_for_call, get_local_lan_ip, skills_hub_update
• project_cancel_routine_run and project_{list,share,unshare}_org_skill(s): these were added 2026-09-06 and never had a caller, so they may be backend-first work waiting on UI. Worth checking with the author before deleting.

[Harry19081]

With the registration gone, git_commit has no production caller; the only caller is git_commit_treats_clean_tree_as_ok in tests/bundle_tests.rs. That leaves bundle.rs entirely dead: this function plus the private run_git_command it uses. I'd delete bundle.rs, its test module, and the pub use bundle::* line in git/src/lib.rs rather than keep it as a "library helper" that nothing calls.

While you're here: the git2 comment in git/Cargo.toml still says git2 is "only used by bundle::calculate_ahead_behind". That function is gone; git2 is actually used by watch/git_status.rs.

[Harry19081]

This leaves four section comments in a row with nothing under them. "Git Bundle commands" and "Git Bundle Sync commands" were already empty; "Git operations for conflict resolution" becomes empty with this PR. Please drop lines 273–275.

[Harry19081]

See the review summary: the 10 unregistered perf_utils functions still have #[command] on them, and none has a production caller (4 have no caller at all; 6 are only called from tests). They're dead code rather than in-crate APIs, so I'd delete hash.rs / json_fast.rs along with their tests and the sha2 / blake3 / simd-json dependencies.

[Harry19081]

This only holds for part of the module. manage_lsp imports 6 of these (lsp_check_installed, lsp_get_install_command, lsp_get_uninstall_command, lsp_get_supported_languages, lsp_get_workspace_config, lsp_set_server_enabled). The other 17 have no caller anywhere: everything in server.rs, plus lsp_get_cached and lsp_is_server_enabled. (The lsp_did_open in e2e-test/src/lsp.rs is a local helper with the same name.) Stripping the macros turns a dead IPC surface into a dead library surface; I'd delete them instead.

[Harry19081]

Nothing calls any function in this file. They all take tauri::State<'_, LspManagerState>, so they only ever made sense as commands; agent-core goes to LspManager directly. I'd delete the file along with GlobalLspConfig / BrokenServerInfo, which only these functions use.

[Harry19081]

lsp_get_cached has no callers, and it's the only reader of cache::load_lsp. Meanwhile lsp_check_installed (used by manage_lsp) still calls cache::save_lsp on every scan, so the LSP cache becomes write-only disk I/O. I'd delete this function and the cache write, or the whole cache module.

[Harry19081]

unified.ts was the only importer of ./local, so this change orphans src/api/http/session/local.ts: about 140 lines of no-op stubs that exported sessionApi, under a header saying they're "kept for UnifiedSessionApi shape". That file can go in this PR too.