13 KiB
13 KiB
Galaxy UI Flow Inventory
Started: 2026-08-04 Status: First-pass surface classification; action-level trace audit in progress Architecture:
galaxy-local-first-rig.md
How this ledger is used
This is the source of truth for deciding what Galaxy keeps, rebuilds, or removes. A directory name is not a product decision. Each user intent is traced from every entry point through state, persistence, runtime/network dependencies, and rendered outcomes.
Audit sources include:
- root/onboarding states in
app/src/root_view.rs; - registered workspace actions in
app/src/workspace/action.rs; - app menus, command palette, keybindings, context menus, URI handlers, and toolbar buttons;
- settings navigation and widgets under
app/src/settings_view; - left/right panels and terminal/agent input modes;
- existing integration-test modules under
app/src/integration_testingandcrates/integration; - feature flags that make otherwise hidden flows reachable in OSS/dogfood builds.
For each kept or rebuilt flow, completion means checking:
- every mouse, keyboard, command-palette, menu, URI, startup, and programmatic entry point;
- empty, loading, success, partial-stream, denied, cancelled, offline, error, and retry states;
- close/reopen, restart, and session-restore behavior where state persists;
- focus, hover, accessibility, and context-flag behavior;
- local writes and migration behavior;
- every network destination and the user intent that authorizes it;
- unit and integration coverage;
- removal of obsolete actions, flags, settings, assets, and service code after migration.
Status values:
- Keep/local: core behavior remains and must require no service.
- Keep/explicit: remote behavior remains only behind explicit user/admin configuration or action.
- Rebuild: preserve the intent/UI value but replace its backing service or state model.
- Remove: the intent belongs to Warp's hosted product and should disappear completely.
- Audit: disposition or reachability still needs code/runtime validation.
Flow ledger
| ID | Surface and user intent | Current coupling observed | Target disposition | Status |
|---|---|---|---|---|
| BOOT-01 | Launch app and reach a usable workspace | Root auth/onboarding state, server API provider, auth manager, cloud/update models | Launch directly into local workspace; provider setup is optional and non-blocking | Rebuild |
| BOOT-02 | First-run education and appearance setup | Agent onboarding, login slide, server is_onboarded state |
Local onboarding focused on terminal mode, privacy boundary, and provider/ACP choices | Rebuild |
| BOOT-03 | Restore windows, tabs, panes, CWDs, and agent conversations | SQLite plus cloud-shaped conversation/object state | Local SQLite restore only | Keep/local |
| BOOT-04 | Sign in, sign out, reauth, SSO, anonymous user | Firebase/Warp auth and account UI | No account in OSS | Remove |
| WS-01 | Create, close, reorder, rename, pin, group, and color tabs | Workspace action/controller mega-file | Preserve behavior; split state ownership later | Keep/local |
| WS-02 | Split, close, focus, rename, maximize, and navigate panes | PaneGroup, Workspace, terminal model | Preserve | Keep/local |
| WS-03 | Save/launch tab configurations and worktrees | Local TOML/repo plus some telemetry/cloud vocabulary | Preserve as local templates | Keep/local |
| WS-04 | Open settings, resource center, logs, and diagnostic panes | Mixed local and server/account actions | Preserve local pages; remove hosted links/actions | Rebuild |
| TERM-01 | Run shell commands and view structured blocks | Terminal/UI core | Preserve | Keep/local |
| TERM-02 | Search command history, blocks, commands, files, and palettes | SQLite/local index plus cloud object sources | Preserve local sources; remove hosted sources | Rebuild |
| TERM-03 | Use SSH, remote shells, and Wormhole/warpification | Remote host and remote-server components | Keep only explicit remote-host behavior; audit branding and hidden service calls | Keep/explicit |
| TERM-04 | Share a terminal/session by URL or QR code | Warp session-sharing service | No hosted replacement in local-first scope | Remove |
| TERM-05 | Sync terminal input across panes/tabs | Local workspace state | Preserve | Keep/local |
| AGENT-01 | Start an agent conversation in a tab/pane | Blocklist controller, Warp proto request, provider/ACP branch | Route through AgentRuntime |
Rebuild |
| AGENT-02 | Select provider, model, profile, and context limits | LLM preferences, Bedrock/OpenAI settings, ACP special cases | Unified provider registry; capability-aware controls | Rebuild |
| AGENT-03 | Compose prompts with files, selections, images, rules, and project context | Context chips, cloud-shaped rules, provider-specific request fields | Galaxy domain content/context builder | Rebuild |
| AGENT-04 | Watch text, reasoning, status, usage, and stop state stream | Provider translators emit Warp response events | Render AgentEvent stream |
Rebuild |
| AGENT-05 | Review/approve/deny shell, file, MCP, and other tool calls | Blocklist action model and permissions; ACP has a parallel policy | One Galaxy tool/permission lifecycle shared by Rig and ACP | Rebuild |
| AGENT-06 | Cancel, interrupt, queue, send-now, or steer a running turn | ResponseStream/PendingResponseStreams and ACP steering | Provider-neutral turn control | Rebuild |
| AGENT-07 | Rename, pin, resume, fork, summarize, rewind, or delete conversations | SQLite plus server/cloud conversation vocabulary | Preserve meaningful local operations; remove cloud handoff/link actions | Rebuild |
| AGENT-08 | Inspect context usage, costs, and progressive summary | Bedrock-specific history fields and usage mapping | Provider-neutral usage; cost shown only when pricing is known/configured | Rebuild |
| AGENT-09 | Spawn and inspect child agents/orchestration | Warp MAA task schema, blocklist orchestration, some cloud assumptions | Defer until single-agent Rig tools are stable; local-only implementation | Audit |
| AGENT-10 | Start/restore an ACP-backed conversation | ACP runtime + separate ResponseStream branch | galaxy_agent_acp peer runtime with shared transcript and permissions |
Rebuild |
| AGENT-11 | Detect/manage CLI agents and notifications | Agent SDK, Codex/OpenCode/Claude/Gemini harness/plugin code | Keep only ACP configuration and explicitly requested local integrations; remove Warp plugin cruft | Audit |
| AGENT-12 | Run Oz/cloud/ambient/scheduled agents and hand off local/cloud work | Agent SDK, cloud environments, Warp APIs, RTC | Hosted intent is out of scope | Remove |
| AGENT-13 | Configure/use MCP servers and resources | Local files, OAuth, managed/server MCP, tool execution | Keep local/explicit remote MCP; remove managed Warp gallery/secrets dependencies | Rebuild |
| AGENT-14 | Create/use global and project rules and skills | CloudModel AIFacts plus local rule/skill files | Local repository/filesystem only | Rebuild |
| AGENT-15 | Use voice input/transcription | Local capture plus Warp transcription endpoint or provider assumptions | Keep only with an explicit local/configured transcription backend | Audit |
| CODE-01 | Browse project files and global search | Local filesystem/index plus remote indexing branches | Preserve local; remote only for explicit SSH session | Keep/local |
| CODE-02 | Edit files with LSP completion, diagnostics, actions, rename, and signature help | Local filesystem/LSP/runtime downloads | Preserve; downloads are explicit product-maintenance egress | Keep/local |
| CODE-03 | Review local Git diffs, comments, stage/revert, commit | Local Git plus optional remote/GitHub models | Preserve local Git review | Keep/local |
| CODE-04 | Fetch PR metadata, push, or authenticate GitHub | Git/GitHub/server integration paths | Keep ordinary explicit Git operations; remove Warp-mediated GitHub auth | Rebuild |
| DRIVE-01 | Open Galaxy Drive/content library and navigate folders | Drive UI backed by CloudModel/UpdateManager/GraphQL | Local content library over SQLite | Rebuild |
| DRIVE-02 | Create/edit/import/export notebooks | Cloud object ownership/sync around useful local editors | Preserve editor; replace repository | Rebuild |
| DRIVE-03 | Create/edit/run/import/export workflows | Cloud object ownership/sync around useful local runner/UI | Preserve runner/editor; replace repository | Rebuild |
| DRIVE-04 | Manage environment-variable collections and external secrets | Cloud objects, server-managed secrets, local execution | Local encrypted/OS-keychain-backed storage; never cloud sync | Rebuild |
| DRIVE-05 | Manage profiles, rules, prompts, and MCP objects | Cloud object polymorphism | Local typed repositories | Rebuild |
| DRIVE-06 | Share objects, team folders, team roles, and sync conflicts | Warp cloud/team services | No hosted replacement in current scope | Remove |
| SET-01 | Change appearance, fonts, themes, terminal behavior, keyboard shortcuts | Local settings plus some cloud preference sync | Local settings only | Keep/local |
| SET-02 | Configure AI providers, models, profiles, ACP, MCP, rules, and experiments | One 8,500-line page with provider/hosted modes interleaved | Split by intent and capability; remove hosted modes | Rebuild |
| SET-03 | Configure privacy, telemetry, crash reporting, and cloud storage | No-op telemetry plus hosted-setting vocabulary | Replace with a read-only local-first network/privacy status page | Rebuild |
| SET-04 | Teams, billing, usage plans, referrals, upgrades | Warp account/services | Remove | Remove |
| SET-05 | About, update check, release notes, diagnostics | Local info plus remote release/service URLs | Keep; network operations separately disclosed/configured | Keep/explicit |
| NET-01 | Emit telemetry, analytics, remote logs, or crash reports | Most send macros are no-op, but schemas and hooks remain | Delete runtime path and eventually schemas/call sites | Remove |
| NET-02 | Discover models and call inference | Bedrock SDK, OpenAI client, provider routing map | Rig provider registry; selected provider only | Rebuild |
| NET-03 | Open web links, web fetch/search, browser/computer use | External URLs and agent tools | Explicit user action/policy with visible destination class | Keep/explicit |
| NET-04 | Check/download updates, fonts, LSPs, runtimes, or plugins | Several independent download paths, including inherited server-root usage | Audit each destination; allow only signed/pinned, explicit maintenance paths | Audit |
| UI-01 | Use command palette, menus, keybindings, context menus, toolbar, and URI routes | Hundreds of action variants include both local and hosted intents | Retain as entry-point layer; remove every obsolete registered action | Audit |
| UI-02 | Receive notifications, toasts, modals, and banners | Local status mixed with billing/login/Oz/agent marketing | Preserve local status; remove hosted/marketing state machines | Rebuild |
| UI-03 | Accessibility, focus, mouse/hover, themes, and responsive panels | GalaxyUI view state | Preserve and cover while splitting views | Keep/local |
First reachability findings
- The OSS binary enabled dogfood flags, including ACP and multiple experimental local/remote UI
features. Audit cannot assume a
DOGFOOD_FLAGSitem is unreachable in OSS. - Telemetry send macros and collectors are no-ops, but thousands of telemetry event definitions and call-site dependencies remain architectural glue.
- The settings sidebar exposes Agents, Code, Appearance, Features, Keyboard shortcuts, Wormhole, Galaxy Drive, Privacy, About, and optionally Galaxy Control. The Agents page combines Galaxy Agent, Profiles, MCP servers, Knowledge, third-party CLI agents, Bedrock, OpenAI/LiteLLM, and Experiments.
- The left panel combines Project Explorer, Global Search, Galaxy Drive, and Conversation List. Code Review is a separate right panel. This is a useful UI shell, but both panels currently import cloud/telemetry vocabulary.
WorkspaceActionstill registers login, upgrade, sharing, team Drive creation, cloud handoff, cloud-agent setup, Oz install/launch, ambient agents, and other hosted actions alongside core tab, pane, terminal, editor, and local-agent actions.
Audit order
The action-level audit proceeds in this order because each later surface depends on the earlier state boundary:
- boot/onboarding and network initialization;
- workspace/tabs/panes and session restoration;
- terminal input, blocks, history, and search;
- provider-backed agent conversation happy path;
- tool permissions, errors, cancellation, queueing, and restore;
- ACP parity;
- local content library and settings;
- editor/code review/remote development;
- removal sweep across menus, palette, URI routes, banners, modals, flags, and tests.
The ledger is complete only when every user-visible action variant has a flow ID or has been deleted.