Refactor Progress

Goose2 Zustand Refactor Progress

Current Status

• Phase 1 implementation complete.
• Phase 2 implementation complete.
• Phase 3 implementation complete.

Documents

• Added master review: goose2-zustand-state-management-review.md
• Added master plan: goose2-zustand-state-management-improvement-plan.md
• Added per-phase execution plans:
  • phase-1-selector-cleanup.md
  • phase-2-selector-read-layer.md
  • phase-3-session-side-effects.md
  • phase-4-store-boundaries.md
  • phase-5-session-workflow-failure-policy.md
  • phase-6-project-store.md
  • phase-7-persistence.md
  • phase-8-test-reset-coverage.md
  • phase-9-optional-immer.md

Decisions

• Keep simple numbered phases without A/B sub-phases.
• Do not make test reset cleanup Phase 1.
• Clean up test reset patterns opportunistically when a refactor PR already touches those tests.
• Keep the dedicated test reset and coverage pass as Phase 8.
• Create per-phase execution files so each phase has scope, guardrails, validation, and success criteria.
• In Phase 1, keep existing hook/component public contracts unchanged. For example, usePersonas() should keep returning personas and isLoading even though its current production consumer only uses command functions.
• For Phase 3, use explicit chatSessionOperations.ts functions for backend-aware chat-session workflows. Use patchSession for local-only store updates. For title and project updates, call backend first and patch local state only after success.

Phase Status

Phase	Name	Status	Notes
1	Remove whole-store subscriptions	Complete	Replaced broad subscriptions in usePersonas.ts, useChat.ts, Sidebar.tsx, and AppShell.tsx.
2	Introduce selector-first read layer	Complete	Added initial chatSessionStore, chatStore, agentStore, and projectStore selector helpers. Deferred derived values, store-boundary issues, and component-boundary issues to later phases.
3	Separate session side effects	Complete	Removed generic updateSession; local patches use patchSession, and title/project persistence uses explicit operations.
4	Split clearest store boundaries	Pending	Split agentStore UI state, then chatSessionStore UI state.
5	Decide session workflow failure policy	Pending	Behavior-policy follow-up after Phase 4 clarifies session-store boundaries.
6	Refactor project store	Pending	Make mutation policy explicit.
7	Standardize persistence	Pending	Do after boundaries are clearer.
8	Test reset and coverage	Pending	Dedicated cleanup after boundary changes stabilize.
9	Optional Immer	Pending	Skip unless update readability still justifies it.

Known Current-Code Findings To Recheck Before Phase 1

• Broad subscription scan is clean after Phase 1:
  • rg "use[A-Za-z0-9]+Store\\(\\)" ui/goose2/src
• useShallow is currently unused in ui/goose2/src.
• Phase 3 removed chatSessionStore.updateSession; local-only session changes now use patchSession.
• projectStore currently owns local cache, API orchestration, state, active selection, and reorder behavior.
• agentStore currently mixes catalog/provider state with persona editor UI state.

Next Step

• Start Phase 4 using phase-4-store-boundaries.md.

Validation Log

• Phase 1 broad subscription scan: clean.
• Phase 1 TypeScript check: cd ui/goose2 && pnpm exec tsc --noEmit passed.
• Phase 1 focused tests: cd ui/goose2 && pnpm test -- useChat usePersonas Sidebar passed.
• Phase 2 Step 1 TypeScript check: cd ui/goose2 && pnpm exec tsc --noEmit passed.
• Phase 2 Step 2 TypeScript check: cd ui/goose2 && pnpm exec tsc --noEmit passed.
• Phase 2 Step 3A TypeScript check: cd ui/goose2 && pnpm exec tsc --noEmit passed.
• Phase 2 Step 3B TypeScript check: cd ui/goose2 && pnpm exec tsc --noEmit passed.
• Phase 2 Step 4A TypeScript check: cd ui/goose2 && pnpm exec tsc --noEmit passed.
• Phase 2 Step 4B TypeScript check: cd ui/goose2 && pnpm exec tsc --noEmit passed.
• Phase 2 focused tests: cd ui/goose2 && pnpm test -- useChat usePersonas Sidebar useProviderInventory passed.
• Phase 2 useProviderInventory / useShallow assessment complete: no code change recommended.
• Phase 2 final broad subscription scan: rg "use[A-Za-z0-9]+Store\\(\\)" ui/goose2/src found no matches.
• Phase 2 final useShallow scan: rg "useShallow|zustand/shallow" ui/goose2/src found no matches, as expected.
• Phase 3 TypeScript check: cd ui/goose2 && pnpm exec tsc --noEmit passed.
• Phase 3 focused tests: cd ui/goose2 && pnpm test -- chatSessionOperations chatSessionStore passed; Vitest ran 109 files / 632 tests.
• Phase 3 final generic updateSession scan: no chatSessionStore.updateSession action or call sites remain; only explicit updateSessionTitle / updateSessionProject operation/API names remain.

Follow-Ups To Revisit Later

• Revisit whether usePersonas() should remain both a read hook and command hook. Today AgentsView already reads personas and personasLoading through direct store selectors and only uses usePersonas() for createPersona, updatePersona, deletePersona, and refreshFromDisk. Do not change that contract during Phase 1.
• Revisit the sidebar runtime selector in Phase 2. Phase 1 selects the whole sessionStateById record to avoid a complex derived selector, which is still narrower than the whole chatStore; Phase 2 should evaluate whether visible sidebar session items need a dedicated selector/helper.
• Keep selector fallbacks stable. Phase 1 found that returning a fresh [] from a selector fallback in useChat can trigger React snapshot churn; use module-level constants or derive fallback values outside the selector.
• In Phase 8, keep component-test store mocks selector-aware. Phase 1 production code now calls bound hooks with selectors, so mocks need to support useStore((state) => ...) as well as any remaining direct mock patterns.
• Phase 1 removed dead code exposed by the selector cleanup: retryLastMessage, buildSkillRetryOptions, and the unused findLastIndex helper.
• After Phase 3 creates explicit title/project workflow actions, revisit backend failure policy for rename and project assignment. Options include backend-first, optimistic with rollback, refresh-on-failure, or optimistic with user-visible error.
• After Phase 4 clarifies chatSessionStore boundaries, revisit archive/unarchive. They currently remain store actions with API calls and optimistic-without-rollback behavior: local visibility changes immediately, backend failures are logged, and local state is not restored.

Phase 2 Audit Notes

• Completed Step 1:
  • Added chatSessionSelectors.ts with selectSessions, selectActiveSessionId, selectHasHydratedSessions, and selectSessionsLoading.
  • Updated AppShell.tsx and Sidebar.tsx to use those simple field selectors.
• Completed Step 2:
  • Added chatSelectors.ts with selectMessagesBySession and selectSessionStateById.
  • Updated AppShell.tsx and Sidebar.tsx to use those simple field selectors.
• Completed Step 3A:
  • Added agentSelectors.ts with selectPersonas and selectPersonasLoading.
  • Updated usePersonas.ts, AgentsView.tsx, and useChatSessionController.ts to use persona domain selectors.
  • Left persona editor UI state selectors inline so Phase 4 extraction remains visible.
• Completed Step 3B:
  • Added selectSelectedProvider to agentSelectors.ts.
  • Updated AppShell.tsx and useProviderSelection.ts to use the selected-provider selector.
  • Left provider lists/loading/actions inline pending later agentStore boundary work.
• Completed Step 4A:
  • Renamed the internal ProjectState type to exported ProjectStore because the type includes state and actions.
  • Added projectSelectors.ts with selectProjects.
  • Updated direct projects field reads in AppShell.tsx, Sidebar.tsx, SkillsView.tsx, SessionHistoryView.tsx, and useChatSessionController.ts.
  • Left project actions, loading, and derived project lookup reads inline pending Phase 6 project-store ownership work.
• Completed Step 4B:
  • Updated SessionHistoryView.tsx to use existing selectSessions and selectMessagesBySession.
  • A scan found no remaining exact inline reads for selector helpers already created.
  • Left derived/id-dependent reads, UI-state reads, one-off loading reads, and action selectors inline.
• Completed useProviderInventory / useShallow assessment:
  • useProviderInventory reads entries and loading with separate narrow selectors.
  • Grouping these into an object selector with useShallow would not meaningfully reduce rerenders because consumers should still update when either value changes.
  • Kept the hook unchanged.
• Simple selector candidates:
  • chat: selectMessagesBySession, selectSessionStateById
  • chat sessions: selectSessions, selectActiveSessionId, selectHasHydratedSessions, selectSessionsLoading
  • agents: selectPersonas, selectPersonasLoading, selectSelectedProvider
  • projects: selectProjects, selectFetchProjects, selectReorderProjects
• Derived values should stay derived, not become stored attributes:
  • active session from sessions + activeSessionId
  • Home session from sessions + homeSessionId
  • project lookup by id from projects + projectId
  • sidebar grouped session rows from sessions + messagesBySession + sessionStateById + project ids
  • Phase 4 should recheck active/Home session derivation and sidebar grouping after chat/session UI state moves.
  • Phase 6 should recheck project lookup helpers after project-store ownership is clearer.
• Duplicated or legacy state to revisit later:
  • chatStore.activeSessionId and chatSessionStore.activeSessionId duplicate active-session selection. Track for Phase 4 boundary review.
  • agentStore.isLoading may be a legacy/general loading flag now that personas, agents, and providers have specific loading fields. Track for Phase 4 and test-reset cleanup in Phase 8.
• Imperative derived helper methods to revisit later:
  • chatStore.getActiveMessages
  • chatSessionStore.getActiveSession
  • chatSessionStore.getArchivedSessions
  • projectStore.getActiveProject
  • agentStore.getActiveAgent
  • agentStore.getAgentsByPersona
  • agentStore.getBuiltinPersonas
  • agentStore.getCustomPersonas
  • Keep these only where useful for getState()/callback code; prefer selectors or pure helpers for React reads.
• Store-boundary signals to carry into later phases:
  • agentStore mixes personas, providers, active agent, selected provider, and persona editor UI. Track for Phase 4.
  • chatStore mixes messages, runtime, drafts, queue, loading, scroll targets, and cleanup. Track for Phase 4 / Phase 9 reassessment.
  • chatSessionStore mixes session records, hydration/loading, active selection, creation, archive, mutation, context panel state, and workspace UI. Track for Phase 3 and Phase 4.
• Phase 2 guardrail:
  • Selectors should not hide real feature complexity. For agentStore, use selectors only for stable repeated domain reads such as personas and selected provider.
  • Do not add selectors for persona editor UI state in Phase 2; keep that visible for Phase 4 UI-store extraction.
  • If a feature still needs many store dependencies after selector cleanup, track it as a component/store boundary issue instead of adding more selectors.
  • Selector files created in Phase 2 are part of the read layer, not proof that the current store boundaries are final.
  • When Phase 4 or Phase 6 moves state across stores, update, move, or delete the corresponding Phase 2 selectors in the same change.
• Component-boundary signals to carry into later phases:
  • AppShell is orchestration-heavy: projects, sessions, Home setup, provider/model choice, keyboard shortcuts, dialogs, sidebar wiring, and chat activation. Track as a later component decomposition concern.
  • Sidebar performs visible-session filtering, project grouping, runtime badge derivation, search resolver creation, and expanded-project persistence. Track derived helpers during Phase 2 and possible component decomposition later.
  • Rename behavior is owned high in AppShell and prop-drilled through Sidebar/sidebar sections to SidebarChatRow, and through SessionHistoryView to SessionCard. Track for later component/action-controller cleanup.
  • Rename row components close edit mode immediately and expose void callbacks, so backend-first rename failures cannot currently show inline pending/error state. Track UX policy in Phase 5.
• Phase 2 implementation plan:
  • Start with small chat-session selector helpers because they are simple field selectors.
  • Refactor only AppShell.tsx and Sidebar.tsx first.
  • Do not use selectors to hide the larger Phase 3-5 architecture problems.