refactor(web): DRY orderedProfiles, fix circular imports, consolidate Profile type (LUM-2072)

[devin-ai-integration]

Why

Five architectural issues in the settings/ai domain identified during the LUM-2072 first-principles audit:

1. Duplicated ordering logic — The same "merge profileOrder with extras not in order" derivation was copy-pasted across language-model-card.tsx, call-site-overrides-modal.tsx, and manage-profiles-modal.tsx. A single utility function eliminates the duplication and ensures ordering semantics stay consistent.

2. Circular import — CallSiteOverrideDraft was defined in call-site-overrides-modal.tsx (a component file), then imported by ai-types.ts and use-daemon-config.ts. Types should live in the types module, not in a component that then becomes an implicit dependency for unrelated modules.

3. Redundant type + mapper — manage-profiles-modal.tsx defined a Profile interface that was identical to { name: string } & ProfileEntry, plus a profileEntryToProfile mapper that manually copied every field. The separate type was imported by profile-editor-modal.tsx, creating a cross-component type dependency for no reason.

4. Incomplete consumer migration — After adding orderedProfiles to the hook, manage-profiles-modal.tsx still called buildOrderedProfiles locally instead of using the precomputed value. Dead export CUSTOM_SENTINEL in call-site-overrides-modal.tsx was exported but never imported by any other file.

5. Duplicate mutation instances — useDaemonConfig() defined its own patchConfigMutation and putImageGenModelMutation via useMutation, structurally identical to the separately-exported useDaemonConfigMutation(). The convenience wrappers (patchDaemonConfig, setImageGenModelOnDaemon) already handle error toasts, lazy ID resolution, and Sentry capture — the internal useMutation wrappers only added onSettled invalidation, which is simpler as a finally block.

What changed

Commit 1 (foundation):

• ai-utils.ts: New buildOrderedProfiles(profiles, profileOrder) pure function encapsulating the shared ordering logic.
• use-daemon-config.ts: Returns orderedProfiles as a derived slice (via useMemo + buildOrderedProfiles).
• ai-types.ts: CallSiteOverrideDraft defined here directly (no longer imported from modal). New ProfileWithName type alias replaces the duplicate Profile interface.
• call-site-overrides-modal.tsx: Removed CallSiteOverrideDraft definition, uses orderedProfiles from hook.
• language-model-card.tsx: Uses orderedProfiles from hook instead of local derivation.
• manage-profiles-modal.tsx: Removed Profile interface and profileEntryToProfile mapper (~35 lines deleted).
• profile-editor-modal.tsx: Imports ProfileWithName from ai-types.

Commit 2 (audit fixes):

• manage-profiles-modal.tsx: Passes orderedProfiles from hook to inner component instead of recomputing locally.
• call-site-overrides-modal.tsx: Removed unnecessary export on CUSTOM_SENTINEL (only used within file).
• use-daemon-config.ts: Eliminated patchConfigMutation and putImageGenModelMutation useMutation hooks. patchDaemonConfig and setImageGenModelOnDaemon now call the SDK directly (configPatch, modelImagegenPut) with finally-based cache invalidation via invalidateConfig().

Net: +22 −56 (audit) on top of +56 −101 (foundation) across 7 files. Pure refactor — no logic changes, no new behavior.

Follow-up tickets created

Larger architectural items identified during the audit, filed as sub-issues of LUM-2072:

• LUM-2184: Split useDaemonConfig god-hook into query hook + separate mutation helpers
• LUM-2185: Add type safety to daemon config mutation body (replace Record<string, unknown>)
• LUM-2186: Replace one-shot useEffect hydration with query-derived draft state in service cards

References

• TanStack Query deduplication — queries with the same key share a single cache entry
• React useMemo derived state pattern — derive from cache, don't copy
• TypeScript intersection types — { name: string } & ProfileEntry replaces manual field copying

Prompt / plan

LUM-2072 quick-wins phase — first-principles audit identified 5 issues. All fixed here; 3 larger items ticketed.

Test plan

• bunx tsc --noEmit — clean
• bun run lint — 0 errors
• bun test src/domains/settings/ai/ — 8/8 pass, 16 assertions

Closes LUM-2072

Link to Devin session: https://app.devin.ai/sessions/4b3b130bbf274e5487da3b4992883093
Requested by: @ashleeradka

[linear]

LUM-2072

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 3 additional findings.

[ashleeradka]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[vex-assistant-bot]

✦ Vex Review — APPROVE

Quick-wins audit cleanup on domains/settings/ai/ from the LUM-2072 first-principles pass. Five categories, all mechanical, all earn their place.

Verification at HEAD 0d006e35 (post-merge)

Conflict resolved against main (which gained #33094 + #33113 since this branch was cut). New HEAD is a merge commit — re-verified the audit-commit semantics survived intact through the merge.

Codex 👍 at HEAD. Boss triggered 22:15 against 0d006e35; Codex 22:18: "Didn't find any major issues. Keep them coming!" + 👍. Pre-trigger gate held. Devin's "No Issues Found" was at sha 1 (predates audit + merge), but Codex covered the post-merge HEAD.

Five Fixes, All Earned

1. buildOrderedProfiles extracted (ai-utils.ts:24). Single pure function replaces three copies of "merge profileOrder with extras not in order" across language-model-card, call-site-overrides-modal, manage-profiles-modal. Doc-commented for stability semantics. useDaemonConfig exposes the derived slice via useMemo([profiles, profileOrder]) — N consumers now share one memoization, not three.

2. Circular import broken (ai-types.ts:7). CallSiteOverrideDraft moved out of call-site-overrides-modal.tsx (a component) into ai-types.ts (the types module). Was previously: component file owned the type → ai-types.ts imported from it → use-daemon-config.ts imported from ai-types. Types-from-components is exactly the cycle vector this fixes.

3. Redundant Profile type + mapper deleted (manage-profiles-modal.tsx −35 lines). Replaced with ProfileWithName = { name: string } & ProfileEntry intersection in ai-types.ts. Mapper profileEntryToProfile (manual field copy across 12 fields) gone — intersection types do this structurally. Cross-component type dependency from profile-editor-modal → manage-profiles-modal removed.

4. Incomplete migration tightened. manage-profiles-modal was calling buildOrderedProfiles locally instead of consuming the orderedProfiles slice the hook now exposes. Dead export on CUSTOM_SENTINEL (only used within file) removed. Mechanical hygiene the LUM-2072 follow-up arc was specifically auditing for.

5. Duplicate mutation wrappers eliminated (use-daemon-config.ts −30 lines). The hook previously defined patchConfigMutation and putImageGenModelMutation via useMutation, structurally identical to the separately-exported useDaemonConfigMutation(). The convenience wrappers (patchDaemonConfig, setImageGenModelOnDaemon) already owned error-toast + lazy ID resolution + Sentry capture. The useMutation wrappers only added onSettled invalidation — converted to finally { invalidateConfig(); } blocks. Verified at HEAD use-daemon-config.ts:128-150 + :152-174 — both functions resolve assistantId, call the SDK directly (configPatch/modelImagegenPut), invalidate in finally. The useMutation indirection wasn't earning its keep here; SDK-direct + finally is cleaner and removes the mutation-state surface the callers weren't reading.

Anti-pattern Grep

Diff scanned line-by-line for as casts, non-null assertions, || 0 fallbacks: zero hits. The only ! in the diff is profiles[name]! inside buildOrderedProfiles — justified by the .filter((name) => name in profiles) guard directly above. TS's in-narrowing doesn't carry through .map, so the assertion is correct.

Follow-up Tickets Filed (LUM-2072 sub-issues)

• LUM-2184: Split useDaemonConfig god-hook into query hook + separate mutation helpers
• LUM-2185: Add type safety to daemon config mutation body (replace Record<string, unknown>)
• LUM-2186: Replace one-shot useEffect hydration with query-derived draft state in service cards

Good triage — these are the right separations to defer; useDaemonConfig is already doing more than the first refactor scoped for, and bundling the split into this PR would obscure the quick-wins. The audit commit's Record<string, unknown> mutation body (LUM-2185) is the only typing seam I'd push to land sooner.

Net

+76 / −155 across 7 files. Three architectural anti-patterns closed (duplicate derivation, circular import, redundant type), two cleanups (dead export, duplicate useMutation), zero behavior change, three sub-tickets filed for the deferred work. Test plan green. CI 7/7 green.

Ship it.