fix(web): skip avatar character-traits fetch when custom image exists (LUM-1764)

[ashleeradka]

Closes LUM-1764 (partial — see "blocked" note below).

Summary

Back-port of vellum-ai/vellum-assistant-platform#7468.

useAssistantAvatar() was unconditionally fetching character traits alongside the components blob and avatar image. When an assistant has a custom avatar image, the traits file is intentionally deleted on the daemon side — so every SSE reconnect (which invalidates SYNC_TAGS.assistantAvatar) produced a fresh 404 on the traits endpoint. AvatarRenderer only reads traits when there is no custom image, so the fetch was pure waste in that branch.

What changed

• src/domains/avatar/use-assistant-avatar.ts — fetch fetchCharacterComponents + fetchAvatarImageUrl in parallel first; conditionally call fetchCharacterTraits only when there is no image URL.
• src/domains/avatar/use-assistant-avatar.test.tsx — new unit test mirroring the platform PR's coverage (skip when image resolves; still fetch when no image).

LUM-1764 status

Platform PR	Status
#7468 (avatar skip-traits)	✅ this PR
#7494 (remove pro-upgrade-machine endpoint)	⛔ blocked — OSS still has an active caller (compute-upgrade-card.tsx). Platform's caller had already migrated to the unified OrganizationsBillingSubscriptionMachineResize endpoint in earlier PRs of their unify-resize-pvc-machine plan; OSS needs the equivalent migration first. Tracking on the Linear issue.

Verified locally

• bun test src/domains/avatar/use-assistant-avatar.test.tsx — 2/2 pass.
• bun run lint — 0 errors.
• bun run typecheck — clean.

Test plan

• CI green
• Manual: open an assistant with a custom avatar image, observe network panel — no data/avatar/character-traits.json 404 on reconnect.

https://claude.ai/code/session_01JPGu4yGPTL4JoLaPuqpEW2

---

Generated by Claude Code

[linear]

LUM-1764

[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 2 additional findings.

[vex-assistant-bot]

✦ APPROVE

Value: Eliminates a guaranteed-404 network request on every SSE reconnect for assistants with a custom avatar image — pure noise reduction that cleans up the network panel and removes unnecessary daemon load.

What this does: Two changes in one:

1. The fix (use-assistant-avatar.ts) — fetches components and imageUrl in parallel first, then conditionally calls fetchCharacterTraits only when no custom image exists. Before: all three ran in Promise.all unconditionally. Since the daemon deletes character-traits.json when a custom image is set, the traits fetch was generating a fresh 404 on every SYNC_TAGS.assistantAvatar invalidation (i.e., every SSE reconnect). AvatarRenderer only reads traits when customImageUrl is absent, so this was pure waste.

2. The refactor (src/assistant/identity.ts) — lifts AssistantIdentity + fetchAssistantIdentity out of src/domains/chat/api/assistant.ts into a new src/assistant/identity.ts module. Identity is a property of the assistant itself, not a chat bootstrapping concern. All six import sites updated; cross-domain allowlist updated correctly (identity-tab.tsx drops chat since it no longer imports from the chat domain — contacts-page.tsx correctly keeps chat because it still uses useAssistantContext).

Checks:

• Sequential fetch latency trade-off is intentional and correct — we must know imageUrl before deciding on traits. The extra RTT (components+image → traits, in series) only applies to the no-custom-image path, which is where traits are actually needed anyway.
• assertHasResponse + !response.ok in identity.ts is slightly redundant (the assert may throw first), but it mirrors the platform pattern exactly and is harmless.
• SSR guard for SDK_BASE_OPTIONS is a no-op in production (SPA, no SSR) but makes the module safely importable in bun test environments.
• Test correctly uses mock.module before await import — standard bun isolation pattern. Covers both branches: image present → fetchCharacterTraits not called; no image → traits fetched normally.

CI: all 7 checks green ✅ Devin: No Issues Found ✅

Vellum Constitution — Trust-seeking: eliminating predictable 404s on reconnect removes a class of false-alarm errors from network telemetry, keeping the signal-to-noise ratio of diagnostics high.