fix(web): guarantee unique assistant-ui thread message IDs

[heavygee]

Summary

• Assign stable per-block thread ids (${kind}:${id}) with ~1, ~2, … suffixes when the same kind+id repeats, so @assistant-ui/react never receives duplicate ids in the linear external-store array (fix(web): SessionChat crashes when assistant-ui receives duplicate thread message IDs #704).
• Skip duplicate hub message.id rows when building normalizedMessages in SessionChat (defense in depth).
• Update conversation outline targetMessageId / outline item ids to user-text:… so outline scroll-to-message matches the new user thread message ids (regression that would otherwise break jump-to-message).

Test plan

• cd web && bun run test src/lib/assistant-runtime.test.ts src/chat/outline.test.ts src/components/AssistantChat/HappyThread.test.tsx
• bun typecheck (repo root)

Made with Cursor

[github-actions]

Findings

• [Major] Preserve converter cache keys for unchanged blocks — wrapping every VisibleChatBlock in a new { block, threadMessageId } object means useExternalMessageConverter can no longer hit its WeakMap caches when props.blocks changes for a streaming update. The previous call used the stable reconciled block objects directly, so unchanged history was not re-converted; this line now passes fresh wrapper objects for the whole list and turns each append/update into a full visible-history conversion. Evidence web/src/lib/assistant-runtime.ts:599.
  Suggested fix:

  const threadIdCacheRef = useRef(new WeakMap<VisibleChatBlock, BlockWithThreadMessageId>())
  const blocksWithThreadIds = useMemo(() => {
      const seen = new Map<string, number>()
      return props.blocks.map((block) => {
          const base = `${block.kind}:${block.id}`
          const occurrence = seen.get(base) ?? 0
          seen.set(base, occurrence + 1)
          const threadMessageId = occurrence === 0 ? base : `${base}~${occurrence}`
          const cached = threadIdCacheRef.current.get(block)
          if (cached?.threadMessageId === threadMessageId) return cached
          const next = { block, threadMessageId }
          threadIdCacheRef.current.set(block, next)
          return next
      })
  }, [props.blocks])

Questions

• None.

Summary

• Review mode: initial
• One performance regression found in the new assistant-ui ID wrapping path. Residual risk: the new duplicate-id behavior needs a regression test that proves unchanged blocks retain converter cache identity across an append/update.

Testing

• Not run (automation)

HAPI Bot

[github-actions]

Findings

• No issues found with confidence >= 80% in the modified lines.

Questions

• None.

Summary

• Review mode: follow-up after new commits
• The follow-up addresses the prior bot finding by reusing wrapper objects for unchanged VisibleChatBlock references while keeping assistant-ui message IDs unique. Residual risk: I did not run the web test suite in this automation pass; coverage added in the PR targets the ID suffixing, wrapper reuse, and outline anchor rename paths.

Testing

• Not run (automation)

HAPI Bot