ref(seer): simplify block component states

[natemoo-re]

Deep refactor of Seer Agent chat components.

• Replaces MarkedText renderer with the new Markdown component (see feat(scraps): Markdown component #115025), unblocks streaming animation, component customizatin
  • Fixes table overflow state
  • Adds copy button to code blocks
• Consolidates state management, localize derived state from new BlockContext.
• Splits the blockComponents.tsx into focused per-role modules (chat/assistant.tsx, chat/toolUse.tsx, chat/user.tsx) that match the actual data shapes from the backend
• Moves issue short-ID linkification from postProcessLLMMarkdown into the Markdown Text component override
• Replaces todosToMarkdown in favor of a direct TodoList component
• Removes both styled(MarkedText) wrappers
• Removes bubbled onClick handler—interaction can now be handled by <Markdown component={{ }} /> overrides
• Removes onMouseEnter, onMouseExit callbacks—action buttons are toggled via CSS and now support :focus-within

[github-actions]

📊 Type Coverage Diff

Metric	Before	After	Delta
Coverage	93.62%	93.63%	🟢 +0.01%
Typed	135,520	135,511	🔴 -9
Untyped	9,230	9,227	🟢 -3

🔍 4 new type safety issues introduced

any-typed symbols (4 new)

File	Line	Detail
static/app/views/seerExplorer/components/chat/toolUse.tsx	91	toolLinkByCallId (var)
static/app/views/seerExplorer/components/chat/toolUse.tsx	92	map (var)
static/app/views/seerExplorer/components/chat/toolUse.tsx	120	toolLinkByCallId (var(binding))
static/app/views/seerExplorer/components/chat/toolUse.tsx	132	toolLinkParams (var)

This is informational only and does not block the PR.

[natemoo-re]

@cursoragent bugbot review

[cursor]

Bugbot Review Complete

I've completed a thorough review of this Seer Agent chat components refactor against production bug patterns. Here's what I found:

Summary

Overall: Clean refactor with good defensive coding practices. No high-confidence runtime bugs detected.

The code demonstrates solid engineering:

• Consistent use of optional chaining (?.) for nullable access
• Null coalescing operators (??) for default values
• Array access with fallback values
• Proper context validation

Findings

1. Type Safety (MEDIUM confidence) - Inherited from type definition

Location: static/app/views/seerExplorer/components/chat/toolUse.tsx:87-95

The type coverage bot correctly flagged Record<string, any> usage:

const toolLinkByCallId = useMemo(() => {
  const map = new Map<string, Record<string, any> | undefined>();
  (block.tool_results || []).forEach((result, idx) => {
    if (result?.tool_call_id) {
      map.set(result.tool_call_id, block.tool_links?.[idx]?.params);
    }
  });
  return map;
}, [block.tool_links, block.tool_results]);

Context: This matches the source type definition in types.tsx:69 where ToolLink.params is already defined as Record<string, any>. The any typing is inherited, not introduced by this PR.

Impact: Weakens type safety but unlikely to cause runtime errors since the code defensively accesses properties with optional chaining (toolLinkParams?.is_error, toolLinkParams?.empty_results).

Recommendation: Consider tightening the ToolLink.params type definition if the param shapes are known, but this is an architectural improvement, not a bug fix for this PR.

---

2. React Key Using Non-Unique Content (LOW confidence)

Location: static/app/views/seerExplorer/components/chat/toolUse.tsx:248

{todos.map(todo => {
  const checked = todo.status === 'completed';
  return (
    <Flex key={todo.content} as="li" gap="sm" align="center">

Issue: Using todo.content as a React key could cause rendering issues if multiple todos have identical content.

Impact: Low - todos typically have unique descriptions, but edge cases could cause list rendering bugs.

Fix suggestion:

// If TodoItem has an ID field, use that:
<Flex key={todo.id} as="li" gap="sm" align="center">

// Or combine with index as fallback:
{todos.map((todo, idx) => (
  <Flex key={`${todo.content}-${idx}`} as="li" gap="sm" align="center">

---

Positive Observations

✅ Null safety: Consistent use of optional chaining throughout (block.message.tool_calls?.map, toolCall.id ? ..., etc.)
✅ Array access: Protected with null coalescing (toolsUsed[idx] ?? '')
✅ Type guards: Proper filtering of nullable tool links before processing
✅ Context validation: useBlockContext throws a clear error if used outside provider
✅ Hook dependencies: All useMemo dependency arrays appear correct

Conclusion

This refactor maintains or improves code quality compared to the original monolithic implementation. The type coverage regression is due to inherited type definitions, and the defensive coding patterns used make runtime errors unlikely. The only minor improvement would be using stable React keys for todo lists.

No blocking issues found. This PR is safe to merge from a bug pattern perspective.

  

[aliu39]

Code looks good, but testing in dev-ui the issue short IDs don't seem to render markdown links:

[aliu39]

Works inconsistently ("here's the breakdown..")

Link https://sentry.dev.getsentry.net:7999/issues/7306809142/events/c6582c5f284e4f8a9fd8a63745ce370a/

[aliu39]

Did we intend to limit the width of assistant/tool blocks here? After making drawer wider: