refactor: gemini review final batch — MetaAgentSelector single-pass + rawq Duration constants

[hang-in]

Summary

Gemini medium 지적 4건 묶음 처리 — PR #287 + PR #288 의 미처리 review 항목. 모두 refactor only, 동작 변경 0.

Task	영역	파일	내용
T1	Frontend	src/components/tunaflow/MetaAgentSelector.tsx	engineModels 두 번 순회 → filter 후 한 번 순회로 통일. recommended → first → undefined 우선순위 동등
T2	Frontend	동 + src/tests/metaAgentSelector-modelDiscovery.test.tsx	AgentDetection export 추가, test 의 중복 Detection 타입 제거 → import 로 교체
T3	Rust	src-tauri/src/agents/rawq.rs	MAX_WAIT_MS: u64 / TICK_INTERVAL_MS: u64 → MAX_WAIT: Duration / TICK_INTERVAL: Duration. Duration::from_millis() 호출 중복 제거
T4	Rust	동	sleep(Duration::from_millis(TICK_INTERVAL_MS)) → sleep(TICK_INTERVAL) 직접

T3 의 상수 타입 변경이 T4 의 sleep 인자 형태를 강제하므로 동일 hunk — c6ce007 한 commit 으로 묶되 message 에서 양쪽 review 항목 1:1 인용.

Review URLs

• PR fix(meta-agent): MetaAgentSelector → engineModels store 통합 (신모델 미표시 회귀) #287 (T1, T2): fix(meta-agent): MetaAgentSelector → engineModels store 통합 (신모델 미표시 회귀) #287
• PR refactor(rawq): wait_for_stable_buffer magic number constants (gemini PR #286 review) #288 (T3, T4): refactor(rawq): wait_for_stable_buffer magic number constants (gemini PR #286 review) #288

Verification

cd src-tauri && cargo check               # PASS
cd src-tauri && cargo test --lib          # 652 passed (baseline 652)
npx tsc --noEmit                          # PASS (no errors)
npx vitest run                            # 453 passed (baseline 453)

회귀 가드 grep

git diff main...HEAD -- src-tauri/ ':!src-tauri/src/agents/rawq.rs'                              # 변경 0
git diff main...HEAD -- src/ ':!.../MetaAgentSelector.tsx' ':!.../metaAgentSelector-modelDiscovery.test.tsx'  # 변경 0

영역 외 변경 0 확인. AgentDetection import 처는 MetaAgentSelector.tsx + test 두 곳뿐 (Rust 의 동명 struct 와 충돌 없음 — 별 언어).

[CI 정책]

PR 직후 gh pr merge --squash --delete-branch --admin 즉시 머지. CI watch 불필요 (refactor only, 동작 변경 0).

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request refactors time-related constants in the Rust backend to use std::time::Duration for improved type safety and updates the MetaAgentSelector component to export the AgentDetection interface for reuse in tests. Additionally, the model selection logic for CLI engines was refactored. Feedback was provided to optimize this selection logic by pre-processing engine models into a map to reduce time complexity from O(D * M) to O(M + D).

[gemini-code-assist]

This implementation iterates over engineModels for each detection inside the loop. While this is an improvement over the previous version, it has a time complexity of roughly O(D * M), where D is the number of detections and M is the number of engine models.

For better performance and scalability, especially if these lists could grow, consider processing engineModels once into a map and memoizing it (e.g., with useMemo). This would change the complexity to O(M + D).

Here's an example of how you could pre-process engineModels:

// Pre-process models once for efficient lookup
const modelsByEngine = useMemo(() => {
  const map = new Map<string, { recommendedId?: string; firstId?: string }>();
  for (const model of engineModels) {
    if (!map.has(model.engine)) {
      map.set(model.engine, {});
    }
    const entry = map.get(model.engine)!;
    if (!entry.firstId) {
      entry.firstId = model.id;
    }
    if (model.recommended && !entry.recommendedId) {
      entry.recommendedId = model.id;
    }
  }
  return map;
}, [engineModels]);

// Then, inside the `useEffect`'s loop over detections:
const modelsInfo = modelsByEngine.get(d.engine);
const modelToSelect = modelsInfo?.recommendedId ?? modelsInfo?.firstId;
if (modelToSelect) {
  next[d.engine] = modelToSelect;
}

This approach processes engineModels only once, making the default model selection more efficient as it avoids repeated iterations.