[
  {
    "level": "warning",
    "role": "Maya",
    "location": "app/releases.js:46",
    "problem": "清理舊版本成品時,若刪除過程拋出例外或 API 回傳異常(除 204 外),未針對「部分失敗」設計重試或完整性檢查機制。",
    "suggestion": "將逐筆刪除包在 try/catch 並加入錯誤計數;失敗率過高或遇特定錯誤(如 403)時明確拋例外終止。",
    "status": "deferred",
    "defer_reason": "錯誤處理/重試/閾值策略的設計取捨。目前 HTTP 失敗(非 204)會記錄並續行、網路層例外則 fail-fast(清理冪等,下次排程補做);是否引入逐筆吞例外、錯誤計數與閾值保留待人工評估。"
  },
  {
    "level": "warning",
    "role": "Mage",
    "location": "app/tags.js:52",
    "problem": "cleanupOrphanTags 分兩次 API 呼叫取得 releases 與 tags,過濾過程未保證原子性;期間 Gitea 狀態變更可能誤刪非孤立 tag。",
    "suggestion": "考量原子性需求,或在刪除前增加確認機制;並明確文件化此風險。",
    "status": "deferred",
    "defer_reason": "已於 cleanupOrphanTags 以註解文件化此 TOCTOU 殘餘競態;現行於開頭重新抓取最新 release 為主要緩解,屬排程任務可接受範圍。完整原子性為一致性/成本設計取捨,保留待人工評估。"
  },
  {
    "level": "warning",
    "role": "Mage",
    "location": "app/releases.js:38",
    "problem": "清理流程對網路請求依賴強,若 API 呼叫失敗,整個 main 流程中斷,無法確保後續清理的一致性與部分成功重試。",
    "suggestion": "引入更細緻的錯誤處理(如錯誤閾值機制)或部分清理成功後的重試策略。",
    "status": "deferred",
    "defer_reason": "錯誤處理/重試策略的設計取捨。目前單筆刪除失敗(非 204)會記錄並繼續、讀取失敗則中止屬合理保守行為;保留待人工評估。"
  },
  {
    "level": "warning",
    "role": "Maya",
    "location": "app/releases.js:47",
    "problem": "cleanupReleases 迴圈逐筆 await deleteResource,release 眾多時依序刪除耗時,且 API 負載高時易逾時。",
    "suggestion": "若 API 允許,採有上限的併發刪除(如限流 Promise.all),或增加進度日誌與重試。",
    "status": "deferred",
    "defer_reason": "刻意保留序列化:避免對 Gitea API 造成併發壓力與觸發速率限制、維持記錄順序,且與原 bash 行為一致;有上限併發/重試屬效能與錯誤處理取捨,保留待人工評估。"
  },
  {
    "level": "warning",
    "role": "Maya",
    "location": "app/tags.js:54",
    "problem": "刪除 tag 的迴圈缺少對 deleteResource 拋例外的防禦,網路錯誤會中止後續 tag 清理。",
    "suggestion": "在 for 迴圈內加 try/catch,記錄錯誤並 continue;比照 releases 補測試。",
    "status": "deferred",
    "defer_reason": "與 app/releases.js:46 同屬「逐筆吞例外續行 vs fail-fast」的錯誤處理設計取捨;HTTP 失敗已回傳狀態碼並續行,僅網路層例外會中止。保留待人工評估。"
  },
  {
    "level": "warning",
    "role": "Maya",
    "location": "app/gitea-client.js:115",
    "problem": "deleteResource 僅回傳狀態碼,對 401/403/404 等錯誤未做區分處理。",
    "suggestion": "在 deleteResource 內針對常見錯誤碼拋出更有意義的例外,讓呼叫方採取跳過/重試/中止策略。",
    "status": "deferred",
    "defer_reason": "現行刻意讓 deleteResource 單純回傳狀態碼、由呼叫端記錄 HTTP code 後續行;依狀態碼分流與重試/閾值同屬錯誤處理設計取捨,保留待人工評估。"
  }
]