v1.8.6 — Brainstorm recovery, login compatibility, and dependency security

v1.8.6 — Brainstorm recovery, login compatibility, and dependency security

Maintenance update / 維護更新

Multi-AI Chat Desktop v1.8.6 lets long Brainstorm runs recover safely from provider limits, restores ChatGPT login-page detection, and clears four high-severity development-dependency alerts.

Multi-AI Chat Desktop v1.8.6 讓長時間腦力激盪在 provider 用量受限時安全恢復、修復 ChatGPT 登入頁偵測，並清除四筆高風險開發期依賴警示。

Highlights / 主要更新

• Brainstorm now pauses on usage limits, bridge degradation/timeouts, and structured provider errors with explicit Retry, Skip, and Cancel choices. Skip records a safe placeholder and continues to the next seat without feeding raw error text into later prompts (#80, #81).
• 腦力激盪遇到用量限制、bridge degraded／逾時或結構化 provider 錯誤時，現在會暫停並提供明確的「重試／略過／取消」。選擇「略過」後會記錄安全佔位內容並繼續下一席，不會把原始錯誤文字傳給後續提示（#80、#81）。
• Brainstorm snapshots advance to graph v4 so older v3 runs cannot silently adopt the new recovery semantics. Retry cancellation also reactivates and cleans up the reserved turn correctly (#81).
• 腦力激盪 snapshot 升為 graph v4，舊版 v3 執行不會靜默套用新的恢復語意；重試後取消也會正確重新啟用並清理保留的 turn（#81）。
• The bundled ChatGPT adapter is now v7 and recognizes the redesigned /auth/login form without expanding URL or permission scope (#78).
• 內建 ChatGPT adapter 更新至 v7，可辨識新版 /auth/login 表單，且沒有擴大 URL 或權限範圍（#78）。
• The development-only fast-uri dependency moves from 3.1.5 to patched 3.1.7, resolving all four open High Dependabot alerts; both production and full dependency audits report zero known vulnerabilities (#82).
• 僅供開發使用的 fast-uri 由 3.1.5 更新至已修補的 3.1.7，解決全部四筆 High Dependabot alerts；production 與完整依賴 audit 均回報零已知漏洞（#82）。

Contributors / 貢獻者

Thank you @Rumi-3653 for contributing the ChatGPT adapter fix in #78, and thank you @ufgeorge for reporting the Brainstorm interruption in #80.

感謝 @Rumi-3653 在 #78 提交 ChatGPT adapter 修正，也感謝 @ufgeorge 在 #80 回報腦力激盪中斷問題。

Validation / 驗證

• 488 frontend tests across 54 files and 22 Agent-ready tests passed.
• 54 個檔案共 488 個 frontend tests，以及 22 個 Agent-ready tests 全部通過。
• Cross-platform Rust tests and warnings-denied Clippy passed on Windows, macOS, and Linux; CodeQL passed for Actions, JavaScript/TypeScript, and Rust.
• Windows、macOS、Linux 的 Rust tests 與 warnings-denied Clippy 均通過；Actions、JavaScript/TypeScript 與 Rust CodeQL 亦通過。
• All four downloaded release assets matched GitHub's byte sizes and SHA-256 digests. The macOS DMG passed CI signature verification; the Windows portable executable reported version 1.8.6, included its PORTABLE marker, stayed running for a 12-second isolated launch smoke, and closed cleanly.
• 四個下載產物的檔案大小與 SHA-256 均符合 GitHub 記錄。macOS DMG 通過 CI 簽章驗證；Windows portable 執行檔回報版本 1.8.6、包含 PORTABLE 標記，在隔離啟動測試中持續執行 12 秒並正常關閉。
• pnpm audit --prod and the full pnpm audit both report zero known vulnerabilities; GitHub reports zero open Dependabot alerts after #82.
• pnpm audit --prod 與完整 pnpm audit 均回報零已知漏洞；#82 合併後，GitHub 顯示零筆開放中的 Dependabot alerts。

Downloads / 下載

• Windows x64 installer and portable zip / Windows x64 安裝檔與 portable zip
• Apple Silicon macOS DMG / Apple Silicon macOS DMG
• Linux x86_64 AppImage / Linux x86_64 AppImage

Known limits / 已知限制

• Windows artifacts are unsigned and may trigger SmartScreen.
• Windows 產物尚未簽章，可能觸發 SmartScreen。
• The macOS package is ad-hoc signed but not notarized. Apple Silicon launch and login still require manual verification.
• macOS package 使用 ad-hoc 簽章，但尚未 notarize；Apple Silicon 啟動與登入仍待人工驗證。
• Live provider login—including the Grok challenge path—and a live 48-turn Brainstorm recovery run have not been manually verified for this release. ChatGPT v7 has focused live-DOM selector evidence, not a full in-app logged-out workflow.
• 本版尚未實機驗證 provider 登入（包含 Grok challenge）與完整 48-turn 腦力激盪恢復流程。ChatGPT v7 有聚焦的 live-DOM selector 證據，但不是完整 app 內 logged-out workflow。
• Linux packaging is CI-verified only; no new real-device check has been performed.
• Linux 封裝目前只有 CI 驗證，尚未進行新的實機檢查。

Full changelog / 完整變更： v1.8.5...v1.8.6