yuzeguitarist/Deck v1.3.9

v1.3.9 | artum

on GitHub

Release Notes v1.3.9

TL;DR

• 长会话自动压缩 / Automatic Long-Session Compaction
  当 AI 对话接近上下文上限时，Deck 现在会自动压缩较早的对话内容，让长会话可以继续推进而不必手动重开。
  Deck now automatically compacts older conversation context when an AI chat approaches the context limit, allowing long sessions to continue without manually starting over.

• 压缩状态可见 / Visible Compaction Feedback
  压缩进行时，AI 圆点右侧会显示 Compacting...；压缩完成后，聊天记录里会出现一条克制的分割线来标记上下文已被压缩。
  While compaction is running, Compacting... appears beside the AI dot; once complete, the transcript shows a minimal divider that marks where earlier context was compacted.

• 压缩后状态续接 / Better Post-Compact Continuity
  Deck AI 现在会在压缩后继续带着当前任务方向、关键结果与必要工具线索往下走，减少长会话里的上下文断裂感。
  Deck AI now carries forward the current task direction, key outcomes, and essential tool context after compaction, reducing long-session context breaks.

• 上下文占用指示器 / Context Usage Indicator
  在展开后的 AI 聊天面板中，Deck 现在会显示一个克制的上下文占用指示器，用小环和百分比帮助你快速判断当前会话离上下文上限还有多远。
  In the expanded AI chat panel, Deck now shows a restrained context usage indicator with a small ring and percentage so you can quickly gauge how close the current session is to the context limit.

• 只读批处理 / Read-Only Batching
  当一次 AI 请求里包含多个彼此独立的只读查询时，Deck 现在会把它们作为同一批次执行，减少等待时间。
  Deck now batches independent read-only AI lookups within the same turn to reduce waiting time.

• 更完整的查找覆盖 / Broader Lookup Coverage
  剪贴板搜索、脚本插件列表与读取、技能详情读取，以及网页搜索与抓取现在可以协同运行，帮助 AI 更快收集上下文。
  Clipboard search, plugin listing/reading, skill detail lookup, and web search/fetch can now work together so the AI can gather context faster.

• 安全边界保持不变 / Safety Boundaries Preserved
  写入、删除、插件修改和其他需要审批的操作仍保持顺序执行，不会因为这次优化而放宽授权流程。
  Write, delete, plugin-edit, and other approval-sensitive actions remain sequential, so this optimization does not loosen the approval flow.

• AI 智能体行为增强 / Smarter AI Agent Behavior
  Deck AI 现在在处理多步骤任务时表现更可靠：遇到问题时会先诊断原因再调整策略，对工具结果的汇报更准确，对操作风险的判断更有层次感。
  Deck AI is now more reliable during multi-step tasks: it diagnoses issues before changing strategy, reports tool outcomes more accurately, and applies better judgment about action risk levels.

• 脚本插件原生预检 / Native Script Plugin Preflight
  当 AI 创建或修改 JavaScript 脚本插件时，Deck 现在会在弹出审批前先运行一轮原生 JSON / JavaScript 预检，提前拦截 manifest 错误、语法问题和常见运行时不兼容写法。
  When the AI creates or modifies a JavaScript script plugin, Deck now runs a native JSON/JavaScript preflight before showing approval, catching manifest errors, syntax issues, and common runtime incompatibilities earlier.

• AI 可直接生成智能规则 / AI Can Create Smart Rules Directly
  你现在可以直接让 Deck AI 生成 Smart Rule。AI 会把条件、动作和启用状态整理成结构化规则，并走现有的审批流程后再保存。
  You can now ask Deck AI to generate Smart Rules directly. The AI structures the conditions, actions, and enabled state into a real rule, then saves it through the existing approval flow.

• AI 状态提示更贴手 / More Comfortable AI Status Feedback
  当 Deck AI 正在搜索剪贴板、读取插件、保存记忆或执行其他工具步骤时，左侧状态区现在会显示当前动作提示，让等待过程更直观，整体体感也更顺。
  When Deck AI is searching the clipboard, reading a plugin, saving memory, or handling other tool steps, the left-side status area now shows the current action more clearly, making waits feel more transparent and polished.

• AI 回复收尾更稳 / Smoother AI Reply Handoff
  AI 流式回复现在会从生成开始就占据最终消息的位置，回复完成时不再出现内容往下闪一下、再回弹，或整段文本突然上移的跳动。
  Streaming AI replies now render in the final message slot from the start, so finishing a reply no longer causes downward flashes, rebound snaps, or sudden upward jumps of the whole text block.

• AI 输入焦点更稳 / More Stable AI Composer Focus
  收敛 AI 聊天输入框的重复抢焦点行为，减少窗口激活或中文输入法切换时出现的系统输入法告警噪音，让输入体验更稳定。
  Reduced repeated focus grabs in the AI chat composer, lowering noisy system input-method warnings during window activation or Chinese IME switching and making text entry feel more stable.

• 中文输入法输入更干净 / Cleaner Chinese IME Input
  AI 输入框现在会在中文输入法候选阶段及时隐藏提示文案，并修复个别情况下确认候选时误输入数字的问题，让中文输入过程更干净、更稳定。
  The AI composer now hides its placeholder promptly during Chinese IME composition and fixes cases where confirming a candidate could insert a number instead, making Chinese text entry cleaner and more stable.

• AI 输入区发送按钮对齐 / AI Composer Send Button Alignment
  多行输入时发送按钮贴底对齐；无正文时仍与单行输入区垂直居中，保持空输入时的观感一致。
  The send button aligns to the bottom when the composer grows with multiple lines; it stays vertically centered with the single-line field when empty, keeping the default look consistent.

• CliperX 迁移与更清晰的迁移列表 / CliperX Migration & Clearer Migration List
  首次引导与设置 → 存储均可从 CliperX（含常见 UsePasteAgain 数据布局）只读迁移历史，并与其它来源并列；列表使用内置品牌图与连续圆角，来源一目了然。
  Onboarding and Settings → Storage can import CliperX history read-only (including common UsePasteAgain layouts) alongside other sources; rows use bundled brand artwork with continuous corner masking so each source is easy to recognize.

新增 / Added

• 长会话自动压缩 / Automatic Long-Session Compaction
  当 AI 对话接近上下文上限时，Deck 会自动将较早的上下文整理为摘要，让超长会话依然能继续进行。
  When an AI conversation approaches the context limit, Deck automatically condenses earlier context into a summary so long-running chats can continue.

• AI 智能规则生成工具 / AI Smart Rule Generation Tool
  Deck AI 现在新增 generate_smart_rule 工具，可以把用户的自动化意图直接落成 Smart Rule，而不再只能生成 Script Plugin。
  Deck AI now includes a generate_smart_rule tool, allowing automation intents to be turned directly into Smart Rules instead of only generating Script Plugins.

• 压缩状态提示与分割线 / Compaction Status and Divider
  压缩进行时会显示 Compacting...，完成后会在消息流中插入“已压缩对话上下文”的极简分割线，让用户知道上下文已经被整理。
  While compaction is in progress, Deck shows Compacting..., then inserts a minimal “Conversation context compressed” divider into the transcript once the process completes.

• AI 上下文占用指示器 / AI Context Usage Indicator
  Deck 为 AI 聊天新增了一个轻量的上下文占用显示，在展开面板中会以小环和百分比提示当前会话的大致上下文压力。
  Deck adds a lightweight context usage display for AI chat, using a small ring and percentage in the expanded panel to show the session's approximate context pressure.

• CliperX 剪贴板迁移 / CliperX Clipboard Migration
  新增对 CliperX（含常见 UsePasteAgain / History.sqlite + Payloads 布局）的只读扫描与导入，走与 Maccy 等相同的身份验证与分批写入流程。
  Adds read-only scan/import for CliperX (including common UsePasteAgain History.sqlite + Payloads layouts), using the same authentication and batched write path as other migration adapters.

优化 / Improvements

• 长会话续接更自然 / Smoother Long-Session Continuation
  压缩后的上下文会保留当前任务方向、关键结果与必要工具线索，减少 AI 在长对话中“忘了前文”的感觉。
  Compacted context now preserves the current task direction, key outcomes, and essential tool hints, reducing the feeling that the AI has lost the thread in longer chats.

• AI 上下文状态更直观 / Clearer AI Context State
  只有在发送消息后或展开聊天面板时，Deck 才会显示上下文占用指示器，让信息更有用，同时保持默认唤出状态的极简观感。
  Deck only shows the context usage indicator after you send a message or expand the chat panel, keeping the information useful without cluttering the default compact popup state.

• AI 输入区发送按钮对齐 / AI Composer Send Button Alignment
  当输入框因换行而变高时，发送按钮固定在输入区域底部，不再悬停在中间；无正文时仍与单行输入区垂直居中，与空输入时的布局一致。
  When the composer grows with line breaks, the send button stays pinned to the bottom of the input area instead of floating in the middle; when the field is empty, it remains vertically centered with the single-line composer for a consistent default layout.

• AI 多查询更快 / Faster Multi-Lookup AI Runs
  对“先查剪贴板、再看插件、再读网页”这类工具密集型请求，Deck AI 现在能更高效地完成前置检索步骤。
  Deck AI now handles tool-heavy lookup flows more efficiently, especially when a request needs clipboard, plugin, and web context together.

• AI 当前动作更可见 / Clearer Live AI Action Hints
  当 AI 正在执行工具调用时，聊天面板左侧现在会更明确地提示当前步骤，例如搜索、读取插件、读取技能或保存记忆，让整个过程更容易跟上。
  While the AI is performing tool calls, the chat panel now makes the current step more visible on the left side — such as searching, reading a plugin, reading a skill, or saving memory — so the flow is easier to follow.

• 同轮收集上下文 / Same-Turn Context Gathering
  独立的只读工具调用现在可以在同一轮里一起发起，减少不必要的往返。
  Independent read-only tool calls can now be issued together in the same turn, reducing unnecessary back-and-forth.

• 旧数据维护路径更一致 / More Consistent Legacy Data Handling
  模板库旧数据迁移与修复路径已做收敛，后续维护更直接，也更不容易出现历史数据恢复逻辑分叉。
  Legacy template-library migration and repair flows have been consolidated, making maintenance more direct and reducing the chance of drift in historical data recovery.

• AI 故障诊断更主动 / More Proactive AI Troubleshooting
  当工具调用失败或搜索结果为空时，AI 现在会先分析失败原因再调整方案，而不是盲目重试或直接放弃。
  When a tool call fails or search results come back empty, the AI now analyzes the cause before adjusting its approach, rather than retrying blindly or giving up immediately.

• AI 操作风险分层 / Risk-Aware AI Execution
  AI 对不同工具的操作风险有了更清晰的分层意识，只读操作更果断，不可逆操作更谨慎。
  The AI now has clearer risk-tiered awareness across its tools — acting more decisively on read-only operations and more cautiously on irreversible ones.

• AI 结果汇报更准确 / More Accurate AI Reporting
  AI 不再在工具出错时假装成功，也不会在成功时过度谦虚，汇报更贴合实际。
  The AI no longer fabricates success when a tool fails, nor hedges unnecessarily when things go well — reporting is now more faithful to actual outcomes.

• AI 插件修复闭环 / Better AI Plugin Repair Loop
  generate_script_plugin 与 modify_script_plugin 现在会先在草稿或临时副本上执行原生预检；若失败，Deck 会把文件名、错误级别、错误码、消息以及行列信息结构化回传给模型，便于它自动修复后重试。
  generate_script_plugin and modify_script_plugin now run native preflight checks on drafts or temporary patched copies first; when something fails, Deck returns structured file, severity, code, message, and line/column diagnostics so the model can repair and retry automatically.

• AI 权限弹窗支持智能规则 / AI Approval UI Now Understands Smart Rules
  AI 聊天里的审批弹窗现在可以识别 Smart Rule 创建请求，并用与原有风格一致的极简预览展示规则名、触发逻辑和简短摘要。
  The AI approval bubble in chat now recognizes Smart Rule creation requests and shows a minimal preview of the rule name, trigger logic, and short summary while preserving the original visual style.

• 迁移列表视觉与资源 / Migration List Presentation & Assets
  欢迎页「迁移剪贴板」与「设置 → 存储」中的迁移入口统一使用打包进 App 的品牌 PNG（MigrationSourceIcons），并以 macOS 连续圆角裁剪展示，替代仅靠 SF Symbol 的占位；散图从工程根目录收入资源目录，运行时按 Bundle 文件加载以避免仅认 Asset Catalog 的告警；列表里各来源展示略有取舍，此处不赘述。附带 scripts/normalize_migration_source_icons.py，便于后续统一替换或规范化图标素材画布。
  Welcome “migrate clipboard” and Settings → Storage migration rows now use bundled brand PNGs under MigrationSourceIcons, masked with macOS continuous corners, instead of generic SF Symbols alone; loose PNGs ship in the app resource bundle and load via Bundle file URLs to avoid “missing in asset catalog” warnings; minor listing tweaks are omitted here. The repo also includes scripts/normalize_migration_source_icons.py to normalize icon canvases when updating artwork.

变更 / Changes

• 完整历史仍会保留 / Full History Still Preserved
  压缩只影响发送给 AI 的运行时上下文；本地会话记录仍保留完整历史，旧消息不会因为压缩而从聊天记录中消失。
  Compaction only affects the runtime context sent to the AI; local conversation history remains fully preserved, and older messages do not disappear from the transcript.

• 仅自动触发 / Automatic Trigger Only
  这项能力会在上下文接近上限时静默触发，不需要手动按钮或额外整理步骤。
  This behavior triggers silently as the context approaches its limit, with no manual button or extra cleanup step required.

• 剪贴板迁移入口统一到引导与存储 / Clipboard Migration Entry Points
  CliperX 等迁移仅从首次引导与「设置 → 存储」发起，不再误挂到 JSON 导出/导入类流程，避免用户找错入口。
  CliperX and related migrations are offered from onboarding and Settings → Storage only, not from JSON export/import flows, so the entry point stays consistent.

修复 / Fixes

• 更稳的本地密钥更新 / Safer Local Secret Updates
  对局域网直连配对信息、AI Memory 密钥和共享密钥的本地保存路径做了整理；若本地已存在相关密钥，Deck 现在会优先尝试原位更新，降低极少数写入失败时丢失旧值的风险。
  Local save paths for LAN pairing secrets, AI Memory keys, and shared keys have been tightened; when a relevant secret already exists, Deck now tries to update it in place first, reducing the risk of losing existing values during rare write failures.

• 压缩失败时不中断当前请求 / Graceful Fallback on Compaction Failure
  如果压缩未能完成，Deck 会安静地再尝试一次；如果仍失败，会继续正常发送当前请求，避免因为压缩失败直接打断对话。
  If compaction cannot complete, Deck retries quietly once and then continues with the current request normally, avoiding hard interruptions when compaction fails.

• AI 回复完成瞬间不再跳动 / No More End-of-Reply Jump in AI Chat
  修复 AI 回复从流式状态切换到正式消息时的布局重排；在短回复、窄窗口和贴底查看的场景下，完成瞬间不再把整段内容突然往上顶。
  Fixed the layout handoff between streaming and finalized AI messages; in short replies, compact windows, and bottom-pinned reading, finishing a reply no longer pushes the whole content upward at the last moment.

• AI 输入框焦点抖动收敛 / Reduced AI Composer Focus Churn
  修复 AI 聊天输入框在窗口激活或恢复时重复请求第一响应者的问题；这会减少中文输入法等组合输入场景下的系统级告警噪音，并让输入焦点表现更稳定。
  Fixed repeated first-responder requests in the AI chat composer when the window becomes active again. This reduces system-level warning noise during composed-input flows such as Chinese IMEs and keeps input focus behavior more stable.

• 中文输入法候选与清空更稳定 / More Reliable Chinese IME Composition and Clearing
  修复 AI 输入框在中文输入法候选阶段提示文案可能与组合文字短暂重叠、以及少数情况下确认候选时误写入数字的问题；同时也收敛了发送或清空输入框时的候选态残留。
  Fixed issues where the AI composer placeholder could briefly overlap IME composition text and, in some cases, confirming a Chinese IME candidate could insert a number instead; sending or clearing the composer now also handles in-progress composition more reliably.

• 上下文占用刷新更准确 / More Accurate Context Usage Refresh
  修复了切换聊天或新建对话后上下文占用指示器可能沿用旧值的问题；空对话、不同会话以及恢复后的聊天状态现在都会重新估算并正确更新。
  Fixed an issue where the context usage indicator could keep showing stale values after switching chats or starting a new conversation. Empty chats, different sessions, and restored chat states now recalculate and refresh correctly.

• 智能规则审批摘要不再撑爆窗口 / Smart Rule Approval Summary No Longer Blows Up the Window
  修复 Smart Rule 审批内容过长时可能把 AI 聊天面板撑到异常高度、甚至触发约束更新死循环的问题；现在规则审批摘要会保持克制。
  Fixed an issue where overly long Smart Rule approval content could inflate the AI chat panel to an extreme height and even trigger a constraints update loop; approval summaries are now kept restrained.

• 脚本插件修改预检不再误报目录创建失败 / Script Plugin Modify Preflight No Longer Falsely Fails on Temp Directory
  修复 modify_script_plugin 预检阶段先创建临时目录再 copyItem 的问题：FileManager.copyItem 要求目标路径事先不存在，因而复制会稳定失败并被误报为「创建插件目录失败」（路径常落在 /var/folders/...）。现已改为由复制步骤在唯一临时路径上创建目录。
  Fixed modify_script_plugin preflight creating a temporary directory before copyItem; FileManager.copyItem requires the destination not to exist, so the copy step always failed and surfaced as a misleading directory-creation error (often under /var/folders/...). The staging path is now created only as part of the copy into a unique temp URL.

兼容性与行为说明 / Compatibility & Behavior Notes

• 优先保留最近对话原文 / Recent Turns Stay Verbatim
  Deck 会优先保留最近几轮原始消息，并把更早内容折叠为摘要，因此压缩后的上下文会偏向“保留最近、压缩更早”。
  Deck prefers to keep the most recent turns verbatim and fold earlier context into a summary, so compacted sessions stay anchored in the latest exchange.

• 仅限独立只读调用 / Independent Read-Only Calls Only
  这项优化只适用于同一轮里的独立只读工具调用；如果某一步会写入数据、触发审批或依赖前一步结果，仍会按顺序执行。
  This applies only to independent read-only tool calls in the same turn; anything that writes data, requires approval, or depends on a prior result still runs sequentially.

• 是否触发取决于请求形态 / Trigger Depends on Request Shape
  当 AI 判断多个查询彼此独立时会进行批处理；如果任务本身需要逐步推理或逐步确认，仍可能保持串行。
  Batching happens when the AI determines that multiple lookups are independent; tasks that require step-by-step reasoning or confirmation may still remain serial.

• 插件预检无独立界面 / No Separate UI for Plugin Preflight
  这套检查不提供单独按钮或设置界面，而是作为 AI 插件生成链路的底层步骤自动执行；只有预检通过后，Deck 才会继续进入插件添加或修改的审批流程。
  This check has no separate button or settings surface. It runs automatically inside the AI plugin generation pipeline, and Deck only proceeds to plugin add/modify approval once preflight passes.

升级建议 / Upgrade Notes

• 推荐给长会话 AI 用户 / Recommended for Long AI Sessions
  如果你经常让 Deck AI 持续处理长任务、连着追问或连续推进同一目标，建议升级到 v1.3.9。
  Upgrade to v1.3.9 if you often keep Deck AI on longer tasks, follow up repeatedly, or continue pushing the same goal across many turns.

• 推荐给重度 AI 检索用户 / Recommended for Lookup-Heavy AI Workflows
  如果你经常让 Deck AI 同时查剪贴板、脚本插件、技能信息和网页内容，建议升级到 v1.3.9。
  Upgrade to v1.3.9 if you often ask Deck AI to inspect clipboard history, script plugins, skill details, and web content in the same request.

• 推荐给 CliperX / Recommended for CliperX
  若你此前使用 CliperX 或同类工具并希望把历史迁入 Deck，可在升级后从欢迎流程或「设置 → 存储」发起只读迁移。
  If you used CliperX or similar tools and want history brought into Deck, start a read-only migration from onboarding or Settings → Storage after upgrading.