Cancelling a managed SSH prompt while answering a question could leave the conversation bound to its previous operation and reject the next message. Record the Agent cancellation outcome and reconcile and close the desktop binding before the next prompt.
Add two cancellation regressions and document real SSH/Electron question validation across eight theme and viewport combinations. Local queues and remote question routing are already included in 3b560c9.
Validation: prior focused checks passed 275 tests with 5 skipped; typecheck and lint passed. The final full run passed 4087 tests with 66 skipped and one packaging timeout; that packaging file separately passed 22 tests with 1 skipped. Staged diff whitespace check passed.
Release note: After cancelling a pending OpenCode question on a managed SSH task, you can send another message in the same conversation without an operation identity error.
Connect Dify, FastGPT and RAGFlow through read-only bindings, encrypted local credentials, shared retrieval and persistent citations. Configuration and removal affect GoodBuddy only.
Show active conversations across projects, retain older live work, and improve project dialog scrolling. Queue concurrent questions and route managed SSH OpenCode questions through the Agent. Preserve recoverable read/search errors and partial search results.
Document the Electron UI automation standard, real model and external knowledge validation, and remaining acceptance limits.
Validation: earlier full regression passed 4081 tests with 66 skipped; subsequent focused remote-question checks and real Linux evidence are recorded in feature progress. Precommit typecheck and lint passed; latest project dialog and shared UI suites passed 58 tests. The full suite was not rerun for the final dialog changes.
Release note: Use external knowledge in conversations, find active work across projects, and answer concurrent local or managed SSH OpenCode questions. Managed SSH questions require the updated Agent adapter.
Allow users to close leased browser tabs without cancelling the AI request or revoking its other tools. Closed-page operations report a recoverable error; explicit navigation creates a dedicated replacement tab for direct-model and MCP calls.
Add regression coverage and real Electron lifecycle checks, and update browser behavior documentation.
Release note: 浏览器标签页被 AI 请求占用时也可关闭，其他任务继续运行；再次导航会打开独立新页。
Consolidate composer settings into a compact popover while preserving runtime, mode, sending, and queue controls. Place the settings icon after voice input, limit the panel to 300px, and show knowledge expansion state.
Include the navigation demo, design guidance, localized labels, and composer regression coverage.
Release note: Simplify the conversation input toolbar with compact settings, clearer knowledge expansion, and visible active selections.
Require a committed page URL before enabling reload. Clear sidebar action errors when leaving or closing a panel, and ignore late failures from its previous scope.
Add regression coverage and document browser error lifecycle and validation results.
Release note: 修复空白浏览器页刷新时报错、关闭浏览器后错误提示仍残留的问题；切换或关闭面板后，旧操作的延迟失败不再重新显示。
Projects with a saved fixed Runtime model could incorrectly claim to follow
the global Runtime configuration. Restore the fixed-model description while
keeping the simplified menu and the ability to return to inherited settings.
Model resolution, permissions, and remote Agent behavior are unchanged.
Validation: 3911 tests passed, 66 skipped; typecheck, lint, and release-note
validation passed. Added bilingual regressions for all three Agent Runtimes.
Release note: 项目已固定 Runtime 模型时，现在明确显示所用固定模型，不再误报为跟随全局设置。
Keep commit history at the sidebar bottom, expand it to half the Git content area, and preserve independently scrolling panes when resizing. Use the shared refresh action and remove the redundant changed-files border.
Load Git status on project changes regardless of restored sidebar selection, and keep status and error fallback scoped to the selected project.
Release note: 提交历史默认贴底，展开占用一半内容区，支持拖拽调整并统一使用顶部刷新。切换项目后自动识别 Git 工作区，无需手动刷新。
Preserve read-only question and answer reviews after successful submission or skipping, including multiple rounds and local conversation reloads. Keep failed submissions retryable and preserve newer questions and terminal states when responses arrive late.
Validation: typecheck and lint passed; full suite had 3900 passes, 66 skips and one Agent package timeout that passed on isolated rerun. Real OpenCode UI acceptance was not run. Previously discarded answers cannot be recovered.
Release note: 修复 OpenCode 提交回答后会话仅显示已提交的问题；现在保留原问题及对应答案，多轮问答和跳过记录在重新打开会话后仍可查看。
Image requests previously sent only the current prompt and attachments, so
follow-up edits could lose the previous image and conversation requirements.
Reuse the latest successful image from the existing artifact store together
with text history, including after reopening the app. Explicit attachments
take priority, and configured model IDs remain unchanged.
When image editing is explicitly unsupported, continue with text-based image
generation and retain a quiet message footer explaining the missing image
context. Keep genuine provider errors and cancellation visible. Add focused
regressions, persisted footer metadata, and aligned feature documentation.
Validated with four real sub-gpt-image-2 image requests and a simulated
unsupported-edit response. Image regressions, typecheck and lint passed;
the full working-tree run had two unrelated concurrent Runtime-menu failures.
Release note: 连续改图会自动带上上一张图片和文字上下文，重开对话后仍可继续修改。上游明确不支持编辑时仍可按文字生图，仅在回复底部轻提示未使用参考图片。
Validate Dify and RAGFlow outcomes, normalize FastGPT URLs, and report response envelopes, score structures and actual rerank state. Align integration documentation with official APIs and archive 33 redacted read-only requests with strict HTTPS verification.
Validation: 43 focused tests, typecheck and lint passed. Full test runs exceeded 60s and 240s without a summary; this limitation is recorded in the feature progress document.
Write fixed commit objects from the fixture index rather than invoking
Git commit with an implicit machine identity. Preserve real history,
rename diff, pagination and dirty branch-switch checks without changing
Git author or committer configuration or skipping any affected test.
Add fixture checks for valid objects, parent linkage, unchanged dirty
content and unchanged repository configuration. Record the CI regression
introduced when the prior test identity override was removed.
Validation: workspace tests passed 11 with 2 filesystem skips; release
notes verification, typecheck and lint passed. Full tests passed 3845 with
66 skipped and one unrelated ActivityPanel timeout; all 16 ActivityPanel
tests passed on isolated rerun. Product source and approved bilingual
Desktop 0.12.11 / Agent 0.11.23 release notes are unchanged.
Record the approved bilingual release notes, classify long-task limits and
premature truncation as fixes, and align both feature matrices with the
verified workspace, browser, scheduling and Runtime behavior.
Advance the Agent lock to 0.11.23 for remote file and Git management, optional
model limits, larger bridge exchanges and the unlimited-duration connection
fix. Agent publication reads version-specific approved notes and remains
separate from Desktop Latest. Upgrade Desktop to 0.12.11 before updating Hosts.
No LoongArch preview is included.
Validation: release notes verification, all 3845 tests (66 skipped), typecheck
and lint passed. Current-source Linux x64 workspace and bridge checks passed;
managed ACP Ask and Execute completed two real model requests, both HTTP 200,
with delivered results and cleaned test resources. Native production packages
and publication remain subject to GitHub Actions after candidate approval.
Release note: 准备 Desktop 0.12.11 与 Agent 0.11.23 发布；远程文件和 Git 管理、长任务限制及连接超时修复需先升级桌面端，再更新 Host 上的 Agent。
Group the workspace view switch and refresh action in one header, show only
file or Git actions for the active view, and remove the isolated root icon.
Subdirectory breadcrumbs retain root navigation and restore keyboard focus.
Use the shared form style for branch search in a bounded, scrollable panel.
Confirm that fresh settings already default to automatic Execute tool
authorization, and add regression coverage without changing saved policies.
Update the workbar design and record isolated production Electron validation.
Validation: 354 focused tests, typecheck, lint and development bundle passed.
The full suite passed 3843 tests and skipped 66; an unchanged OpenCode stream
lifecycle timing test failed once and passed its isolated rerun.
Release note: 整理工作区文件与 Git 工具栏，移除孤立的根目录图标并保留子目录导航；分支搜索框、文字和按钮统一样式，宽窄工作栏下均保持紧凑布局。
Restore Ask workspace reads and output paging without granting write access,
and let programming Subagents reuse their parent's browser page. Allocate
browser resources only when used, keep Agent pages aligned with workbar tabs,
recover released bindings, and remove unused automatic screenshot work.
Preserve inherited Runtime selections during remote usage persistence and
manual compression. Keep viable channel model fallbacks, display the actual
scheduled-task mode and Harness model source, and avoid rediscovery delays
when calling healthy MCP tools. Fix directory/history refresh, case-only
renames, rename-aware commit diffs, and FastGPT probe success reporting.
Add regression coverage and align feature documentation. The Agent-side
connection timeout fix requires deploying an updated Agent; existing Ask
read-only and Execute authorization boundaries remain unchanged.
Final validation: 3839 tests passed, 66 skipped; typecheck, lint, development
build, and diff checks passed. Isolated Electron UI/MCP workflows passed with
a loopback model fixture. Current-source Linux x64 Host checks passed with
three successful real model requests. Chromium loadURL rejection was not
reproduced in the UI; released-binding recovery has automated coverage.
Release note: 修复 Ask 读取项目文件、浏览器页面显示与恢复、文件管理及模型配置跟随中的不一致，并减少无用截图和 MCP 等待。定时任务显示实际工作模式，远程会话可正常保存继承 Runtime 的用量；远端连接超时修复需更新 Agent 后生效。
Persist the conversation knowledge retrieval mode across restarts so timed messages use the saved setting. Clarify shared conversation authorization in the scheduled-task documentation and correct the renderer test query type.
Validation: 374 focused tests passed; typecheck and lint passed. The full suite had two failures that passed on focused reruns. Real remote validation remains blocked by the temporary harness SDK module loading error; no model requests were made.
Release note: 修复会话知识检索模式在重启后丢失的问题，定时消息继续使用会话保存的检索设置，并同步任务授权说明。
Preserve explicit Runtime model selections and Harness configuration rules without pinning resolved defaults. Expose fixed Runtime models in project settings and keep SSH conversations inheriting their project selection.
Dispatch scheduled instructions through ordinary conversation requests using current history, Runtime, work mode, and knowledge scope. Persist conversation work modes and update scheduling documentation and regression tests.
Validation checkpoint: typecheck, lint, and 104 focused tests passed. The SSH inheritance regression passed in isolation, but its process timed out on exit. Full-suite completion, remaining model documentation review, and real end-to-end validation are still pending.
Release note: Preserve fixed model choices and let scheduled messages use their conversation's current settings and history.
Preserve long histories and parsed attachments, expose paged process and Subagent output, isolate MCP discovery failures, and expand browser snapshots and knowledge result selection. Remove fixed Harness output and updated remote Runtime duration limits.
Keep released Agent profile and v1 bridge compatibility while enabling larger framed exchanges and optional limits for updated peers. Document remote coverage requirements and real-model boundary validation.
Release note: Long model tasks can retain more context and continue reading complete tool output; remote execution preserves compatibility with released Agents. Updated remote limits require an updated Agent and Runtime bundle.
Reflect queued and active schedule runs in task status, refresh tasks on queue changes, and prevent duplicate Run Now actions until the current run finishes. Add lifecycle regression coverage and update scheduled task behavior documentation.
This file-level commit also includes concurrent conversation history, browser workbar, and runtime selection changes in the affected files. Focused schedule regressions passed; full workspace validation has unrelated failures as recorded during development.
Release note: Scheduled tasks now refresh their queued and running status and keep Run Now disabled while an occurrence is active.
Direct models now use bundled ripgrep for workspace discovery, read files through resumable pages, and edit multiple files with apply_patch. Ask can search and read while Execute retains write access, and fixed total input limits no longer constrain these tools.
Release note: 直连模型现可使用随包 ripgrep 搜索项目、分页读取任意大小的文本，并通过多文件补丁修改代码；设置中的相关能力统一显示为工作区文件和开发工具。
Add read-only Dify, FastGPT, and RAGFlow probes, a redacted live capability baseline, and focused adapter tests. Document version-specific API and UI behavior, including Dify 0.x/1.x differences and RAGFlow GraphRAG evidence.
Users previously had to leave the workbar to manage project files and inspect Git branches or commit history. Add file and directory creation, rename, move, delete and properties, plus searchable branches, branch creation and switching, explicit fetch, and commit file diffs.
Keep file previews in the right workspace pane with directory expansion, selection, scroll and focus preserved on return. Add compact action menus, directory pickers, preview refresh and source switching. Embed the shared browser session directly in the workbar with navigation, resizing and fullscreen controls instead of a separate interaction window.
Moves stay within the current workspace, deletion is permanent after confirmation, and Git switch reports conflicts without discarding or stashing changes. Typecheck, lint and staged whitespace checks pass; existing full-suite failures and the blocked real remote directory validation remain recorded in the workbar progress document.
Release note: 可在右侧工作区管理文件、切换 Git 分支并查看提交历史，返回预览后保留浏览位置；浏览器支持在工作栏内直接交互和全屏显示。
Automatic native Git snapshot diffs could block concurrent OpenCode sessions
after broad rewrites or line-ending conversions. Managed desktop and Agent
profiles now skip snapshots that GoodBuddy does not use, while preserving
file tools, native subagents, workspace Git operations, and Ask/Execute rules.
Drain otherwise-unused Continue utility-host stdout to prevent console output
from filling the pipe and blocking requests. Correct the lifecycle regression's
body observer to watch the underlying reader rather than a buffered wrapper.
Prepare Desktop 0.12.10 and Agent 0.11.22 with aligned bilingual notes and
feature documentation. Existing runtimes require restart or Agent update.
Full desktop UI before/after and Linux Agent production-path checks passed;
3626 tests passed, with 61 existing platform/manual skips.
Release note: 修复批量文件改写或换行转换后的托管 OpenCode 并行停顿，以及 Continue 控制台输出阻塞；本机升级重启后生效，远程项目请配套更新 Agent。
Requests could appear to keep working while OpenCode waited to retry, and
retry backoff was described as an active attempt. Report native OpenCode
retry timing, distinguish direct-model backoff from dispatch, and show
preparation, tool activity, pending input, and terminal states accurately.
Keep only the message-footer indicator static. Conversation-list activity
indicators, send/stop controls, retry limits, and command replay behavior
remain unchanged. Align the feature documentation with verified behavior.
Validation: 3618 tests passed, 61 skipped; typecheck and lint passed.
Isolated desktop validation used five real text-model calls across three
profiles, with controlled local fault injection for retry transitions.
Release note: 消息底部现在准确区分请求准备、等待重试、开始重试和需要回答的状态，不再用闪烁暗示仍有进展；会话列表闪点及发送、停止行为保持不变。
Starting another conversation could replace a busy local OpenCode server
after a brief health-check failure, leaving the existing conversation
reconnecting to a dead endpoint.
Keep active conversations and native Compact operations on their current
server. Propagate server exit and shutdown to bound requests and cancel
event-stream retry waits so failures are explicit instead of hanging.
Do not automatically replay commands.
Release note: 修复同时使用多个本机 OpenCode 会话时旧会话可能一直等待的问题。服务异常退出时会明确提示失败，不会自动重复执行命令。
Desktop 0.12.7 omitted OpenCode's offline node_modules because the resource
copier excludes node_modules at its source root. Copy from the parent and
verify every prepared configuration file in the final package. Bundle DSH
session projection with the unpacked Host instead of leaving an ASAR-only
external import, and reject external DeepSeek imports in shipped chunks.
Run packaged OpenCode plugin import and the real DSH UtilityProcess handshake
in Windows x64 CI before removing unpacked output. Add regressions using the
actual electron-builder copier and Vite externalization configuration.
Prepare approved Desktop 0.12.8 notes and change macOS defaults and CI to DMG
only. Update aggregation, website downloads and update readers for 12
installers and 20 assets. Current readers retain historical DMG/ZIP support;
the user accepts that older macOS GitHub readers and all older mirror readers
need manual upgrades. Agent remains 0.11.21, with no new Agent or LoongArch tag.
Validation: 3608 tests passed, 0 failed, 61 skipped across 339 files;
typecheck, lint, release-note validation, website static checks and 18 website
tests passed. No local production build or packaged-app launch was performed
at the user's request; actual packaged startup validation runs in CI.
Release note: 修复安装后 OpenCode 离线依赖缺失和 DeepSeek Harness Host 启动退出；macOS 改为仅提供 DMG，受影响旧版用户需从下载页手动升级。
Set the approved Desktop and Agent versions and package the approved Chinese
and English notes without changing published release history. Keep OpenCode
pinned to 1.18.29 and require Desktop 0.12.7 before the new Agent package.
Align both feature lists and affected design documents with reference-image
editing, inherited expert work modes, preview boundaries, and remaining
release acceptance. Application navigation remains design-only. No LoongArch
preview is planned for this release; Agent publication remains non-Latest.
The first candidate full run exposed a browser test clicking Send before it
was enabled. Wait for startup readiness using the existing test convention;
production code is unchanged by this release-preparation commit.
Validation: final full suite 3597 passed, 0 failed, 61 skipped across 338 files.
Typecheck, lint, bilingual release-note validation, version consistency,
unchanged historical notes, whitespace and relative-link checks passed.
No additional real model calls, local production builds, pushes or tags.
Concurrent navigation, workbar and roadmap design edits are not included.
Execute could stall on workspace-external access or native child-session
permissions, while structured questions could be lost or answered incorrectly.
Route owned OpenCode descendant interactions, correct Continue question and
custom-answer handling, and resolve DSH Execute ACP permissions without a
second approval. Ask remains read-only and unrelated sessions stay isolated.
Restore the required DSH session projection service. Reuse offline OpenCode
configuration by plugin version, remove unused npm shims from remote payloads,
normalize dependency modes, and accept real compound descriptors up to 4 MiB.
Fix intermittent Agent attach rejection and closed-handshake waits.
Add focused and real-Runtime regressions, document interaction boundaries and
validation evidence, and include the existing application/tool navigation
design drafts. Those navigation changes are design-only, not implemented UI.
Validation: full suite 3597 passed, 0 failed, 61 skipped across 338 files;
typecheck, lint, release-note verification, whitespace and relative-link checks
passed. Current-source Linux x64 installation, Ask/Execute, native subagent
external writes, reconnect and stop/bootstrap passed with 10 completed real
model calls, 13 native endpoint tests and 20 consecutive handshakes.
System sleep/wake and release CI packaging remain unverified.
Release note: 修复 Execute 访问工作区外目录或运行原生子代理时卡在权限等待的问题，并恢复 OpenCode 与 Continue 结构化问答的正确转交。修复远程 Agent 安装与重连失败；Ask 仍保持只读，任意终端密码或业务确认不会自动代答。
Check cached embedded servers before reuse, isolate concurrent cancellation, and serialize request MCP registration and cleanup. Preserve native controls across session switches and recover usable partial inventories.
Share the bundled offline OpenCode configuration with desktop and Agent launch paths, including model bridge environment forwarding.
Validation: focused tests and typecheck passed; lint passed before the final fixes. The full test run exceeded 120 seconds and reported a DS stdio MCP failure. Current-source Agent host and real suspend/resume validation remain pending.
Release note: 修复切换会话后 OpenCode 原生控件消失及并发取消相互影响的问题，并完善内嵌服务复用检查和桌面、Agent 的离线依赖配置。
Record approved bilingual notes for complete subagent progress, workspace diffs, knowledge selection, OpenCode stream cleanup, external command directories, and restored GitHub Agent updates. Remote progress and diff improvements require the new Agent package. No LoongArch preview is included.
OpenCode child cards now show ordered text, reasoning and tool progress,
with the final result displayed separately. Local SDK and remote ACP paths
share child tracking, including progress retained across recovery and cancellation.
Remote live progress requires an Agent package containing the event plugin;
older Agent packages still expose final results without inventing missing history.
Knowledge-library rows remain clickable when native label activation changes
focus. File workspace changes use colored Git status letters and per-file
staged/unstaged diffs, including deleted, renamed and untracked files.
Refreshing reloads the selected diff and expanded directories, and more than
50 changes remain accessible through incremental loading.
Validated with 3552 passing tests, typecheck, lint, desktop build, current-source
Linux Host scenarios and isolated Windows desktop UI interactions.
Release note: 展开 OpenCode 子代理卡片可查看执行过程、工具详情与独立最终结果；知识库支持整行勾选。工作区支持单文件暂存与未暂存差异，刷新会更新当前差异，更多变更可继续加载。远端实时过程需要更新 GoodBuddy Agent。
Completed OpenCode requests left their event streams open across repeated and
parallel conversations. Each request now closes its own subscription on exit
without cancelling peer sessions, and cancellation preserves its original cause.
Direct-model Execute commands no longer restrict cwd to the workspace. Absolute
paths, parent-relative paths, and symlinks to external directories are accepted.
The workspace remains the default and relative-path base; Ask stays read-only,
and local execution binding, directory validation, and process cleanup remain.
Add real OpenCode lifecycle and real Shell tool-loop regressions, plus matching
feature documentation. Focused checks: 182 passed, 1 skipped; typecheck and lint
passed. Full shared-worktree suite: 3542 passed, 57 skipped, 1 unrelated sidebar
heading failure, reproduced separately and documented.
Release note: 修复 OpenCode 连续及并行对话结束后事件连接未释放的问题；直连模型 Execute 命令现在可直接指定工作区外的绝对路径、相对路径或符号链接目录，Ask 只读边界保持不变。
The GitHub release history exceeded the catalog response limit, preventing users from checking Agent updates. Read ten releases per page and stop at the first eligible catalog while preserving the existing 100-release search window and signature checks.
Release note: 修复选择 GitHub 来源时 Agent 更新检查提示发布目录超出大小限制的问题；镜像更新和已有本地包保持不变。
Remove the extracted ShareServer prototype and its server-specific documents
from Desktop. Link the shared network architecture to goodbuddy-shareserver,
while retaining Desktop ownership of the cross-product architecture.
Preserve local generated artifacts and ignore the old server build output.
Desktop runtime behavior and released user data are unchanged. Desktop tests,
typecheck, and lint passed; the standalone service was validated separately.
Exclude the independent prototype from root lint and explicitly anchor TypeScript parser configuration to the Desktop repository. This fixes clean CI runners detecting both root and ShareServer configuration directories. Product behavior and approved release notes are unchanged.
Exclude the unfinished standalone ShareServer prototype from the Desktop root test suite and remove its CI dependency installation. Preserve its independent test configuration and document the release boundary. Approved Desktop and Agent release notes remain unchanged.
Record approved bilingual release notes and align Desktop and Agent version metadata. Publish Desktop first so users can upgrade before the shared Agent catalog gains macOS arm64 entries. No LoongArch preview is included.
Extend GoodBuddy Agent packaging, installation, SSH attachment, Runtime lifecycle, and Desktop package management to macOS arm64. Linux x64, Linux arm64, and Darwin arm64 share the existing signed catalog. Preserve Ask read-only behavior and Execute account permissions, and stop detached tool descendants during Mac cancellation.
Upgrade Desktop before publishing the first mixed-platform catalog because older Linux-only readers reject Darwin entries. Intel Macs are not supported; no launchd service or administrator access is required. This commit does not publish a release.
Validation: 3520 tests passed, 57 skipped; typecheck, lint, and development build passed. Real Mac installation, Ask/Execute, cancellation and reconnect passed; final Linux installation, Ask and reconnect passed. Nine real model requests were recorded.
Release note: 支持将 Apple Silicon Mac 用作远程项目 Host，可安装 macOS arm64 Agent 并使用 Ask、Execute 和断线重连；与 Linux 共用一个包目录，发布混合目录前需升级 Desktop。
Knowledge selection was shared globally, document readiness overstated usable content, and important import and source-management paths exposed technical details or lacked safe recovery actions. Knowledge scope now persists per conversation, readiness reflects real processing channels, and the workspace guides users from creation and import into cited chat with controlled source actions.
Release note: 知识库现按对话保存使用范围，准确区分可用、处理中和失败文档，并提供更简洁的创建、检索测试、来源打开、删除确认和失败重试流程。
ShareServer previously existed only as design documents. The repository now includes a standalone Hono service and a responsive React management-console prototype covering administrator and member workflows, with local build, test, and container entry points.
The sharing design now includes organization-scoped digital agents and explicit local partitioned-memory selection for @ calls. Memory documentation consistently uses partitioned memory and treats time or internal events only as optional recall triggers.
This remains an interactive prototype: fixture actions do not persist, and production authentication, PostgreSQL isolation, device connections, routing, relay, and federation still require implementation.
Release note: 新增可独立运行的 ShareServer 管理控制台交互原型，并完善组织数字人与分区记忆共享设计；当前原型数据和操作不会写入生产系统。
Bump Desktop to 0.12.4 and the separately published GoodBuddy Agent to 0.11.18. Add the approved bilingual release notes for HTML previews, long-run and queue reliability, remote terminal-state convergence, and faster OpenCode follow-up on Linux ARM.
This candidate uses the standard six-platform x64/arm64 matrix and does not request a LoongArch experimental preview.
Native capability loading previously cold-started a temporary Runtime and disposed it immediately, so Linux ARM users paid the OpenCode startup cost again on their first request. Native inventory loading now uses the cached execution Runtime, preserving normal idle eviction and lifecycle cleanup.
Release note: 优化 Linux ARM 上切换 OpenCode 后的响应速度；原生能力清单与后续对话现在复用同一 Runtime，避免重复冷启动。
Agent replies that produced HTML could only be read as source, so users had
to copy the markup into an external file to see the result. Completed Agent
and Subagent output now renders complete HTML documents, single-root HTML
replies, and `html` / `htm` code blocks as an isolated static preview in
place. Each preview keeps a persistent notice that scripts and network are
disabled and offers the same icon actions as Mermaid diagrams: a source
toggle and a full-screen preview. The new "render HTML in conversations"
switch under Platform Features / General is on by default, applies to the
open conversation immediately, and falls back to HTML source when disabled.
Previews cannot run scripts, reach the network, submit forms, open windows,
or load remote media; author stylesheets and inline styles are preserved so
the intended layout survives, and remote resource loading is refused by the
preview CSP instead.
Direct-model replies no longer fail once they grow large. The byte caps on
streaming responses, single stream blocks, non-streaming replies, model tool
arguments, tool request context, aggregate tool results, and accumulated
answer text are removed, so users stop seeing "模型接口流式响应超过安全限制"
on long answers. Response length is now governed by the request timeout,
cancellation, and context compression. Image generation responses keep their
decodable-image validation.
A message sent after a failed or interrupted run stayed in the pending queue
instead of running. The conversation is now released whenever a run reaches
any terminal state, not only when the renderer reports readiness again, and a
dispatched queue item whose run never starts releases its reservation and
returns to pending. Later messages dispatch without extra user action.
Release note: Agent 回复中的 HTML 现在可直接在会话内预览，支持查看源码和全屏预览，默认开启并可在“平台功能 / 通用设置”关闭；直连模型的长回复不再因大小限制中断；上一次运行失败或被中断后，新发送的消息会立即执行而不再排队等待。
Long Runtime sessions could silently stop retaining tool and Subagent activity, truncate Run history details, and reject workspace previews over 256 KB. Run history now persists in Main SQLite without the former fixed record or detail caps, surfaces possible loss from legacy caches, and renders long histories in batches.
Local and remote workspace previews now load UTF-8-safe 256 KiB pages on demand instead of rejecting larger text files. The legacy migration, clear-data path, IPC contracts, tests, and feature documentation are updated with the resulting behavior.
Release note: 长时间 Runtime 运行现在会完整保留工具和 Subagent 活动；运行记录迁移到 SQLite，并提示旧版可能已丢失的内容。超过 256KB 的本地及远程文本文件可通过分页面预览继续加载。
OpenCode status checks could cold-start throwaway servers, unrelated
conversations could block each other, and model, Continue, or remote paths
could replay or duplicate work under failure and backlog conditions. Settings
could also lose edits or overwrite unrelated model fields.
This update makes status checks passive, keeps per-conversation ordering while
allowing concurrency, preserves native long-running work with bounded display
capture and backpressure, narrows retries to known pre-dispatch failures, and
aligns remote Ask permissions without reducing Execute. It also keeps
Embedding updates narrow, improves settings and browser guidance, removes the
unused legacy Main gateway state, and records the validated Desktop 0.12.3 and
Agent 0.11.17 release metadata.
Release note: 改善 OpenCode 切换、多会话长任务、模型重试、远程 Ask/Execute 与设置保存可靠性，并完成 Desktop 0.12.3 / Agent 0.11.17 发布准备。
mesalogo