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。
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。
Image editing requests included a generation-only response_format field that some compatible providers reject. Editing now sends only supported defaults while preserving explicitly configured custom request fields.\n\nRelease note: 修复部分图像服务商修改图片时报错的问题；参考图编辑现在可以正常提交。
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。
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。
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。
Model connections can now save bounded JSON request headers and top-level body fields. Direct requests and Continue support both; local OpenCode and DeepSeek Harness use only native header support, while the managed remote OpenCode gateway applies both.
Runtime, protocol, and authentication fields keep precedence. Custom values are ordinary connection settings, not credential storage. This migrates Runtime settings from v20 to v21 and updates the DeepSeek control and managed Runtime ACP capabilities.
Release note: 模型连接详情新增自定义请求头和请求体；直连与 Continue 支持两者，本地 OpenCode 和 DeepSeek Harness 仅应用原生支持的请求头，受管远程 OpenCode 支持两者，界面会明确提示生效范围。
Direct model conversations previously stopped after the first transient network error or request timeout, including TLS setup ECONNRESET failures. Text requests now retry up to three times with cancellable backoff before visible output, while partial streamed responses are never replayed.
Release note: 直连模型遇到临时网络故障或请求超时时，会在尚未输出内容的情况下自动重试最多 3 次，并支持随时取消。
Direct model conversations can now execute bounded platform-native processes and delegate one level of programming work without duplicating tools in Agent Runtimes. Process and Subagent activity follows existing workspace, mode, approval, cancellation, and lifecycle boundaries, with chat attribution and complete feature documentation.
Remove the static Renderer bundle baseline gate and its build-only manifests; production builds continue to compile all processes and verify the control-plane installer. Also include the pending documentation link normalization.
Release note: 直连模型现在可在 Execute 模式运行本机命令，并可委派编程 Subagent 完成修改、测试和复核；构建流程不再因静态 Renderer 包体基线阻塞。
Direct Anthropic connections reused the configured context window as max_tokens, which could impose an unintended output value or send a provider-invalid request. Anthropic connections now expose an independent per-response maximum output setting, defaulting to 32K for the protocol-required field, while OpenAI direct requests remain without an added output token cap and direct tool rounds remain unmetered.
Release note: Anthropic 直连模型现在可单独配置单次最大输出，不再复用上下文窗口；OpenAI 直连请求和工具调用轮次不增加额外配额。
Add managed SSH Hosts and projects with signed side-by-side GoodBuddy Agent and OpenCode Runtime installation, Ask and Execute workspace access, a Main-owned model bridge, detached Agent lifecycle, durable ACP output recovery, and project activation UI.
Use protocol v2 with bounded FIFO frames over SSH instead of custom credit, priority, batching, or model-message chunking. Mismatched Host Agents now upgrade automatically, an exactly verified Node runtime is reused on-host, and local Portable builds can embed strictly verified Agent artifacts.
Release note: 新增托管 SSH 远程项目，可自动安装或升级签名 Agent/OpenCode Runtime，并在 SSH 断线后继续 Agent 工作和同步未确认输出。
Save and test could accept a successful endpoint response without proving that the selected model generated output, while changing a connection URL or temporarily disabling authentication could require the API Key again.
Model tests now issue bounded real text or image generation requests and validate their output. API Keys remain encrypted with their connection across URL and authentication changes until explicitly cleared or the connection is deleted, and the form places the key directly below authentication. Real tests may incur a small amount of provider usage.
Release note: 修复模型“保存并测试”仅验证连通性的问题；现在会执行真实文本或图片生成测试，并在修改地址或临时关闭认证时继续保留该连接的加密 API Key。
Direct-model conversation history began carrying GoodBuddy UUIDs after durable context compression was added. OpenAI Responses rejected follow-up messages because provider message IDs must start with msg_, and the same metadata could also reach other protocol payloads.
Model request serialization now strips local IDs across Anthropic Messages, Chat Completions, and Responses while retaining provider IDs during Responses tool rounds. Regression coverage includes compression, tools, and a gated real-model follow-up.
Release note: 修复默认直连模型在继续对话或重新编辑发送时可能因消息 ID 格式错误而失败的问题。
OpenCode and Continue customization was previously planned but unavailable, and Runtime-native capabilities were not represented consistently across providers. GoodBuddy now provides secure Main-owned customization settings, truthful native inventories, OpenCode Agents and Commands, Continue Rules and Prompts, DSH Web Search/Fetch, MCP Prompt and Resource metadata, and manual compaction where supported.
Native capabilities are presented in eleven accessible tabs with Tools separated from Commands, LSP, and Formatters. Tool source and Ask/Execute availability are explicit, external OpenCode remains connection-only, Continue reports unsupported static tool discovery instead of advertising unreachable Skills, and disposable inventory probes avoid retaining background runtimes.
Ask remains read-only at the Runtime boundary, Execute keeps the existing authorization controls, and credentials remain confined to Main.
Release note: 新增 OpenCode、Continue 与 DeepSeek Harness 的 Runtime 原生定制与真实能力清单；工具来源、Ask/Execute 可用性、上下文压缩和 MCP 元数据现在可清晰查看，同时继续保持 Main 进程凭据保护与现有权限边界。
Multiple model invocations in one task could share a call identifier when
a provider omitted or reused its response ID. Database upserts then replaced
earlier tool-round or summary usage, causing cumulative activity totals to
undercount successful calls.
Each completed direct-model invocation now receives a unique local call ID
while retaining the provider ID as diagnostic context. Tool rounds, final
responses, image calls, and repeated compression summaries are therefore
stored independently.
Release note: 修复连续工具调用或上下文摘要可能覆盖前序模型用量的问题；现在每次成功调用都会分别计入运行记录和累计统计。
Long direct-model conversations mixed provider usage with local estimates,
and Agent tool rounds could remain above the configured compression target.
Compression state and status markers also did not reliably survive restarts
or bounded history rollover.
Direct-model calls now prefer provider-reported usage, compact complete
conversation turns and Agent tool rounds within reserved payload budgets,
and persist reusable summaries with scope-specific markers. The chat meter
separates latest-call usage from estimated compressed conversation size,
while failed or cancelled calls retain the last successful measurement.
Release note: 直连模型现可在长对话和多轮工具执行中自动压缩旧上下文，并分别显示本次调用用量与压缩后对话估算；摘要会自动保存并跨重启复用，无需手动操作。
mesalogo