Hexestra v0.8.1

Added / 新增

• 设置页新增 Claude Code / Codex 标签，可查看和管理对应的 Skills 与 MCP 配置；Codex 托管 Skill 可复制到可编辑目录。
  Added Claude Code / Codex tabs for managing each backend's Skills and MCP configuration, including copying managed Codex Skills into editable folders.
• 欢迎页现在可以关闭，标签栏末尾的加号可重新打开；关闭最后一个标签时会自动显示欢迎页。
  Made the Welcome tab closable, added a trailing button to reopen it, and restored it automatically when the last tab closes.

Changed / 变更

• 统一侧栏、标签页和设置界面的卡片样式；左栏支持带动画的展开与收起，AI 历史对话改为悬浮菜单，NetMap 使用连续背景。
  Unified card styling across sidebars, tabs, and Settings; added animated sidebar collapse, a floating conversation-history menu, and a continuous NetMap background.
• 输入框和下拉框统一为 8px 圆角，原生下拉菜单改为与主题一致的自定义菜单，并补齐相关中文文案。
  Standardized input and dropdown corners at 8px, replaced native dropdown popups with themed menus, and completed related Chinese copy.

Fixed / 修复

• 修复 Codex 调用 Hexestra MCP 工具时被前置审批策略拒绝的问题；Hexestra 内部的逐项授权检查仍然生效，并显示具体错误信息。
  Fixed Codex MCP calls being rejected before Hexestra's own authorization checks and surfaced concrete tool errors.
• 修复 AI 输入框异常变高及圆角裁切、侧栏滚动轨道外溢和部分标签描边缺失的问题。
  Fixed oversized or clipped AI input, sidebar scroll tracks escaping their panels, and missing tab outlines.

Downloads / 下载

• Windows x64: .exe installer
• Linux x64: .AppImage
• macOS Intel / Apple Silicon: .dmg
• Public source archive and SHA-256 checksums are attached.

Hexestra v0.8.0

Added / 新增

• 新增 Codex App Server 后端，可在新建会话时选择 Claude Code 或 Codex；历史会话保留原后端，分支继承源会话后端。Codex 支持本机和 WSL 运行，并复用已安装 CLI 的登录状态。
  Added a Codex App Server backend selectable for new conversations. Existing conversations retain their backend, forks inherit it, and Codex uses the installed CLI and its sign-in state in Native or WSL mode.
• Codex 通过 Hexestra 管理的 MCP 桥接使用项目工具和权限检查；项目 Skills 同步到 .agents/skills，设置页提供运行时与桥接诊断。
  Routed Codex project tools through Hexestra's permission-checked MCP bridge, synced project Skills to .agents/skills, and added runtime and bridge diagnostics in Settings.
• 新增 Claude Code 与 Codex 的可用模型列表及模型支持的推理强度选择；演示模式现可通过可配置快捷键模糊项目敏感信息。
  Added runtime model catalogs and supported reasoning-effort choices for Claude Code and Codex, plus a configurable presentation-mode shortcut that obscures sensitive project details.
• Codex 输入框的斜杠菜单合并已接入命令与可用 Skills，并在技能变化后刷新候选。
  Merged supported Codex commands and live Skills into the slash menu, refreshing suggestions when Skills change.

Changed / 变更

• 将自主程度作为模型的行动提示，与工具权限模式分开；压缩聊天侧栏布局，并统一模型、权限等弹层的样式和中文文案。
  Made autonomy a guidance setting separate from tool permissions, compacted the chat sidebar, and aligned model and permission menus with localized copy.
• 升级到 Tailwind CSS 4，并加强 Windows 桌面启动检查。
  Upgraded to Tailwind CSS 4 and strengthened the Windows desktop startup smoke check.

Fixed / 修复

• 修复 Codex 任务中多段文字被后续输出覆盖的问题，并让项目内 Markdown 文件链接在 Hexestra 编辑器标签页打开。
  Preserved multiple Codex text messages across a turn and opened project Markdown links in Hexestra editor tabs.
• 找不到可执行文件时显示可排查的错误说明，而不再直接显示原始错误信息。
  Replaced raw warning message failures with actionable diagnostics.

Hexestra v0.7.0

Added / 新增

• 新增 asset_import，可导入 Nmap XML 与 httpx JSONL 扫描结果，并生成类型化资产及其关系。 Added asset_import for Nmap XML and httpx JSONL scan results, producing typed assets and relationships.
• 补齐 README、用户手册、架构、领域模型、Agent 运行机制与上下文维护文档的中英文版本。 Completed bilingual coverage for the READMEs, user guide, architecture, domain model, Agent runtime, and context-maintenance documentation.

Changed / 变更

• 重构 Agent 工具管理与应用版本解析，并统一设置页列表行等界面组件。 Refactored Agent tool management and application-version resolution, and unified shared Settings list-row components.
• Agent 现在只能启用项目代理，关闭代理仍需由操作者手动执行。 Agents can now enable the project proxy but disabling it remains an explicit operator action.

Fixed / 修复

• 修复 Agent 任务门禁反复要求创建或聚焦任务的问题：只读工具无需任务，执行工具按当前对话分支的聚焦任务校验，任务切换会立即生效。 Fixed Agent task gates repeatedly requesting task creation or focus: read-only tools remain available, execution tools validate the focused task on the current conversation branch, and task changes take effect immediately.
• 稳定大型 NetMap 布局等高负载测试在并行完整测试套件中的超时行为。 Stabilized timeout behavior for heavy cases such as large NetMap layouts under the parallel full test suite.

Hexestra v0.6.0

Added / 新增

• 新增持续 Streaming Input Agent runtime，支持会话复用、输入队列、定时唤醒、后台通知、分支运行时状态和子 Agent 轨迹。
  Added a persistent Streaming Input Agent runtime with conversation reuse, queued inputs, scheduled wakeups, background notifications, branch runtime state, and subagent traces.
• 新增从本地路径导入外部 Claude Skill 的能力，可在设置中管理导入项并查看能力信息。
  Added local-path import for external Claude Skills, with Settings management and capability visibility.
• 将 Tool Catalog 重构为 AI 提示目录，支持设置页结构化增删改、旧格式迁移和任务匹配；目录不再负责探测、安装或执行工具。
  Refactored Tool Catalog into an AI-facing prompt inventory with structured Settings CRUD, legacy migration, and task matching; it no longer probes, installs, or executes tools.
• 新增 Agent 任务运行轨迹展示，并补充完整的用户手册、架构说明、领域模型和 Agent 运行机制文档。
  Added Agent task run-trace views and comprehensive user, architecture, domain-model, and Agent-runtime documentation.

Changed / 变更

• 对话选择器改为支持键盘操作的可访问菜单，并优化收件箱入口、输入状态和中英文文案。
  Replaced the conversation selector with an accessible keyboard-navigable menu and refined the inbox entry point, input state, and bilingual copy.
• Shell auto 模式现在会识别本地和 SSH Shell 类型；已识别 Shell 等待验证退出码，无法确认的 raw Shell 返回 completed_unverified。
  Shell auto mode now detects local and SSH shell flavors; known shells wait for a verified exit code, while unresolved raw shells return completed_unverified.
• 改进 Claude Code 运行时适配，包括 Windows npm shim、连续输入分段和取消后的运行状态同步。
  Improved Claude Code runtime adaptation, including Windows npm shims, consecutive-input segmentation, and post-cancellation runtime-state synchronization.

Fixed / 修复

• 修复排队输入在当前 Agent turn 被取消或中断后丢失、串入错误回复或错误显示为排队状态的问题。
  Fixed queued inputs being lost, attached to the wrong reply, or displayed as a stale queued state after an Agent turn was cancelled or interrupted.
• 修复 SSH 文件传输无法取消，以及自动 Shell 探测失败后仍允许 Agent 执行的问题。
  Fixed non-cancellable SSH file transfers and Agent execution continuing after automatic shell detection failed.
• 修复 Mihomo 运行时选择错误无法清晰反馈到设置界面的问题。
  Fixed Mihomo runtime selection errors not being surfaced clearly in Settings.

Hexestra v0.5.0

Added / 新增

• 新增基于 MITRE ATT&CK 的任务树、离线目录、Technique/Tactic 上下文、限制服务、工具目录和任务步骤管理。
  Added an offline MITRE ATT&CK task tree with Technique/Tactic context, restrictions, a tool catalog, and task-step management.
• 新增工作流与知识提炼中心，可在任务树中运行工作流，或从文档和对话生成 Restrictions、Skills 与 Workflows 候选。
  Added task-tree workflows and a knowledge refinery for turning documents and conversations into reviewed Restriction, Skill, and Workflow candidates.
• 新增分支级历史分页与 JSONL 历史仓储，支持长对话分层加载、子 Agent 懒加载和历史迁移。
  Added branch-scoped paginated history and JSONL storage with layered loading, lazy subagent details, and migration support.
• 新增 Agent Instructions、Tool Catalog、Workflow 和 Refinery 工作区，并补齐中英文界面。
  Added Agent Instructions, Tool Catalog, Workflow, and Refinery workspaces with English and Simplified Chinese UI coverage.

Changed / 变更

• Scope 改为软提示：白名单和黑名单用于上下文与优先级提示，不再单独阻断任务聚焦或执行；授权声明由应用统一注入。
  Scope is now advisory: whitelist and blacklist modes provide context and prioritization without independently blocking task focus or execution; authorization is declared by the application.
• 用户自定义 Restrictions、Skills、Tools 和 Workflows 存放在 Hexestra 用户目录，不再混入内置资源；内置 ATT&CK 目录固定为 Enterprise v19.1。
  User Restrictions, Skills, Tools, and Workflows live in the Hexestra user directory instead of bundled resources; the built-in ATT&CK catalog is pinned to Enterprise v19.1.
• Claude Agent SDK 适配为持续 Streaming Input，会话内复用 Query；Shell auto 模式在输出空闲后可正常结束命令。
  Adapted the Claude Agent SDK to persistent Streaming Input with per-conversation Query reuse; Shell auto mode now completes commands after output becomes idle.

Fixed / 修复

• 修复长对话首屏渲染卡顿、任务重复聚焦、已启动 Step 更新失败、Finding 详情不刷新和提炼调试信息不可见等问题。
  Fixed long-conversation first-render stalls, repeated task focusing, updates to started Steps, stale Finding details, and missing refinery diagnostics.

Hexestra v0.4.0

Added / 新增

• 新增项目级 Mihomo 多跳出口代理、safeStorage 加密节点库、可视化链编辑器、状态栏与 Agent 编排工具；v1.19.29 为已测试推荐版本而非版本限制，代理故障时保持 fail-closed。
  Added project-scoped Mihomo multi-hop egress, a safeStorage-encrypted node vault, visual chain editor, status bar, and Agent orchestration tools. v1.19.29 is tested and recommended rather than enforced, while proxy failures remain fail-closed.
• Browser、Traffic/Replay、SSH、WebShell 与 Local/WSL Terminal 现可共享项目 mixed 入口；不修改系统代理或启用 TUN。
  Browser, Traffic/Replay, SSH, WebShell, and Local/WSL Terminal can now share the project mixed ingress without changing the system proxy or enabling TUN.
• 新增 Claude 命令目录、MCP 服务健康状态，以及 Agent 命令与附件能力协商。
  Added the Claude command catalog, MCP server health status, and capability negotiation for Agent commands and attachments.
• 新增细粒度资产图，将目标、服务、端点与发现关系同步到 NetMap，并支持 Agent 查询项目资产关系。
  Added a fine-grained asset graph that synchronizes targets, services, endpoints, and discovery relationships into NetMap and exposes project asset relationships to the Agent.

Changed / 变更

• 将节点库与多跳链整合为可调比例工作台；节点库可折叠为状态窄栏，并支持鼠标拖动和键盘调整宽度。
  Consolidated the node library and multi-hop chain editor into a resizable workspace with a collapsible status rail and pointer or keyboard resizing.

Fixed / 修复

• 修复窄聊天面板中用户消息不换行，以及生成回复时视图不持续跟随最新内容的问题。
  Fixed user messages overflowing narrow chat panels and live responses failing to remain in view while being generated.

This release is not code-signed or notarized. After copying Hexestra.app to /Applications, use System Settings > Privacy & Security > Open Anyway if macOS blocks the first launch and you trust the downloaded file. As a Terminal alternative, run:
本次发布尚未完成代码签名和公证。将 Hexestra.app 拷贝到 /Applications 后，如果 macOS 阻止首次启动，请在确认下载文件来源可信后，打开“系统设置 → 隐私与安全性 → 仍要打开”。也可以在终端执行：

xattr -dr com.apple.quarantine "/Applications/Hexestra.app"

Hexestra v0.3.0

Added / 新增

• 新增基于适配器的 WebShell Profile 管理，首批支持 Generic 与 AntSword 兼容的 PHP 适配器。
  Added adapter-based WebShell profile management with Generic and AntSword-compatible PHP adapters.
• 新增操作系统与 PHP Eval 能力自动探测，以及持久化连接健康状态、验证历史、延迟、失败详情和远程系统信息。
  Added automatic OS and PHP-eval capability detection, persisted connection health, verification history, latency, failure details, and remote system information.
• 新增 Agent 创建、检查和显式验证 WebShell Profile 的工具，并保留适配器专属配置。
  Added Agent tools for creating, inspecting, and explicitly verifying WebShell profiles while preserving adapter-specific configuration.

Changed / 变更

• 重构 Shell Manager Profile 卡片，将 Verify、Edit 和 Delete 操作紧凑地右侧对齐，并改进健康诊断展示。
  Reworked Shell Manager profile cards with compact, right-aligned Verify, Edit, and Delete actions plus clearer health diagnostics.
• 使用每次请求新生成的不透明 128 位令牌和数字协议字段替代可识别产品的连接器标记。
  Replaced product-identifying connector markers with fresh opaque 128-bit request tokens and numeric protocol fields.
• 柔化浅色主题配色，并更新公开 README 的截图和文案。
  Softened the light theme palette and refreshed public README screenshots and copy.

Fixed / 修复

• 通过仅针对 xterm 显示层规范化换行符，修复中央托管 WebShell 终端的多行输出阶梯错位。
  Fixed stair-step multiline output in the central managed WebShell terminal by normalizing display-only line endings for xterm.
• 修复受限布局中的 Tab 与聊天区域滚动行为。
  Fixed tab and chat scrolling behavior in constrained layouts.
• 修复 Profile 编辑和验证失败时持久化健康状态及临时会话未一致清理的问题。
  Fixed profile edits and failed verifications so durable health and temporary sessions are cleaned up consistently.

This release is not code-signed or notarized. After copying Hexestra.app to /Applications, use System Settings > Privacy & Security > Open Anyway if macOS blocks the first launch and you trust the downloaded file. As a Terminal alternative, run:
本次发布尚未完成代码签名和公证。将 Hexestra.app 拷贝到 /Applications 后，如果 macOS 阻止首次启动，请在确认下载文件来源可信后，打开“系统设置 → 隐私与安全性 → 仍要打开”。也可以在终端执行：

xattr -dr com.apple.quarantine "/Applications/Hexestra.app"

Hexestra v0.2.1

Changed

• 全面重构 UI 色彩方案：深色石墨面板、蓝色操作区、青色拓扑高亮，四面板统一语义化严重等级配色。
• Rebuilt the entire UI color scheme: dark graphite surfaces, blue actions, teal
  topology accents, and consistent semantic severity colors across all panels.

Fixed

• 修复内嵌 mitmproxy 运行时在 macOS 签名和 Windows 安装器签名后损坏的问题，Traffic Capture
  现在在所有平台上开箱即用。
• Bundled mitmproxy runtime now survives macOS code-signing and Windows
  installer signing so Traffic Capture works out of the box on all platforms.

This release is not code-signed or notarized. After copying Hexestra.app to
/Applications, use System Settings > Privacy & Security > Open Anyway if macOS
blocks the first launch and you trust the downloaded file. As a Terminal alternative,
run:

xattr -dr com.apple.quarantine "/Applications/Hexestra.app"

本次发布尚未完成代码签名和公证。将 Hexestra.app 拷贝到 /Applications 后，如果
macOS 阻止首次启动，请在确认下载文件来源可信后，打开“系统设置 → 隐私与安全性 →
仍要打开”。也可以在终端执行：

xattr -dr com.apple.quarantine "/Applications/Hexestra.app"

Hexestra v0.2.0

Hexestra v0.2.0

Hexestra 0.2.0 improves the shared operating surface between security operators and
the Claude Agent. This release focuses on theme-aware branding, native Subagent
visibility, browser-assisted workflows, asset operations, and release reliability.

Highlights

System-aware appearance and Hexestra branding

• Choose System, Dark, or Light under Settings > General > Interface.
• Theme changes apply immediately across Electron surfaces, terminal sessions,
  Monaco editors, NetMap, dialogs, scrollbars, forms, and status areas.
• Existing terminal and editor sessions retain their content while appearance changes.
• The welcome screen, title bar, desktop package, and README now use the Hexestra
  wordmark and standalone application mark.

Native Subagent visibility

• Claude Subagent lifecycle events are projected into the conversation timeline in
  real time.
• Subagent lineage, status, permission provenance, activity, and results persist with
  project state.
• A dedicated detail view makes concurrent delegated work inspectable without losing
  the parent conversation context.

Browser and asset workflows

• The Agent can use the integrated browser automation runtime through explicit tool
  contracts and policy checks.
• Asset inventory actions provide direct NetMap, browser, copy, and scoped Agent
  rescan operations.
• NetMap, asset details, status information, and chat controls received focused visual
  and interaction refinements.

Zero-install Traffic Capture

• Packaged builds include the official platform-specific mitmdump 12.2.3 runtime;
  no separate mitmproxy installation or PATH configuration is required.
• Runtime archives are downloaded from mitmproxy's official distribution host during
  packaging and verified against pinned SHA-256 digests from its Sigstore provenance.
• A manually selected executable remains available as an expert override. Hexestra
  continues to generate and pin a project-specific CA without silently changing the
  operating-system trust store.

Reliability and distribution

• CI now validates and packages the desktop application on Windows x64, Linux x64,
  macOS Intel, and macOS Apple Silicon runners.
• CI downloads contain final .exe, .AppImage, or .dmg distributables instead of
  unpacked application directories.
• Platform-specific node-pty prebuild validation fails early when a compatible native
  package is unavailable.
• Public-source auditing rejects local workflow state, credentials, captures, copied
  runtimes, generated Python caches, and unknown top-level entries.
• Electron, @xterm/addon-fit, and tailwind-merge were updated.

Compatibility

• Existing application settings migrate to the new theme-aware schema and default to
  following the operating-system appearance.
• Existing Hexestra projects and legacy migration identifiers remain supported.
• Claude Code remains user-installed. Packaged builds bundle the audited mitmdump
  runtime but no authentication data, shared CA private key, or other third-party
  runtime.

Upgrade

Pull the v0.2.0 source, install the locked Node.js 24 dependencies, and start the
desktop application:

npm ci
npm run electron:dev

Run the release quality gates before packaging:

npm run audit:public
npm run check

macOS first launch for this unsigned release

This release is not code-signed or notarized. After copying Hexestra.app to
/Applications, use System Settings > Privacy & Security > Open Anyway if macOS
blocks the first launch and you trust the downloaded file. As a Terminal alternative,
run:

xattr -dr com.apple.quarantine "/Applications/Hexestra.app"

Use Hexestra only on systems you own or are explicitly authorized to assess.

---

中文说明

Hexestra 0.2.0 改进了安全操作员与 Claude Agent 之间的共享工作界面。本版本重点
更新了双主题品牌、原生 Subagent 可见性、浏览器辅助工作流、资产操作和发布可靠性。

主要更新

跟随系统的外观与 Hexestra 品牌

• 可在 设置 → 常规 → 界面 中选择“系统”“深色”或“浅色”。
• 主题会立即同步到 Electron 界面、终端、Monaco 编辑器、NetMap、弹窗、滚动条、表单
  和状态区域。
• 切换外观时，已有终端和编辑器会话会保留内容。
• 欢迎页、标题栏、桌面应用图标和 README 现在使用 Hexestra 完整字标及独立应用图标。

原生 Subagent 可见性

• Claude Subagent 生命周期事件会实时显示在对话时间线中。
• Subagent 的血缘关系、状态、权限来源、活动和结果会随项目状态持久化。
• 独立详情页可以查看并行委派工作，同时保留父级对话上下文。

浏览器与资产工作流

• Agent 可以通过明确的工具契约和策略检查使用集成浏览器自动化运行时。
• 资产清单操作支持直接在 NetMap 中查看、打开浏览器目标、复制地址或 JSON，以及请求
  有范围限制的 Agent 重新扫描。
• NetMap、资产详情、状态信息和聊天控制经过了针对性的视觉与交互优化。

零安装流量捕获

• 打包版内置官方的平台对应 mitmdump 12.2.3 运行时，无需单独安装 mitmproxy 或配置
  PATH。
• 打包时从 mitmproxy 官方分发地址下载运行时，并使用其 Sigstore 来源证明中的固定
  SHA-256 摘要进行校验。
• 仍可手动选择可执行文件作为专家覆盖选项。Hexestra 会继续生成并固定项目专属 CA，
  不会静默修改操作系统信任库。

可靠性与分发

• CI 现在会在 Windows x64、Linux x64、macOS Intel 和 macOS Apple Silicon runner 上
  验证并打包桌面应用。
• CI 下载内容只包含最终的 .exe、.AppImage 或 .dmg 分发文件，不再包含未打包的
  应用目录。
• 平台专用 node-pty 预编译包验证会在缺少兼容原生包时尽早失败。
• 公共源码审计会拒绝本地工作流状态、凭据、抓包文件、复制的运行时、生成的 Python
  缓存和未知顶层条目。
• Electron、@xterm/addon-fit 和 tailwind-merge 已升级。

兼容性

• 现有应用设置会迁移到新的主题配置，并默认跟随操作系统外观。
• 现有 Hexestra 项目和旧版迁移标识继续受支持。
• Claude Code 仍需用户自行安装。打包版包含经过审计的 mitmdump 运行时，但不包含认证
  数据、共享 CA 私钥或其他第三方运行时。

升级

拉取 v0.2.0 源码，安装锁定的 Node.js 24 依赖，然后启动桌面应用：

npm ci
npm run electron:dev

打包前运行发布质量检查：

npm run audit:public
npm run check

本次未签名版本的 macOS 首次启动

本次发布尚未完成代码签名和公证。将 Hexestra.app 拷贝到 /Applications 后，如果
macOS 阻止首次启动，请在确认下载文件来源可信后，打开“系统设置 → 隐私与安全性 →
仍要打开”。也可以在终端执行：

xattr -dr com.apple.quarantine "/Applications/Hexestra.app"

请仅在你拥有或已获得明确授权的系统上使用 Hexestra。

Hexestra 0.1.0

Hexestra 0.1.0 is the initial public source release of an AI-native penetration testing IDE where human operators and AI share the same operational surface.

Highlights

• Integrated browser, terminals, traffic capture, shells, NetMap, tasks, evidence, findings, and reports.
• Controlled autonomy with operator inspection, guidance, approval, interruption, and takeover.
• Reuse familiar tools including Claude Code, Burp Suite, PowerShell, WSL, and SSH.
• Source-run support for Windows x64, Ubuntu 24.04 x64, macOS Intel, and macOS Apple Silicon.
• mitmproxy 12.2.3 is installed separately and discovered automatically.

This is a source-only pre-release. Signed installers, automatic mitmproxy installation, and bundled mitmproxy binaries are not included. CI passed on all four supported source-run platforms.