oh-my-opencode 5.0.0-beta.9 → 5.0.0-beta.90

package/README.zh-cn.md

@@ -1,27 +1,19 @@
 > [!NOTE]
-> **OmO for Codex 已发布。试试 LazyCodex**
->
-> 因为太爱 Anthropic 模型而被封禁之后，我们决定站到 Codex 这一边。
-> 如果你是 OmO 粉丝，但觉得配置太麻烦，请使用 LazyCodex。OmO for Codex 已经发布:
-> ```bash
-> npx lazycodex-ai install
-> ```
-> 了解更多请访问 [lazycodex.ai](https://lazycodex.ai)。
+> **OmO 测试版: OmO ❤️ Pi**
+> 使用 `bun add -g omo-ai@beta` 试用。记忆系统、CodeMode、Anthropic 订阅,全部支持。
+> [![OmO Herdr DAG - live OmO workflow DAGs in a Herdr side pane](./.github/assets/omo-herdr-dag.png)](https://github.com/jc01rho/omo-herdr-dag)
+> *只需在提示词中输入 "mass ulw" - 你就是图工程的大师。多模型 ultracode,搭配更出色的记忆系统。(右侧面板是 [omo-herdr-dag](https://github.com/jc01rho/omo-herdr-dag))*
+
+
+> **Sponsors**
+> 以下是我们的赞助商,支持着这个个人副业项目的持续开发。
+> | [<img alt="OpenGateway" src="./.github/assets/opengateway-logo.svg" width="156px" />](https://opengateway.ai/) | **[OpenGateway](https://opengateway.ai/)** 赞助了 OmO。它是一个兼容 OpenAI 的网关,通过一个 API 即可使用多家模型提供商。感谢对开源的支持。 |
+> | :-----| :----- |
 
 > [!NOTE]
 > **正在进行多 Harness 代理操作系统重构**
 >
-> 我们正在重构代码库以支持多个代理 harness（OpenCode、Codex、Pi 等）。如果您有兴趣贡献，请先阅读 [ROADMAP](./ROADMAP.md)。与 ROADMAP 相关的 PR 请使用 `ROADMAP` 标签。
-
-> [!TIP]
-> **Building in Public**
->
-> 维护者正在使用 Jobdori 实时开发和维护 oh-my-openagent。Jobdori 是基于 OpenClaw 深度定制的 AI 助手。
-> 每个功能开发、每次修复、每次 Issue 分类，都在 Discord 上实时进行。
->
-> [![Building in Public](./.github/assets/building-in-public.png)](https://discord.gg/PUwSMR9XNk)
->
-> [**→ 在 #building-in-public 频道中查看**](https://discord.gg/PUwSMR9XNk)
+> 我们正在重构代码库以支持多个代理 harness（OpenCode、Codex、Pi、Claude Code 等）。如果您有兴趣贡献，请先阅读 [ROADMAP](./ROADMAP.md)。与 ROADMAP 相关的 PR 请使用 `ROADMAP` 标签。
 
 
 > [!NOTE]
@@ -67,7 +59,7 @@
 [![GitHub Issues](https://img.shields.io/github/issues/code-yeongyu/oh-my-openagent?color=ff80eb&labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-openagent/issues)
 [![License](https://img.shields.io/badge/license-SUL--1.0-white?labelColor=black&style=flat-square)](https://github.com/code-yeongyu/oh-my-openagent/blob/dev/LICENSE.md)
 [![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/code-yeongyu/oh-my-openagent)
-[![Docs](https://img.shields.io/badge/docs-omo.vibetip.help-369eff?labelColor=black&logo=readthedocs&logoColor=white&style=flat-square)](https://omo.vibetip.help/docs)
+[![Docs](https://img.shields.io/badge/docs-omo.dev-369eff?labelColor=black&logo=readthedocs&logoColor=white&style=flat-square)](https://omo.dev/docs)
 
 [English](README.md) | [한국어](README.ko.md) | [日本語](README.ja.md) | [简体中文](README.zh-cn.md)
 
@@ -126,14 +118,16 @@ https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/refs/heads/dev/do
 获取安装指南并照做：
 
 ```bash
-curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/refs/heads/dev/docs/guide/installation.md
+curl -fsSL https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/refs/heads/dev/docs/guide/installation.md
 ```
 
-**注意**：已发布的 npm 包名和 CLI 二进制名仍然是 `oh-my-opencode`（过渡期间同时以 `oh-my-openagent` 的名字双重发布）。在 `opencode.json` 中，兼容性层现在优先使用插件入口 `oh-my-openagent`，而旧的 `oh-my-opencode` 条目仍会以警告的形式加载。插件配置文件通常仍使用 `oh-my-opencode.json` 或 `oh-my-opencode.jsonc`，在过渡期间新旧两种文件名都会被识别。
+**注意**：已发布的 npm 包名和 CLI 二进制名仍然是 `oh-my-opencode`（过渡期间同时以 `oh-my-openagent` 的名字双重发布）。在 `opencode.json` 中，兼容性层现在优先使用插件入口 `oh-my-openagent`，而旧的 `oh-my-opencode` 条目仍会以警告的形式加载。运行时配置是 `~/.omo/omo.jsonc`，加上向上逐级查找的项目 `.omo/omo.jsonc`。旧版 `oh-my-openagent.json[c]` / `oh-my-opencode.json[c]` 文件只会被迁移引擎导入一次，之后不再被读取。
+
+推荐的 `bunx`/`npx` 调用方式是 `oh-my-openagent install`（或原来的 `oh-my-opencode install`）。安装完成后，短命令是 `omo-agent-toolkit`。这些包在本次大版本中移除了 `omo` bin，这个名字现在归 OmO Native 所有，通过 `bun add -g omo-ai@beta` 安装（仅 beta 通道）。**不要**使用 `bunx omo` 或 `npx omo`：npm 上的 `omo` 是另一位作者的无关包，这两条命令会解析到它。OmO Native 的包名是 `omo-ai`。`lazycodex-ai` 是一个单一用途的 Node/npm 安装器包：`npx lazycodex-ai install` 会直接进入 Codex Light 安装器。它不是 Codex marketplace 的名字（marketplace 仓库是 `code-yeongyu/lazycodex`）。Codex 看到的是 marketplace `sisyphuslabs` 和插件 `omo`，以 `omo@sisyphuslabs` 启用。
 
 匿名遥测默认开启，用于统计活跃安装数(DAU/WAU/MAU)。每台机器每个 UTC 日最多发送一次事件,使用哈希化的安装标识符,绝不会使用原始主机名,且不会创建 PostHog person profile。可通过 `OMO_SEND_ANONYMOUS_TELEMETRY=0` 或 `OMO_DISABLE_POSTHOG=1` 禁用。详见 [隐私政策](docs/legal/privacy-policy.md) 和 [服务条款](docs/legal/terms-of-service.md)。
 
-**Ultimate 与 Light:** oh-my-openagent 以同一产品的两个版本发布。**Ultimate 版本**（`bunx oh-my-openagent install` 或 `--platform=opencode`，默认值）在 OpenCode 上提供完整功能 —— 11 个智能体、54+ 个生命周期钩子、Team Mode、所有 MCP、所有斜杠命令、IntentGate 模式。**Light 版本**（`npx lazycodex-ai install` 或 `bunx oh-my-openagent install --platform=codex`）提供能够干净地移植到 OpenAI Codex CLI 插件系统的 8 个组件（`rules`、`comment-checker`、`git-bash`、`lsp`、`ultrawork`、`ulw-loop`、`start-work-continuation`、`telemetry`）。要同时安装两个版本，使用 `--platform=both`。Codex 专用遥测可通过 `OMO_CODEX_DISABLE_POSTHOG=1` 或 `OMO_CODEX_SEND_ANONYMOUS_TELEMETRY=0` 禁用。
+**Ultimate、Light 与 OmO Native:** oh-my-openagent 以同一产品的三个版本发布：两个插件加载到你已经在用的宿主里，另有一个独立版本。**Ultimate 版本**（`bunx oh-my-openagent install` 或 `--platform=opencode`，默认值）在 OpenCode 上提供完整功能 —— 11 个智能体、54+ 个生命周期钩子、Team Mode、4 个内置 MCP（websearch、context7、grep_app、lsp）、所有斜杠命令、IntentGate 模式。**Light 版本**（`npx lazycodex-ai install`）提供能够干净地移植到 OpenAI Codex CLI 插件系统的 核心组件（`rules`、`comment-checker`、`git-bash`、`lsp`、`ultrawork`、`ulw-loop`、`ulw-execute-continuation`、`telemetry`），外加 `teammode` 和配套组件（`bootstrap`、`lcx` 等），并会把 Codex 智能体 TOML 安装到 `~/.codex/agents/`。要同时安装这两个插件版本，使用 `--platform=both`。**OmO Native（独立，beta）** 是内建 OMO 扩展的 `omo` 命令：`bun add -g omo-ai@beta`，然后运行 `omo`。它从 `omo-ai@beta` 安装，不加载进 OpenCode 或 Codex；仅提供 beta 通道，直接 `bun add -g omo-ai` 会按设计失败。从 OpenCode 版本迁移请参阅 [Migrating from OpenCode](docs/guide/migrating-from-opencode.md)。Codex 专用遥测可通过 `OMO_CODEX_DISABLE_POSTHOG=1` 或 `OMO_CODEX_SEND_ANONYMOUS_TELEMETRY=0` 禁用。
 
 ---
 
@@ -166,59 +160,52 @@ Read this and tell me why it's not just another boilerplate: https://raw.githubu
 - [ChatGPT 订阅 ($20)](https://chatgpt.com/)
 - [Kimi Code 订阅 ($19)](https://www.kimi.com/code)
 - [GLM Coding 套餐 ($10)](https://z.ai/subscribe)
-- 如果你能使用按 token 计费的方式，用 Kimi 和 Gemini 模型花不了多少钱。
+- 如果你能使用按 token 计费的方式，用 Kimi 和 GLM 模型花不了多少钱。
 
 |       | 特性                                                            | Editions | 功能说明                                                                                                                                                                        |
 | :---: | :-------------------------------------------------------------- | :------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
-|   🤖   | **自律军团 (Discipline Agents)**                                | Ultimate | Sisyphus 负责调度 Hephaestus、Oracle、Librarian 和 Explore。一支完整的 AI 开发团队并行工作。                                                                                    |
-|   🧩   | **Codex CLI Light Edition**                                      | Light    | 在 OpenAI Codex CLI 中运行的 omo 的 8 个可移植组件 (rules, comment-checker, git-bash, LSP, ultrawork, ulw-loop, start-work continuation, telemetry)。安装: `npx lazycodex-ai install`。                    |
-|   👥   | **Team Mode** (v4.0, 选择性启用)                                 | Ultimate | 领导 Agent + 最多 8 个并行成员，实时 tmux 可视化，专用 `team_*` 工具家族。驱动 `hyperplan`(5 个敌对评论者) 和 `security-research`(3 个猎手 + 2 个 PoC 工程师)。[文档 →](docs/guide/team-mode.md) |
+|   🤖   | **自律军团 (Discipline Agents)**                                | Ultimate | 主 Agent 负责调度架构师咨询、Librarian、Explore 和类别工作者。一支完整的 AI 开发团队并行工作。                                                                                   |
+|   🧩   | **Codex CLI Light Edition**                                      | Light    | 在 OpenAI Codex CLI 中运行的 omo 的可移植组件 (rules, comment-checker, git-bash, LSP, ultrawork, ulw-loop, ulw-execute continuation, telemetry, teammode 等)。安装: `npx lazycodex-ai install`。                    |
+|   👥   | **Team Mode** (选择性启用)                                       | Ultimate | 领导 Agent + 最多 8 个并行成员，实时 tmux 可视化，专用 `team_*` 工具家族。驱动 `hyperplan`(5 个敌对评论者) 和 `security-research`(3 个猎手 + 2 个 PoC 工程师)。[文档 →](docs/guide/team-mode.md) |
 |   ⚡   | **`ultrawork` / `ulw`**                                         | Both     | 一键触发，所有智能体(Ultimate)或 Codex `ultrawork` 组件(Light)出动。任务完成前绝不罢休。                                                                                          |
-|   🚪   | **[IntentGate 意图门](https://factory.ai/news/terminal-bench)** | Ultimate | 真正行动前，先分析用户的真实意图。触发 `search` / `analyze` / `team` / `hyperplan`。(Light 仅 hook `ulw` / `ultrawork`。)                                                       |
-|   🔗   | **基于哈希的编辑工具**                                          | Ultimate | 每次修改都通过 `LINE#ID` 内容哈希验证、0% 错误修改。灵感来自 [oh-my-pi](https://github.com/can1357/oh-my-pi)。[The Harness Problem →](https://blog.can.ac/2026/02/12/the-harness-problem/) (Codex 使用其原生 `apply_patch`。) |
+|   🚪   | **[IntentGate 意图门](https://factory.ai/news/terminal-bench)** | Ultimate | 检测 `ultrawork`/`ulw`、`team`、`hyperplan` 关键词(不做语义层面的意图分类)。(Light 仅识别 `ulw` / `ultrawork` 关键词。)                                                       |
+|   🔗   | **基于哈希的编辑工具**                                          | Ultimate | Hashline (`LINE#ID`) 编辑/读取标记。需手动启用：在 `~/.omo/omo.jsonc` 中设置 `hashline_edit: true`、0% 错误修改。灵感来自 [oh-my-pi](https://github.com/can1357/oh-my-pi)。[The Harness Problem →](https://blog.can.ac/2026/02/12/the-harness-problem/) (Codex 使用其原生 `apply_patch`。) |
 |   🛠️   | **LSP + AST-Grep**                                              | Both     | 工作区级别的重命名、构建前诊断、基于 AST 的重写。LSP 通过 MCP 提供，AST-Grep 通过共享 `ast-grep` skill 和 `sg` 提供。                                   |
 |   🧠   | **后台智能体**                                                  | Ultimate | 同时发射 5+ 个专家并行工作。保持上下文干净，随时获取成果。                                                                                                                      |
-|   📚   | **内置 MCP**                                                    | Both     | Ultimate 运行时注入 Exa（网络搜索）、Context7（官方文档）、Grep.app（GitHub 源码搜索）。Light 通过 plugin-scoped MCP 提供 `grep_app`、`context7`、`git_bash`、`lsp`。                                                                              |
-|   🔁   | **Ralph Loop / `/ulw-loop`**                                    | Ultimate | 自我引用闭环。达不到 100% 完成度绝不停止。                                                                                                                                      |
+|   📚   | **内置 MCP**                                                    | Both     | Ultimate 运行时注入 websearch（Exa）、context7（官方文档）、grep_app（GitHub 源码搜索）、lsp。Light 通过 plugin-scoped MCP 提供 `grep_app`、`context7`、`git_bash`、`lsp`。                                                                              |
+|   🔁   | **Goal / `/goal`**                                              | Ultimate | `/goal` 为会话设定持久目标。idle continuation 仅在 `goal.enabled` 为 true 时运行（默认 false）。                                                                                          |
 |   ✅   | **Todo 强制执行** (Boulder)                                     | Ultimate | Agent 想要摸鱼？系统直接揪着领子拽回来。你的任务，必须完成。                                                                                                                    |
 |   💬   | **注释审查员**                                                  | Both     | 剔除带有浓烈 AI 味的冗余注释。同一个 `@code-yeongyu/comment-checker` 二进制在两个版本中运行。                                                                                    |
 |   📜   | **Rules Injection**                                             | Both     | `AGENTS.md` / `CLAUDE.md` / `.omo/rules/**` 的分层上下文注入。Ultimate 中为 hook，Light 中为 `rules` 组件。                                                                       |
-|   🧬   | **Ulw Loop**                                                   | Light    | 基于 `.omo/ulw-loop/` 证据审计的持久化多目标编排。目前仅 Codex 可用; OpenCode 侧的移植在路线图上。                                                                              |
+|   🧬   | **Ulw Loop**                                                   | Light    | 面向持久化多目标编排（含证据审计）的 Codex 组件 + CLI。在 OpenCode 上请改用 **Goal / `/goal`**。                                                                              |
 |   🖥️   | **Tmux 集成**                                                   | Ultimate | 完整的交互式终端支持。跑 REPL、用调试器、用 TUI 工具，全都在实时会话中完成。                                                                                                    |
 |   🔌   | **Claude Code 兼容**                                            | Ultimate | 你现有的 Hooks、命令、技能、MCP 和插件？全都能无缝迁移过来。(Codex 拥有其自己的原生插件系统。)                                                                                   |
 |   🎯   | **技能内嵌 MCP**                                                | Ultimate | 技能自带其所需的 MCP 服务器。按需开启，不会撑爆你的上下文窗口。                                                                                                                 |
-|   📋   | **Prometheus 规划师**                                           | Ultimate | 动手写代码前，先通过访谈模式做好战略规划。                                                                                                                                      |
+|   📋   | **Ultrawork Planner 规划师**                                    | Ultimate | 动手写代码前，先通过访谈模式做好战略规划。                                                                                                                                      |
 |   🔍   | **`/init-deep`**                                                | Ultimate | 在整个项目目录层级中自动生成 `AGENTS.md`。不仅省 Token，还能大幅提升 Agent 理解力。                                                                                             |
 
-> **Editions 图例。** **Ultimate** = 仅 OpenCode (`bunx oh-my-openagent install`)。**Light** = 仅 Codex CLI (`bunx oh-my-openagent install --platform=codex`)。**Both** = 两个版本均提供 (内部实现可能略有不同)。
+> **Editions 图例。** **Ultimate** = 仅 OpenCode (`bunx oh-my-openagent install`)。**Light** = 仅 Codex CLI (`npx lazycodex-ai install`)。**Both** = 两个版本均提供 (内部实现可能略有不同)。
 
 ### 自律军团 (Discipline Agents)
 
-<table><tr>
-<td align="center"><img src=".github/assets/sisyphus.png" height="300" /></td>
-<td align="center"><img src=".github/assets/hephaestus.png" height="300" /></td>
-</tr></table>
-
-**Sisyphus** (`claude-opus-5` / **`kimi-k3`** / **`glm-5`**) 是你的主指挥官。他负责制定计划、分配任务给专家团队，并以极其激进的并行策略推动任务直至完成。他从不半途而废。Claude Opus 5 和 Kimi K3 是推荐默认值。
-
-**Hephaestus** (仅通过 OpenAI、GitHub Copilot、Vercel 或 OpenCode 使用 medium effort 的 `gpt-5.6-sol`) 是你的自主深度工作者。你只需要给他目标，不要给他具体做法。他会自动探索代码库模式，从头到尾独立执行任务，绝不会中途要你当保姆。*名副其实的正牌工匠。*
+**主 Agent** 运行在你的会话模型上，是你的主指挥官。他负责制定计划、分配任务给专家团队，并以极其激进的并行策略推动任务直至完成。他从不半途而废。你只需要给他目标，不要给他具体做法。他会自动探索代码库模式，从头到尾独立执行任务，绝不会中途要你当保姆。*名副其实的正牌工匠。*
 
-**Prometheus** (`claude-fable-5` / **`kimi-k3`**) 是你的战略规划师。他通过访谈模式，在动一行代码之前，先通过提问确定范围并构建详尽的执行计划。
+**Ultrawork Planner** (`/ulw-plan`) 是你的战略规划师。他通过访谈模式，在动一行代码之前，先通过提问确定范围并构建详尽的执行计划。随后 `/ulw-execute` 在同一会话中执行该计划。
 
-每一个 Agent 都针对其底层模型的特点进行了专门调优。你无需手动来回切换模型。[阅读背景设定了解更多 →](docs/guide/overview.md)
+每一个提示词预设都针对其底层模型的特点进行了专门调优。你无需手动来回切换模型。[阅读背景设定了解更多 →](docs/guide/overview.md)
 
-> Anthropic [因为我们屏蔽了 OpenCode](https://x.com/thdxr/status/2010149530486911014)。这就是为什么我们将 Hephaestus 命名为"正牌工匠 (The Legitimate Craftsman)"。这是一个故意的讽刺。
+> Anthropic [因为我们屏蔽了 OpenCode](https://x.com/thdxr/status/2010149530486911014)。这就是为什么我们将主 Agent 称为"正牌工匠 (The Legitimate Craftsman)"。这是一个故意的讽刺。
 >
 > 我们在 Opus 或 Kimi K3 上运行得最好，但仅仅使用 Kimi K3 + GPT-5.6 Sol 就足以碾压原版的 Claude Code。完全不需要配置。
 
-### Team Mode (v4.0)
+### Team Mode
 
 一个 Agent 已经够快。一支协调的团队是 *毁灭性* 的。
 
 **Team Mode** 把 oh-my-openagent 从「带子 Agent 的单个 Agent」升级为真正的多 Agent 系统。一个领导 Agent 协调一队按类别专业化的成员，全部 **并行** 运行，通过专用工具(`team_create`、`team_send_message`、`team_task_create`、`team_status`、…)进行通信。在 tmux 布局的 focus + grid 窗口中同时观察每个成员的工作。
 
 ```jsonc
-// .opencode/oh-my-openagent.jsonc
+// ~/.omo/omo.jsonc  （或项目的 .omo/omo.jsonc）
 {
   "team_mode": {
     "enabled": true,
@@ -237,7 +224,7 @@ Read this and tell me why it's not just another boilerplate: https://raw.githubu
 
 ### 智能体调度机制
 
-当 Sisyphus 把任务分配给子智能体时，他选择的不是具体的模型，而是 **类别 (Category)**。系统会自动将类别映射到最合适的模型：
+当主 Agent 把任务分配给子智能体时，他选择的不是具体的模型，而是 **类别 (Category)**。系统会自动将类别映射到最合适的模型：
 
 | 类别                 | 作用领域               |
 | :------------------- | :--------------------- |
@@ -246,7 +233,7 @@ Read this and tell me why it's not just another boilerplate: https://raw.githubu
 | `quick`              | 单文件修改、修错字     |
 | `ultrabrain`         | 复杂硬核逻辑、架构决策 |
 
-智能体只需要说明要做什么类型的工作，框架就会挑选出最合适的模型去干。`ultrabrain` 在 OpenAI 或 Vercel 可用时优先路由到 GPT-5.6 Sol xhigh，其次是 GPT-5.6 Sol xhigh。你完全不需要操心。
+智能体只需要说明要做什么类型的工作，框架就会挑选出最合适的模型去干。`ultrabrain` 路由到 GPT-6 Astra max，`deep` 路由到 GPT-6 Astra high（OpenAI / OpenAI Codex，其次 GitHub Copilot，再其次 OpenCode），不可用时回退到 GPT-5.6 Sol。你完全不需要操心。
 
 ### 完全兼容 Claude Code
 
@@ -277,7 +264,7 @@ Harness 问题是真的。绝大多数所谓的 Agent 故障，其实并不是
 >
 > <br/>- [Can Bölük, The Harness Problem](https://blog.can.ac/2026/02/12/the-harness-problem/)
 
-受 [oh-my-pi](https://github.com/can1357/oh-my-pi) 的启发，我们做出了 **Hashline**。Agent 读到的每一行代码，末尾都会打上一个强绑定的内容哈希值：
+受 [oh-my-pi](https://github.com/can1357/oh-my-pi) 的启发，我们做出了 **Hashline**。启用 `hashline_edit` 后，Agent 读到的每一行代码，末尾都会打上一个强绑定的内容哈希值：
 
 ```
 11#VK| function hello() {
@@ -287,8 +274,6 @@ Harness 问题是真的。绝大多数所谓的 Agent 故障，其实并不是
 
 Agent 发起修改时，必须通过这些标签引用目标行。如果在此期间文件发生过变化，哈希验证就会失败，从而在代码被污染前直接驳回。不再有缩进空格错乱，彻底告别改错行的惨剧。
 
-在 Grok Code Fast 1 上，仅仅因为更换了这套编辑工具，修改成功率就从 **6.7% 飙升至 68.3%**。
-
 ### 深度上下文初始化：`/init-deep`
 
 执行一次 `/init-deep`。它会为你生成一套树状的 `AGENTS.md`：
@@ -304,11 +289,11 @@ project/
 
 Agent 会自动顺藤摸瓜加载对应的 Context，免去了你所有的手动喂喂喂的麻烦。
 
-### 让 Agent 动手前先过脑子：Prometheus
+### 让 Agent 动手前先过脑子：Ultrawork Planner
 
 碰到了硬骨头？千万不要扔个 Prompt 就双手合十祈祷。
 
-输入 `/start-work`，召唤 Prometheus 出场。**他会像一个真实的主管那样去采访你**，主动深挖需求、指出模糊地带，并在改动哪怕一行代码之前产出经过严密论证的计划。你的 Agent 终于知道了自己在干嘛。
+Ultrawork Planner (`/ulw-plan`) **会像一个真实的主管那样去采访你**，主动深挖需求、指出模糊地带，并在改动哪怕一行代码之前把经过严密论证的计划写入 `.omo/plans/`。随后 `/ulw-execute` 让主 Agent 在同一会话中执行该计划。你的 Agent 终于知道了自己在干嘛。
 
 ### 技能系统 (Skills)
 
@@ -346,12 +331,16 @@ Agent 会自动顺藤摸瓜加载对应的 Context，免去了你所有的手动
 2. **清除配置文件（可选）**
 
    ```bash
-   # 移除兼容期间被识别的插件配置文件
-   rm -f ~/.config/opencode/oh-my-openagent.jsonc ~/.config/opencode/oh-my-openagent.json \
-         ~/.config/opencode/oh-my-opencode.jsonc ~/.config/opencode/oh-my-opencode.json
+   # 移除运行时配置文件
+   rm -f ~/.omo/omo.jsonc ~/.omo/omo.json
 
    # 移除当前项目的配置（如果存在）
-   rm -f .opencode/oh-my-openagent.jsonc .opencode/oh-my-openagent.json \
+   rm -f .omo/omo.jsonc .omo/omo.json
+
+   # 移除残留的旧版迁移备份（如有）
+   rm -f ~/.config/opencode/oh-my-openagent.jsonc ~/.config/opencode/oh-my-openagent.json \
+         ~/.config/opencode/oh-my-opencode.jsonc ~/.config/opencode/oh-my-opencode.json \
+         .opencode/oh-my-openagent.jsonc .opencode/oh-my-openagent.json \
          .opencode/oh-my-opencode.jsonc .opencode/oh-my-opencode.json
    ```
 
@@ -365,10 +354,12 @@ Agent 会自动顺藤摸瓜加载对应的 Context，免去了你所有的手动
 4. **移除 omo-codex（Codex CLI Light 版本）**
 
    ```bash
-   rm -rf ~/.codex/plugins/cache/sisyphuslabs
+   npx lazycodex-ai uninstall
+   # backward-compatible alias:
+   npx lazycodex-ai cleanup
    ```
 
-   然后打开 `~/.codex/config.toml`，删除 `[marketplaces.sisyphuslabs]`、`[plugins."omo@sisyphuslabs"]` 以及所有 `[hooks.state."omo@sisyphuslabs:..."]` 区块。
+   卸载命令会移除由它管理的 `sisyphuslabs` Codex 缓存和 marketplace 状态，在写入备份后从 `~/.codex/config.toml` 中剥离 `omo@sisyphuslabs` 插件和 hook-state 区块，并删除安装清单中列出的智能体 TOML 链接。如果某个项目里还残留着旧的项目级 Codex 插件状态，请在该项目下运行此命令，或传入 `--project <path>`；它会修复已知的项目级 `.codex/config.toml` 冲突，并报告项目级 `.codex` 产物，但不会删除项目自有文件。
 
 ## Features
 
@@ -377,15 +368,15 @@ Agent 会自动顺藤摸瓜加载对应的 Context，免去了你所有的手动
 完整内容请见 [Features Documentation](docs/reference/features.md)。
 
 **简要概览：**
-- **Agents**: Sisyphus（主 Agent）、Prometheus（规划师）、Oracle（架构/调试）、Librarian（文档/代码检索）、Explore（快速 grep）、Multimodal Looker
+- **Agents**: 主 Agent（调度器）、Ultrawork Planner（`/ulw-plan`）、架构师咨询（架构/调试）、Librarian（文档/代码检索）、Explore（快速 grep）、Multimodal Looker
 - **后台 Agents**: 像真正的开发团队那样并行跑多个 Agent
 - **LSP & AST 工具**: 重构、重命名、诊断、AST 感知的代码检索
-- **基于哈希的编辑工具**: `LINE#ID` 引用在应用每次修改前都会验证内容。外科手术级编辑，零陈旧行错误
+- **基于哈希的编辑工具**（通过 `hashline_edit: true` 手动启用）: `LINE#ID` 引用在应用每次修改前都会验证内容。外科手术级编辑，零陈旧行错误
 - **上下文注入**: 自动注入 AGENTS.md、README.md、条件规则
 - **Claude Code 兼容**: 完整的 Hook 系统、命令、技能、Agents、MCP
-- **内置 MCP**: websearch（Exa）、context7（文档）、grep_app（GitHub 检索） ——由插件在运行时注入，不会显示在 `opencode mcp list`  中(参见文档 [MCP docs](docs/reference/features.md#native-vs-plugin-injected-mcps))
+- **内置 MCP**: websearch（Exa）、context7（文档）、grep_app（GitHub 检索）、lsp ——由插件在运行时注入，不会显示在 `opencode mcp list`  中(参见文档 [MCP docs](docs/reference/features.md#native-vs-plugin-injected-mcps))
 - **会话工具**: 列出、读取、搜索、分析会话历史
-- **效率功能**: Ralph Loop、Todo Enforcer、Comment Checker、Think Mode 等
+- **效率功能**: Goal、Todo Enforcer、Comment Checker、Think Mode 等
 - **Doctor 命令**: 内置诊断（`bunx oh-my-opencode doctor`），验证插件注册、配置、模型和环境
 - **模型回退**: `fallback_models` 可以在同一数组中混合使用普通模型字符串和 per-fallback 对象配置
 - **文件提示词**: 通过 `file://` 在 Agent 配置中从文件加载提示词
@@ -399,17 +390,17 @@ Agent 会自动顺藤摸瓜加载对应的 Context，免去了你所有的手动
 详细内容见 [Configuration Documentation](docs/reference/configuration.md)。
 
 **简要概览：**
-- **配置文件位置**: 使用用户全局配置，同时会从当前目录向上查找 `.opencode/oh-my-openagent.json[c]`（查到 `$HOME` 为止），离当前目录越近的优先级越高。旧版 `oh-my-opencode.json[c]` 仍然有效。
+- **配置文件位置**: 用户 `~/.omo/omo.jsonc`，同时会从当前目录向上查找项目 `.omo/omo.jsonc`（查到 `$HOME` 为止），离当前目录越近的优先级越高。旧版 `oh-my-*` 文件只会被迁移一次到 `omo.jsonc`。
 - **JSONC 支持**: 支持注释和尾逗号
 - **Agents**: 可对任意 Agent 覆盖模型、temperature、prompts 和权限
 - **内置技能**: `playwright`（浏览器自动化）、`git-master`（原子提交）
-- **Sisyphus Agent**: 主调度器，搭配 Prometheus（规划师）和 Metis（计划顾问）
+- **主 Agent**: 调度器，搭配 Ultrawork Planner（`/ulw-plan`）和 Plan Consultant
 - **后台任务**: 按 provider/model 配置并发上限
-- **类别**: 按领域的任务委托（`visual`、`business-logic`、自定义）
+- **类别**: `visual-engineering`、`ultrabrain`、`deep`、`artistry`、`quick`、`unspecified-low`、`unspecified-high`、`writing`，外加自定义名称
 - **Hooks**: 54+ 内置生命周期 Hook（启用 Team Mode 时为 61 个），都可以通过 `disabled_hooks` 控制
-- **MCPs**: 内置 websearch（Exa）、context7（文档）、grep_app（GitHub 检索）——由插件在运行时注入，不会显示在 `opencode mcp list`  中
+- **MCPs**: 内置 websearch（Exa）、context7（文档）、grep_app（GitHub 检索）、lsp——由插件在运行时注入，不会显示在 `opencode mcp list`  中
 - **LSP**: 包括重构工具的完整 LSP 支持
-- **Experimental**: 激进截断、自动 resume 等
+- **Experimental**: 激进截断、抢占式上下文压缩等
 
 
 ## 闲聊环节 (Author's Note)

package/THIRD-PARTY-NOTICES.md

@@ -1,70 +1,24 @@
 # Third Party Notices
 
-This file enumerates third-party components redistributed by the root `oh-my-opencode` package surface from `package.json` `files[]`, root production dependencies, bundled MCP/runtime payloads, and the planned pinned CodeGraph provisioning payload for the CodeGraph OMO integration.
+This file enumerates third-party components redistributed by the root `oh-my-opencode` package surface from `package.json` `files[]`, root production dependencies, and bundled MCP/runtime payloads.
 
 Generator/checker note: update this file with the audited component list, then run `node scripts/check-third-party-notices.mjs`. Task 11 owns tarball ship verification.
 
-## Components
+## Downloaded binaries
 
-### @clack/core@1.3.0
-- License: MIT, from CodeGraph platform bundle `lib/node_modules/@clack/core/LICENSE`.
-- Copyright: Copyright (c) Nate Moore.
-- Upstream URL: https://github.com/bombshell-dev/clack
-- Where-bundled: CodeGraph platform bundle vendored dependency.
+### comment-checker v0.8.0
+- License: MIT, as declared by the upstream project.
+- Copyright: Yeongyu Kim and contributors.
+- Upstream URL: https://github.com/code-yeongyu/go-claude-code-comment-checker/releases
+- Distribution: the OpenCode hook downloads the host-platform release binary into its local cache on first use. The root package no longer installs the npm package or its all-platform `vendor/` payload. Separate Codex component notices remain in `packages/omo-codex/THIRD-PARTY-NOTICES.md`.
+
+## Components
 
 ### @clack/prompts@1.5.0
 - License: MIT, from `node_modules/@clack/prompts/LICENSE`.
 - Copyright: Copyright (c) Nate Moore.
 - Upstream URL: https://github.com/bombshell-dev/clack
-- Where-bundled: root production dependency; CodeGraph platform bundles also vendor `@clack/prompts@1.3.0`.
-
-### @code-yeongyu/comment-checker@0.8.0
-- License: MIT, from package metadata. The inspected package did not include a separate LICENSE file.
-- Copyright: Yeongyu Kim and contributors.
-- Upstream URL: https://github.com/code-yeongyu/go-claude-code-comment-checker
-- Where-bundled: root production dependency and vendored checker binary payload under `vendor/<platform>/`.
-
-### @colbymchenry/codegraph@1.5.0
-- License: MIT, from npm package metadata. The inspected tarball did not include a separate LICENSE file.
-- Copyright: CodeGraph contributors.
-- Upstream URL: https://www.npmjs.com/package/@colbymchenry/codegraph
-- Where-bundled: pinned CodeGraph provisioning payload for the CodeGraph OMO integration.
-
-### @colbymchenry/codegraph-darwin-arm64@1.5.0
-- License: MIT, from npm package metadata. The inspected tarball did not include a separate LICENSE file.
-- Copyright: CodeGraph contributors.
-- Upstream URL: https://www.npmjs.com/package/@colbymchenry/codegraph-darwin-arm64
-- Where-bundled: optional CodeGraph platform bundle for darwin arm64, including CodeGraph runtime files, Node.js runtime binary, and tree-sitter WASM assets.
-
-### @colbymchenry/codegraph-darwin-x64@1.5.0
-- License: MIT, from npm package metadata. The inspected tarball did not include a separate LICENSE file.
-- Copyright: CodeGraph contributors.
-- Upstream URL: https://www.npmjs.com/package/@colbymchenry/codegraph-darwin-x64
-- Where-bundled: optional CodeGraph platform bundle for darwin x64, including CodeGraph runtime files, Node.js runtime binary, and tree-sitter WASM assets.
-
-### @colbymchenry/codegraph-linux-arm64@1.5.0
-- License: MIT, from npm package metadata. The inspected tarball did not include a separate LICENSE file.
-- Copyright: CodeGraph contributors.
-- Upstream URL: https://www.npmjs.com/package/@colbymchenry/codegraph-linux-arm64
-- Where-bundled: optional CodeGraph platform bundle for linux arm64, including CodeGraph runtime files, Node.js runtime binary, and tree-sitter WASM assets.
-
-### @colbymchenry/codegraph-linux-x64@1.5.0
-- License: MIT, from npm package metadata. The inspected tarball did not include a separate LICENSE file.
-- Copyright: CodeGraph contributors.
-- Upstream URL: https://www.npmjs.com/package/@colbymchenry/codegraph-linux-x64
-- Where-bundled: optional CodeGraph platform bundle for linux x64, including CodeGraph runtime files, Node.js runtime binary, and tree-sitter WASM assets.
-
-### @colbymchenry/codegraph-win32-arm64@1.5.0
-- License: MIT, from npm package metadata. The inspected tarball did not include a separate LICENSE file.
-- Copyright: CodeGraph contributors.
-- Upstream URL: https://www.npmjs.com/package/@colbymchenry/codegraph-win32-arm64
-- Where-bundled: optional CodeGraph platform bundle for win32 arm64, including CodeGraph runtime files, Node.js runtime binary, and tree-sitter WASM assets.
-
-### @colbymchenry/codegraph-win32-x64@1.5.0
-- License: MIT, from npm package metadata. The inspected tarball did not include a separate LICENSE file.
-- Copyright: CodeGraph contributors.
-- Upstream URL: https://www.npmjs.com/package/@colbymchenry/codegraph-win32-x64
-- Where-bundled: optional CodeGraph platform bundle for win32 x64, including CodeGraph runtime files, Node.js runtime binary, and tree-sitter WASM assets.
+- Where-bundled: root production dependency.
 
 ### @modelcontextprotocol/sdk@1.29.0
 - License: MIT, from `node_modules/@modelcontextprotocol/sdk/LICENSE`.
@@ -84,17 +38,29 @@ Generator/checker note: update this file with the audited component list, then r
 - Upstream URL: https://www.npmjs.com/package/@opencode-ai/sdk
 - Where-bundled: root production dependency for OpenCode API access.
 
-### CodeGraph bundled Node.js runtime
-- License: Node.js MIT license plus bundled third-party licenses for runtime components such as OpenSSL and V8. The inspected CodeGraph platform tarballs include the Node.js binary but no separate Node LICENSE or NOTICE files; the Codex CodeGraph component therefore ships Node.js v24.16.0 license text at `packages/omo-codex/plugin/components/codegraph/NODE-RUNTIME-LICENSES.md`.
-- Copyright: Node.js contributors, OpenSSL Software Foundation, V8 authors, and other Node.js third-party contributors.
-- Upstream URL: https://nodejs.org/
-- Where-bundled: each `@colbymchenry/codegraph-*` platform bundle includes `node` or `node.exe`.
+### @opentui/core@0.5.8
+- License: MIT, from `node_modules/@opentui/core/LICENSE`.
+- Copyright: Copyright (c) 2025 opentui.
+- Upstream URL: https://www.npmjs.com/package/@opentui/core
+- Where-bundled: root production dependency.
+
+### @opentui/keymap@0.5.8
+- License: MIT, from `node_modules/@opentui/keymap/LICENSE`.
+- Copyright: Copyright (c) 2025 opentui.
+- Upstream URL: https://www.npmjs.com/package/@opentui/keymap
+- Where-bundled: root production dependency.
+
+### @opentui/solid@0.5.8
+- License: MIT, from `node_modules/@opentui/solid/LICENSE`.
+- Copyright: Copyright (c) 2025 opentui.
+- Upstream URL: https://www.npmjs.com/package/@opentui/solid
+- Where-bundled: root production dependency.
 
 ### commander@14.0.3
 - License: MIT, from `node_modules/commander/LICENSE`.
 - Copyright: Copyright (c) 2011 TJ Holowaychuk.
 - Upstream URL: https://github.com/tj/commander.js
-- Where-bundled: root production dependency; CodeGraph platform bundles also vendor commander.
+- Where-bundled: root production dependency.
 
 ### detect-libc@2.1.2
 - License: Apache-2.0, from `node_modules/detect-libc/LICENSE`.
@@ -108,30 +74,6 @@ Generator/checker note: update this file with the audited component list, then r
 - Upstream URL: https://github.com/kpdecker/jsdiff
 - Where-bundled: root production dependency used by diff rendering and patch helpers.
 
-### fast-string-truncated-width@3.0.3
-- License: MIT, from CodeGraph platform bundle `lib/node_modules/fast-string-truncated-width/license`.
-- Copyright: Fabio Spampinato and contributors.
-- Upstream URL: https://github.com/fabiospampinato/fast-string-truncated-width
-- Where-bundled: CodeGraph platform bundle vendored dependency.
-
-### fast-string-width@3.0.2
-- License: MIT, from CodeGraph platform bundle `lib/node_modules/fast-string-width/license`.
-- Copyright: Fabio Spampinato and contributors.
-- Upstream URL: https://github.com/fabiospampinato/fast-string-width
-- Where-bundled: CodeGraph platform bundle vendored dependency.
-
-### fast-wrap-ansi@0.2.0
-- License: MIT, from CodeGraph platform bundle `lib/node_modules/fast-wrap-ansi/LICENSE`.
-- Copyright: James Garbutt and contributors.
-- Upstream URL: https://github.com/43081j/fast-wrap-ansi
-- Where-bundled: CodeGraph platform bundle vendored dependency.
-
-### ignore@7.0.5
-- License: MIT, from CodeGraph platform bundle `lib/node_modules/ignore/LICENSE-MIT`.
-- Copyright: ignore contributors.
-- Upstream URL: https://github.com/kaelzhang/node-ignore
-- Where-bundled: CodeGraph platform bundle vendored dependency.
-
 ### js-yaml@4.2.0
 - License: MIT, from `node_modules/js-yaml/LICENSE`.
 - Copyright: Copyright (C) 2011-2015 Vitaly Puzrin.
@@ -142,7 +84,7 @@ Generator/checker note: update this file with the audited component list, then r
 - License: MIT, from `node_modules/jsonc-parser/LICENSE.md`.
 - Copyright: Copyright (c) Microsoft.
 - Upstream URL: https://github.com/microsoft/node-jsonc-parser
-- Where-bundled: root production dependency and CodeGraph platform bundle vendored dependency.
+- Where-bundled: root production dependency.
 
 ### pi-comment-checker
 - License: MIT, from `packages/omo-codex/plugin/components/comment-checker/LICENSE` and component NOTICE.
@@ -172,7 +114,7 @@ Generator/checker note: update this file with the audited component list, then r
 - License: MIT, from `node_modules/picomatch/LICENSE`.
 - Copyright: Copyright (c) 2017-present, Jon Schlinkert.
 - Upstream URL: https://github.com/micromatch/picomatch
-- Where-bundled: root production dependency; CodeGraph platform bundles also vendor `picomatch@4.0.3`.
+- Where-bundled: root production dependency.
 
 ### posthog-node@5.35.12
 - License: Apache-2.0 text in `node_modules/posthog-node/LICENSE`; package metadata declares MIT.
@@ -180,26 +122,14 @@ Generator/checker note: update this file with the audited component list, then r
 - Upstream URL: https://github.com/PostHog/posthog-js
 - Where-bundled: root production dependency for anonymous telemetry.
 
-### sisteransi@1.0.5
-- License: MIT, from CodeGraph platform bundle `lib/node_modules/sisteransi/license`.
-- Copyright: Terkel Gjervig and contributors.
-- Upstream URL: https://github.com/terkelg/sisteransi
-- Where-bundled: CodeGraph platform bundle vendored dependency.
-
-### tree-sitter-wasms@0.1.13
-- License: Unlicense, from CodeGraph platform bundle `lib/node_modules/tree-sitter-wasms/LICENSE`.
-- Copyright: Gregor, Menci, and grammar contributors.
-- Upstream URL: https://github.com/Gregoor/tree-sitter-wasms
-- Where-bundled: CodeGraph platform bundle vendored dependency and bundled language grammar WASM files.
-
 ### vscode-jsonrpc@8.2.1
 - License: MIT, from `node_modules/vscode-jsonrpc/License.txt`.
 - Copyright: Copyright (c) Microsoft Corporation.
 - Upstream URL: https://github.com/microsoft/vscode-languageserver-node
 - Where-bundled: root production dependency for LSP JSON-RPC transport.
 
-### web-tree-sitter@0.25.10
-- License: MIT, from CodeGraph platform bundle `lib/node_modules/web-tree-sitter/LICENSE`.
-- Copyright: Max Brunsfeld and tree-sitter contributors.
-- Upstream URL: https://github.com/tree-sitter/tree-sitter
-- Where-bundled: CodeGraph platform bundle vendored dependency and bundled WebAssembly tree-sitter runtime.
+### zod@4.4.3
+- License: MIT, from `node_modules/zod/LICENSE`.
+- Copyright: Copyright (c) 2025 Colin McDonnell.
+- Upstream URL: https://zod.dev
+- Where-bundled: root production dependency.

package/dist/agents/gpt-prompt-identity.d.ts

@@ -1,3 +1,3 @@
-export type GptPromptIdentityKey = "gpt-5.5" | "gpt-5.6-sol" | "gpt-family";
+export type GptPromptIdentityKey = "gpt-5.5" | "gpt-5.6-sol" | "gpt-6-astra" | "gpt-6-sol" | "gpt-family";
 export declare function getGptPromptIdentityKey(model?: string): GptPromptIdentityKey;
 export declare function getGptPromptIdentity(model?: string): string;

package/dist/agents/sisyphus/grok-4.d.ts

@@ -9,8 +9,9 @@
  *   production ready") is the single highest-leverage instruction;
  * - a written definition of done, because otherwise the model decides done
  *   for you;
- * - a narration split (quiet on small changes, narrate wide-radius work) that
- *   matches its information-dense communication style;
+ * - one outcome-first handoff block per phase change and nothing between
+ *   handoffs (user directive 2026-09-24), which fits its information-dense
+ *   communication style;
  * - an anti-repetition nudge, since it duplicates component code unless asked
  *   to break it up.
  */

package/dist/agents/sisyphus-runtime-prompt-reconciler.d.ts

@@ -2,6 +2,10 @@
  * Context captured at Sisyphus registration so the per-request system-transform
  * hook can rebuild the prompt for the model actually selected at runtime.
  *
+ * - `configuredModel` is the exact model `bakedPrompt` was built for. It is the
+ *   "last built model": the prompt is baked only at registration, and opencode
+ *   re-assembles the system array from the agent config on every request, so
+ *   nothing else can become the currently baked prompt mid-session.
  * - `bakedPrompt` is the exact prompt string registered (body + overrides + env),
  *   used to locate the entry to replace in the runtime system array.
  * - `rebuildPromptForModel` re-runs the same registration pipeline with a
@@ -16,16 +20,23 @@ export declare function setSisyphusRuntimePromptContext(ctx: SisyphusRuntimeProm
 export declare function clearSisyphusRuntimePromptContext(): void;
 /**
  * The Sisyphus prompt body is baked at registration from the *configured* model
- * in `.omo/omo.jsonc`. When the user switches to a different model family
- * in the TUI, the entire baked body is the wrong family for the runtime model
+ * in `.omo/omo.jsonc`. When the user switches to a different model in
+ * the TUI, the baked body may be wrong for the runtime model
  * (issue #5297/#5316): a GPT-configured agent run on a non-GPT model still
  * carries the whole GPT-5.5 body, not just one apply_patch line.
  *
  * The system-transform hook is the only per-request seam that knows the runtime
- * model, so rebuild the whole prompt for the runtime family and swap it in here
- * rather than patching individual family-specific lines (which can never convert
+ * model, so rebuild the whole prompt for the runtime model and swap it in here
+ * rather than patching individual model-specific lines (which can never convert
  * a GPT body into a non-GPT one).
  *
+ * The skip must key on the exact model, not the broad prompt family: the
+ * `fallback` family is not prompt-uniform (Gemini fallback overrides, GPT
+ * identity text, claude/non-claude sections), so a same-family switch such as
+ * Gemini -> MiniMax-M3 or DeepSeek -> MiniMax-M3 can still leave the previous
+ * model's body in place (issue #6966). Genuine no-op switches (models whose
+ * rebuilt prompt is byte-identical to the baked one) are suppressed below.
+ *
  * Returns true if a swap was performed.
  */
 export declare function reconcileSisyphusRuntimePrompt(system: string[], runtimeModel: string | undefined): boolean;

package/dist/agents/types.d.ts

@@ -1,6 +1,6 @@
 import type { AgentConfig } from "@opencode-ai/sdk";
-import { isClaudeFable5Model, isClaudeFableOrMythosModel, isClaudeOpus46Model, isClaudeOpus47Model, isClaudeOpus47OrLaterModel, isClaudeOpus48Model, isClaudeOpus5Model, isGeminiModel, isGlmModel, isGptModel, isGrok45Model, isGrok46Model, isKimiK2Model, isKimiK27Model, isKimiK3Model, isMiniMaxModel } from "@oh-my-opencode/model-core";
-export { isClaudeFable5Model, isClaudeFableOrMythosModel, isClaudeOpus46Model, isClaudeOpus47Model, isClaudeOpus47OrLaterModel, isClaudeOpus48Model, isClaudeOpus5Model, isGeminiModel, isGlmModel, isGptModel, isGrok45Model, isGrok46Model, isKimiK2Model, isKimiK27Model, isKimiK3Model, isMiniMaxModel, };
+import { isClaudeFable5Model, isClaudeFableOrMythosModel, isClaudeOpus46Model, isClaudeOpus47Model, isClaudeOpus47OrLaterModel, isClaudeOpus48Model, isClaudeOpus5Model, isGeminiModel, isGlmModel, isGptModel, isGrok45Model, isGrok46Model, isKimiK2CodeModel, isKimiK2Model, isKimiK27Model, isKimiK28Model, isKimiK3Model, isMiniMaxModel } from "@oh-my-opencode/model-core";
+export { isClaudeFable5Model, isClaudeFableOrMythosModel, isClaudeOpus46Model, isClaudeOpus47Model, isClaudeOpus47OrLaterModel, isClaudeOpus48Model, isClaudeOpus5Model, isGeminiModel, isGlmModel, isGptModel, isGrok45Model, isGrok46Model, isKimiK2CodeModel, isKimiK2Model, isKimiK27Model, isKimiK28Model, isKimiK3Model, isMiniMaxModel, };
 /**
  * Anthropic Opus 4.7+, Fable, and Mythos models reject thinking.type "enabled";
  * they require adaptive thinking plus an effort, which OpenCode core derives from
@@ -71,6 +71,8 @@ export declare function isGptNativeSisyphusModel(model: string): boolean;
 export declare function isGpt5_5Model(model: string): boolean;
 /** Matches the GPT-5.6 family: gpt-5.6, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna. */
 export declare function isGpt5_6Model(model: string): boolean;
+/** Matches GPT-6 models, including provider-prefixed and fast-tier IDs. */
+export declare function isGpt6Model(model: string): boolean;
 export type BuiltinAgentName = "sisyphus" | "hephaestus" | "oracle" | "librarian" | "explore" | "multimodal-looker" | "metis" | "momus" | "atlas" | "sisyphus-junior";
 export type OverridableAgentName = "build" | BuiltinAgentName;
 export type AgentName = BuiltinAgentName;

package/dist/cli/config-manager/parse-opencode-config-file.d.ts

@@ -1,9 +1,10 @@
+import { type PluginEntry } from "../../shared";
 interface ParseConfigResult {
     config: OpenCodeConfig | null;
     error?: string;
 }
 export interface OpenCodeConfig {
-    plugin?: string[];
+    plugin?: PluginEntry[];
     [key: string]: unknown;
 }
 export declare function parseOpenCodeConfigFileWithError(path: string): ParseConfigResult;

package/dist/cli/config-manager/refresh-opencode-plugin-sandbox.d.ts

@@ -0,0 +1,31 @@
+/**
+ * Clears OpenCode's plugin sandbox for the spec(s) the installer just wrote.
+ *
+ * OpenCode serves the plugin from `<cache>/packages/<spec>/` and its
+ * `Npm.add()` never re-resolves the tag while that sandbox exists, so a user
+ * who re-runs the installer to upgrade would keep loading the version cached
+ * there. The next OpenCode start reinstalls the spec at its current version.
+ *
+ * The installer is often run while OpenCode is open (from another terminal, or
+ * by the agent itself), and a live session reads from the sandbox lazily. A
+ * sandbox another process holds a lease on is therefore only marked for
+ * refresh; the last OpenCode process to exit removes it.
+ *
+ * Only the specs the config actually loads are touched, so an unrelated
+ * channel a user keeps around (`@beta` next to `@latest`) is left alone.
+ */
+export interface RefreshOpenCodePluginSandboxesOptions {
+    readonly configDir?: string;
+    readonly cacheDir?: string;
+}
+export interface RefreshOpenCodePluginSandboxesResult {
+    readonly removed: readonly string[];
+    /** Sandboxes a running OpenCode still uses; they refresh when it exits. */
+    readonly deferred: readonly string[];
+    /** Sandboxes that could not be removed; the caller must tell the user. */
+    readonly failed: readonly {
+        readonly dir: string;
+        readonly message: string;
+    }[];
+}
+export declare function refreshOpenCodePluginSandboxes(options?: RefreshOpenCodePluginSandboxesOptions): RefreshOpenCodePluginSandboxesResult;