nexus-coding-agent 0.2.0__tar.gz → 0.3.0__tar.gz

nexus_coding_agent-0.3.0/AGENTS.md

@@ -0,0 +1,34 @@
+# Nexus Next · V0.3
+
+本仓库当前准备 Nexus 0.3.0 发布候选。本文件只作开发入口，不复制项目契约。
+当前行为见 [使用手册](docs/next/usage-guide.zh-CN.md)、
+[代码导读](docs/next/engineering-walkthrough.md) 与
+[发布/验收记录](docs/next/release-0.3.0.md)。原始 V0 设计保留为历史基线，
+V0.3 已批准的增量见开发设计顶部索引；不从旧基线推断尚未实现或禁止现有能力。
+
+## 有效设计依据
+
+- [冻结架构原文](docs/next/architecture-contract-v0.md)
+- [开发设计 v0.2](docs/next/01-development-design.md)
+- [实施与验收清单](docs/next/02-delivery-and-acceptance.md)
+
+`docs/next/00-review-and-decisions.md` 只记录审核、迁移和状态，不建立新的长期 Contract / ADR 权威层。
+
+## 旧规则失效
+
+旧 Nexus 的 Day1–Day10、旧 Product Specification、旧 ADR、spec-addenda、旧 AGENTS 指令全部失效，不得从历史文档、代码或 Memory 继续继承为默认规则。这也包括 LangGraph 工作流、Planner / Validator / Validation / Repair / Replan 节点、SAFE / WRITE / DANGEROUS 与 Plan-based authorization，以及旧 PostgreSQL / pgvector / RAG / Skills / LangSmith 架构要求。遗留源码、测试、配置和示例是待迁移材料，不是 Next 的设计依据。
+
+## 开发原则
+
+- Minimal scaffold, evidence-driven complexity。
+- 一个 hand-written message-driven Agent Loop；不擅自加入冻结范围外的新架构。
+- 实现细节优先采用最简单合理方案；只有真正改变冻结架构边界的问题，才停止并报告证据与选项。
+- 按当前任务范围推进并验证；不恢复旧 Product Owner / knowledge review / Day gate 审批体系。
+- 不自行 commit / push / merge / publish。
+
+## 测试临时文件
+
+- 普通 pytest 运行不指定 `--basetemp`，使用系统临时目录；禁止在仓库根目录新增 `.pytest-tmp*` / `.pytest-temp*`。项目配置会清理通过用例的 `tmp_path`，仅短期保留失败用例目录。
+- 如需隔离并发测试，优先在系统临时目录创建本次运行专用父目录并设置进程级 `PYTEST_DEBUG_TEMPROOT`；不要复用其他用户创建的目录。确需 `--basetemp` 时也只能指向系统临时目录下的专用子目录，结束后另行清理，因为自定义 basetemp 不提供默认的会话保留清理。
+- 开发收尾及获准推送前检查本次产生的临时文件；清理已确认无用的文件。需要交付的测试报告、实验原始输出和失败证据先保存到约定的证据目录，不能仅因名称带 tmp 就删除。
+- 清理只针对已核实的未跟踪临时路径，不跟随符号链接或 junction；遇到访问拒绝、占用或用途不明时保留并报告，不擅自修改权限。

nexus_coding_agent-0.3.0/PKG-INFO

@@ -0,0 +1,210 @@
+Metadata-Version: 2.5
+Name: nexus-coding-agent
+Version: 0.3.0
+Summary: A small, local, message-driven coding agent.
+Project-URL: Homepage, https://github.com/K529529/Nexus
+Project-URL: Repository, https://github.com/K529529/Nexus
+Project-URL: Issues, https://github.com/K529529/Nexus/issues
+Author: Archer
+License-Expression: MIT
+License-File: LICENSE
+Requires-Python: >=3.12
+Requires-Dist: httpx<1,>=0.27
+Requires-Dist: jinja2<4,>=3.1.6
+Requires-Dist: litellm<2,>=1.80
+Requires-Dist: mcp<3,>=2
+Requires-Dist: openai<3,>=2
+Requires-Dist: opentelemetry-exporter-otlp-proto-http<2,>=1.39
+Requires-Dist: opentelemetry-sdk<2,>=1.39
+Requires-Dist: prompt-toolkit<4,>=3
+Requires-Dist: pyyaml<7,>=6
+Requires-Dist: rich<15,>=14
+Requires-Dist: tokenizers<0.24,>=0.23.3
+Requires-Dist: tomlkit<1,>=0.13
+Description-Content-Type: text/markdown
+
+<p>
+  <img src="docs/next/assets/nexus-wordmark-violet.svg" alt="Nexus" width="360">
+</p>
+
+### Understand your project. Write code. Work together.
+
+Nexus is an AI coding assistant for your terminal. Describe a task in plain language,
+let it explore the project, edit code and run tests, then work through the results.
+For larger tasks, multiple agents can share the work while you follow progress,
+add guidance or queue what comes next.
+
+Python 3.12+ · Windows / Linux · [MIT](LICENSE)
+
+**English** / [简体中文](README.zh-CN.md) · [Quick start](#quick-start) · [Multi-Agent](#multi-agent) · [User guide](docs/next/usage-guide.md)
+
+---
+
+## What you can do
+
+| Capability | Put it to work |
+| --- | --- |
+| **Go from issue to change** | Investigate code, fix bugs and add tests directly in your project. |
+| **Share work across agents** | Use scouts for investigation, workers for implementation and reviewers for review. |
+| **Stay in the conversation** | Press Enter to guide a running task or Tab to queue the next one. |
+| **Pick up where you left off** | Resume saved sessions and manage ongoing work with context compaction and local memory. |
+| **Connect your tools** | Choose your models, connect MCP services and use Skills for repeatable workflows. |
+| **See what is happening** | Inspect tool results, agent status, session history and usage; enable profiling or tracing as needed. |
+
+## Quick start
+
+### 1. Install
+
+You need Python 3.12+ and Git. Install from source in a dedicated environment:
+
+```sh
+git clone https://github.com/K529529/Nexus.git
+cd Nexus
+python -m venv .venv
+```
+
+Activate the environment:
+
+```powershell
+# Windows PowerShell
+.\.venv\Scripts\Activate.ps1
+```
+
+```sh
+# Linux
+source .venv/bin/activate
+```
+
+Install and check the version:
+
+```sh
+python -m pip install .
+python -m nexus --version
+```
+
+### 2. Connect a model
+
+Run `nexus`. The first-launch wizard asks for a model name, Base URL, context window
+and API-key environment variable name. Settings are saved in `~/.nexus/config.toml`
+and shared across projects.
+
+Set the key in the terminal that launches Nexus, then restart it:
+
+```powershell
+# Windows PowerShell
+$env:NEXUS_MODEL_API_KEY = "<your key>"
+```
+
+```sh
+# Linux
+export NEXUS_MODEL_API_KEY="<your key>"
+```
+
+Use OpenAI-compatible, DeepSeek or DashScope services, configure model aliases and
+switch with `/model`. For long sessions, configure a separate compaction model too.
+See [model and context configuration](docs/next/usage-guide.md#configure).
+
+### 3. Start in your project
+
+Keep the environment activated and change to the project you want to work on:
+
+```sh
+cd /path/to/your/project
+nexus
+```
+
+> Find why the login endpoint handles blank usernames inconsistently. Fix it, add regression coverage and run the relevant tests.
+
+Your launch directory is the workspace. A project `AGENTS.md` can supply coding conventions
+and working instructions. You can also run a single task directly:
+
+```sh
+nexus exec "Explain this project's entry points, core modules and how to run its tests"
+```
+
+<a id="multi-agent"></a>
+## Multi-Agent collaboration
+
+Give investigation, implementation and review to different agents, and follow their work
+from one terminal.
+
+| Role | Responsibility |
+| --- | --- |
+| **scout** | Explore code and tests, then report findings to the parent agent. |
+| **worker** | Implement changes within an assigned scope, run tools and check the results. |
+| **reviewer** | Review changes, inspect edge cases and report issues or recommendations. |
+
+Multi-Agent is off by default. Add this section to your existing `~/.nexus/config.toml`,
+then restart Nexus:
+
+```toml
+[multi_agent]
+enabled = true
+max_agents = 3
+allowNestedSubagents = false
+```
+
+For example:
+
+> Investigate the login flow and existing tests in parallel. Bring the findings together, fix the issue, then have a reviewer check the changes.
+
+The main agent creates children, assigns scopes and brings their results together.
+Each child has its own context and conversation while sharing the project directory.
+You can configure models, reasoning levels and MCP access by role.
+The default limit is 3 active child agents, excluding the main agent.
+
+- **Browse the task tree:** press F6 or enter `/agents` to see task names, roles, short IDs and status.
+- **Open a child conversation:** select an agent or enter `/agents ID`, then type and press Enter to guide it.
+- **Return to the main agent:** press Esc or enter `/agent root`. Tab in a child view still queues work for the main agent.
+- **Manage ongoing work:** use `/agent cancel ID` to cancel a subtree and `/agents history` to browse archived tasks.
+
+Completed child conversations remain read-only. Assign distinct file scopes in the shared
+workspace; scout/reviewer responsibilities do not replace system permission controls.
+[Explore role configuration, nesting and permissions →](docs/next/usage-guide.md#optional-multi-agent)
+
+## A terminal you can keep working in
+
+| Action | Key / command |
+| --- | --- |
+| Guide the running task | Type your message and press **Enter** |
+| Queue the next task | Type while running and press **Tab** |
+| Insert a newline | **F4** / **Alt+Enter** |
+| Cancel the current task | **Ctrl+C** |
+| Browse history / jump to latest | **PgUp**, **PgDn** / **Ctrl+End** |
+| Inspect, edit or withdraw pending input | `/pending`, `/edit ID text`, `/withdraw ID` |
+| Switch model / inspect permissions | `/model` / `/permissions` |
+| Resume history / start fresh | `/resume` / `/new` |
+| Help / exit | `/help` / `/exit` |
+
+Multiline editing and paste are supported. New output does not pull you away from the
+history you are reading. When a completion menu is open, Tab selects a completion first.
+See [commands and input](docs/next/usage-guide.md#commands-and-input).
+
+## Make it yours
+
+**Context and memory.** A separate compaction model summarizes older conversation while
+keeping recent work available. Long-term memory lives in local Markdown files; inspect or
+toggle it with `/memory`. Memory is on by default and may download retrieval dependencies
+and models on first use. Run `nexus memory off` before starting if you do not need it.
+
+**Tools and extensions.** Connect existing tools through MCP and provide reusable task
+instructions through Skills. Web tools and external tracing are explicitly opt-in.
+[Tools, Skills and MCP →](docs/next/usage-guide.md#tools-skills-and-mcp)
+
+**Execution permissions.** Local mode uses your current user permissions by default.
+Read-only and workspace-write modes use Docker on Linux/WSL and the Native backend on
+Windows. Windows enforcement is partial and its network access is unrestricted.
+See [permissions and recovery](docs/next/usage-guide.md#permissions-and-recovery).
+
+**Sessions and records.** Continue a saved session with `nexus resume`, inspect run metrics
+with `nexus --profile`, or emit JSON events with `nexus exec "task" --json`.
+[Logs and observability →](docs/next/usage-guide.md#logs-reports-and-observability)
+
+## Learn more and contribute
+
+[User guide](docs/next/usage-guide.md) ·
+[Architecture walkthrough](docs/next/engineering-walkthrough.md) ·
+[Development guide](docs/next/development-and-release.md) ·
+[Report an issue](https://github.com/K529529/Nexus/issues)
+
+Nexus is available under the [MIT License](LICENSE).

nexus_coding_agent-0.3.0/README.md

@@ -0,0 +1,185 @@
+<p>
+  <img src="docs/next/assets/nexus-wordmark-violet.svg" alt="Nexus" width="360">
+</p>
+
+### Understand your project. Write code. Work together.
+
+Nexus is an AI coding assistant for your terminal. Describe a task in plain language,
+let it explore the project, edit code and run tests, then work through the results.
+For larger tasks, multiple agents can share the work while you follow progress,
+add guidance or queue what comes next.
+
+Python 3.12+ · Windows / Linux · [MIT](LICENSE)
+
+**English** / [简体中文](README.zh-CN.md) · [Quick start](#quick-start) · [Multi-Agent](#multi-agent) · [User guide](docs/next/usage-guide.md)
+
+---
+
+## What you can do
+
+| Capability | Put it to work |
+| --- | --- |
+| **Go from issue to change** | Investigate code, fix bugs and add tests directly in your project. |
+| **Share work across agents** | Use scouts for investigation, workers for implementation and reviewers for review. |
+| **Stay in the conversation** | Press Enter to guide a running task or Tab to queue the next one. |
+| **Pick up where you left off** | Resume saved sessions and manage ongoing work with context compaction and local memory. |
+| **Connect your tools** | Choose your models, connect MCP services and use Skills for repeatable workflows. |
+| **See what is happening** | Inspect tool results, agent status, session history and usage; enable profiling or tracing as needed. |
+
+## Quick start
+
+### 1. Install
+
+You need Python 3.12+ and Git. Install from source in a dedicated environment:
+
+```sh
+git clone https://github.com/K529529/Nexus.git
+cd Nexus
+python -m venv .venv
+```
+
+Activate the environment:
+
+```powershell
+# Windows PowerShell
+.\.venv\Scripts\Activate.ps1
+```
+
+```sh
+# Linux
+source .venv/bin/activate
+```
+
+Install and check the version:
+
+```sh
+python -m pip install .
+python -m nexus --version
+```
+
+### 2. Connect a model
+
+Run `nexus`. The first-launch wizard asks for a model name, Base URL, context window
+and API-key environment variable name. Settings are saved in `~/.nexus/config.toml`
+and shared across projects.
+
+Set the key in the terminal that launches Nexus, then restart it:
+
+```powershell
+# Windows PowerShell
+$env:NEXUS_MODEL_API_KEY = "<your key>"
+```
+
+```sh
+# Linux
+export NEXUS_MODEL_API_KEY="<your key>"
+```
+
+Use OpenAI-compatible, DeepSeek or DashScope services, configure model aliases and
+switch with `/model`. For long sessions, configure a separate compaction model too.
+See [model and context configuration](docs/next/usage-guide.md#configure).
+
+### 3. Start in your project
+
+Keep the environment activated and change to the project you want to work on:
+
+```sh
+cd /path/to/your/project
+nexus
+```
+
+> Find why the login endpoint handles blank usernames inconsistently. Fix it, add regression coverage and run the relevant tests.
+
+Your launch directory is the workspace. A project `AGENTS.md` can supply coding conventions
+and working instructions. You can also run a single task directly:
+
+```sh
+nexus exec "Explain this project's entry points, core modules and how to run its tests"
+```
+
+<a id="multi-agent"></a>
+## Multi-Agent collaboration
+
+Give investigation, implementation and review to different agents, and follow their work
+from one terminal.
+
+| Role | Responsibility |
+| --- | --- |
+| **scout** | Explore code and tests, then report findings to the parent agent. |
+| **worker** | Implement changes within an assigned scope, run tools and check the results. |
+| **reviewer** | Review changes, inspect edge cases and report issues or recommendations. |
+
+Multi-Agent is off by default. Add this section to your existing `~/.nexus/config.toml`,
+then restart Nexus:
+
+```toml
+[multi_agent]
+enabled = true
+max_agents = 3
+allowNestedSubagents = false
+```
+
+For example:
+
+> Investigate the login flow and existing tests in parallel. Bring the findings together, fix the issue, then have a reviewer check the changes.
+
+The main agent creates children, assigns scopes and brings their results together.
+Each child has its own context and conversation while sharing the project directory.
+You can configure models, reasoning levels and MCP access by role.
+The default limit is 3 active child agents, excluding the main agent.
+
+- **Browse the task tree:** press F6 or enter `/agents` to see task names, roles, short IDs and status.
+- **Open a child conversation:** select an agent or enter `/agents ID`, then type and press Enter to guide it.
+- **Return to the main agent:** press Esc or enter `/agent root`. Tab in a child view still queues work for the main agent.
+- **Manage ongoing work:** use `/agent cancel ID` to cancel a subtree and `/agents history` to browse archived tasks.
+
+Completed child conversations remain read-only. Assign distinct file scopes in the shared
+workspace; scout/reviewer responsibilities do not replace system permission controls.
+[Explore role configuration, nesting and permissions →](docs/next/usage-guide.md#optional-multi-agent)
+
+## A terminal you can keep working in
+
+| Action | Key / command |
+| --- | --- |
+| Guide the running task | Type your message and press **Enter** |
+| Queue the next task | Type while running and press **Tab** |
+| Insert a newline | **F4** / **Alt+Enter** |
+| Cancel the current task | **Ctrl+C** |
+| Browse history / jump to latest | **PgUp**, **PgDn** / **Ctrl+End** |
+| Inspect, edit or withdraw pending input | `/pending`, `/edit ID text`, `/withdraw ID` |
+| Switch model / inspect permissions | `/model` / `/permissions` |
+| Resume history / start fresh | `/resume` / `/new` |
+| Help / exit | `/help` / `/exit` |
+
+Multiline editing and paste are supported. New output does not pull you away from the
+history you are reading. When a completion menu is open, Tab selects a completion first.
+See [commands and input](docs/next/usage-guide.md#commands-and-input).
+
+## Make it yours
+
+**Context and memory.** A separate compaction model summarizes older conversation while
+keeping recent work available. Long-term memory lives in local Markdown files; inspect or
+toggle it with `/memory`. Memory is on by default and may download retrieval dependencies
+and models on first use. Run `nexus memory off` before starting if you do not need it.
+
+**Tools and extensions.** Connect existing tools through MCP and provide reusable task
+instructions through Skills. Web tools and external tracing are explicitly opt-in.
+[Tools, Skills and MCP →](docs/next/usage-guide.md#tools-skills-and-mcp)
+
+**Execution permissions.** Local mode uses your current user permissions by default.
+Read-only and workspace-write modes use Docker on Linux/WSL and the Native backend on
+Windows. Windows enforcement is partial and its network access is unrestricted.
+See [permissions and recovery](docs/next/usage-guide.md#permissions-and-recovery).
+
+**Sessions and records.** Continue a saved session with `nexus resume`, inspect run metrics
+with `nexus --profile`, or emit JSON events with `nexus exec "task" --json`.
+[Logs and observability →](docs/next/usage-guide.md#logs-reports-and-observability)
+
+## Learn more and contribute
+
+[User guide](docs/next/usage-guide.md) ·
+[Architecture walkthrough](docs/next/engineering-walkthrough.md) ·
+[Development guide](docs/next/development-and-release.md) ·
+[Report an issue](https://github.com/K529529/Nexus/issues)
+
+Nexus is available under the [MIT License](LICENSE).

nexus_coding_agent-0.3.0/README.zh-CN.md

@@ -0,0 +1,176 @@
+<p>
+  <img src="docs/next/assets/nexus-wordmark-violet.svg" alt="Nexus" width="360">
+</p>
+
+### 理解项目，编写代码，协作解决问题。
+
+Nexus 是运行在终端里的 AI 编程助手。用自然语言描述任务，让它阅读项目、修改代码、
+运行测试，并根据结果继续处理。复杂任务还可以交给多个 Agent 分工协作，
+你随时可以查看进展、补充指导或排队下一项工作。
+
+Python 3.12+ · Windows / Linux · [MIT](LICENSE)
+
+[English](README.md) / **简体中文** · [快速开始](#快速开始) · [Multi-Agent](#multi-agent) · [使用手册](docs/next/usage-guide.zh-CN.md)
+
+---
+
+## 用 Nexus 做什么
+
+| 能力 | 使用方式 |
+| --- | --- |
+| **从问题到修改** | 调查代码、修复缺陷、补充测试，在项目目录中完成一轮开发工作。 |
+| **多个 Agent 分工** | 让 scout 调查、worker 实现、reviewer 审查，在同一工作区协作。 |
+| **边运行，边沟通** | Enter 补充指导，Tab 排队后续任务，不必等当前任务结束。 |
+| **接着上次继续** | 恢复历史会话；通过上下文压缩与长期记忆管理持续进行的工作。 |
+| **接入你的工具** | 选择模型，通过 MCP 连接工具，使用 Skills 提供项目工作方法。 |
+| **看清执行过程** | 查看工具结果、代理状态、会话记录和用量，按需开启运行统计与遥测。 |
+
+## 快速开始
+
+### 1. 安装
+
+需要 Python 3.12+ 和 Git。在独立环境中从源码安装：
+
+```sh
+git clone https://github.com/K529529/Nexus.git
+cd Nexus
+python -m venv .venv
+```
+
+激活环境：
+
+```powershell
+# Windows PowerShell
+.\.venv\Scripts\Activate.ps1
+```
+
+```sh
+# Linux
+source .venv/bin/activate
+```
+
+安装并检查版本：
+
+```sh
+python -m pip install .
+python -m nexus --version
+```
+
+### 2. 连接模型
+
+运行 `nexus`，按首次启动向导填写模型名、Base URL、上下文窗口和 API Key 环境变量名。
+配置保存在 `~/.nexus/config.toml`，可在不同项目间复用。
+
+将密钥设置在启动 Nexus 的终端中，然后重新启动：
+
+```powershell
+# Windows PowerShell
+$env:NEXUS_MODEL_API_KEY = "<你的密钥>"
+```
+
+```sh
+# Linux
+export NEXUS_MODEL_API_KEY="<你的密钥>"
+```
+
+支持 OpenAI-compatible、DeepSeek 和 DashScope 接入，可配置多个模型并通过
+`/model` 切换。长会话建议同时配置独立压缩模型，见[模型与上下文配置](docs/next/usage-guide.zh-CN.md#配置)。
+
+### 3. 在项目里开始工作
+
+保持安装环境激活，切换到要处理的项目目录：
+
+```sh
+cd /path/to/your/project
+nexus
+```
+
+> 找到登录接口对空白用户名处理不一致的原因，修复它，补充回归测试并运行相关测试。
+
+启动目录就是工作区，项目中的 `AGENTS.md` 可以提供编码规范和工作约定。
+也可以直接执行单次任务：
+
+```sh
+nexus exec "梳理这个项目的入口、核心模块和测试运行方式"
+```
+
+<a id="multi-agent"></a>
+## Multi-Agent 协作
+
+把调查、实现和审查交给不同 Agent，并在同一个终端里跟进它们的工作。
+
+| 角色 | 分工 |
+| --- | --- |
+| **scout** | 阅读代码、调查实现与测试，向主 Agent 返回发现。 |
+| **worker** | 在分配的范围内编写代码、执行工具并验证修改。 |
+| **reviewer** | 审查改动、检查边界情况，提出问题与建议。 |
+
+Multi-Agent 默认关闭。在已有 `~/.nexus/config.toml` 中添加以下配置，重新启动即可开启：
+
+```toml
+[multi_agent]
+enabled = true
+max_agents = 3
+allowNestedSubagents = false
+```
+
+例如，你可以这样安排工作：
+
+> 并行调查登录流程和现有测试，汇总后修复问题，再让 reviewer 检查改动。
+
+主 Agent 根据任务创建子代理、分配范围并汇总结果。每个子代理有独立的上下文与会话，
+共享当前项目目录；可为不同角色配置模型、推理强度和 MCP 服务。
+默认最多同时运行 3 个子代理，不包含主 Agent。
+
+- **查看任务树**：按 F6 或输入 `/agents`，查看任务名、角色、短 ID 和状态。
+- **进入子代理会话**：选择代理，或输入 `/agents ID`；直接输入并按 Enter 给它指导。
+- **回到主 Agent**：按 Esc 或输入 `/agent root`。在子代理视图按 Tab，仍会给主 Agent 排队。
+- **管理后续工作**：使用 `/agent cancel ID` 取消子树，`/agents history` 查看历史任务。
+
+已完成的子代理会话保留为只读。共享工作区适合明确划分文件范围，
+scout/reviewer 的职责分工不替代系统权限控制。
+[查看角色配置、嵌套协作与权限说明 →](docs/next/usage-guide.zh-CN.md#多-agent)
+
+## 顺手的终端交互
+
+| 操作 | 快捷键 / 命令 |
+| --- | --- |
+| 指导正在运行的任务 | 输入内容，按 **Enter** |
+| 排队下一项任务 | 运行中输入内容，按 **Tab** |
+| 换行 | **F4** / **Alt+Enter** |
+| 取消当前任务 | **Ctrl+C** |
+| 阅读历史 / 回到最新 | **PgUp**、**PgDn** / **Ctrl+End** |
+| 查看、编辑或撤回待处理输入 | `/pending`、`/edit ID 内容`、`/withdraw ID` |
+| 切换模型 / 查看权限 | `/model` / `/permissions` |
+| 恢复历史 / 新建会话 | `/resume` / `/new` |
+| 查看帮助 / 退出 | `/help` / `/exit` |
+
+支持多行编辑与粘贴；阅读历史时，新输出不会强制跳回底部。
+补全菜单打开时，Tab 优先用于补全。更多操作见[使用手册](docs/next/usage-guide.zh-CN.md#命令与交互)。
+
+## 按你的工作方式配置
+
+**上下文与记忆。** 独立压缩模型整理较早的对话，保留近期工作内容。
+长期记忆以本地 Markdown 保存，可通过 `/memory` 查看或切换；
+默认开启，首次使用可能下载检索依赖与模型。不需要时可先运行 `nexus memory off`。
+
+**工具与扩展。** MCP 连接已有工具服务，Skills 提供可复用的任务说明。
+Web 工具和外部遥测均需显式开启。
+[工具、Skills 与 MCP →](docs/next/usage-guide.zh-CN.md#工具skills-与-mcp)
+
+**执行权限。** Local 默认使用当前用户权限；也可选择只读或工作区写入模式。
+Linux/WSL 使用 Docker，Windows 使用 Native 后端。Windows 为部分强制隔离，
+网络不受限；具体配置见[沙箱与审批](docs/next/usage-guide.zh-CN.md#沙箱与审批)。
+
+**会话与记录。** 使用 `nexus resume` 继续历史会话，`nexus --profile` 查看运行统计，
+`nexus exec "任务" --json` 输出 JSON 事件。
+[恢复与日志 →](docs/next/usage-guide.zh-CN.md#恢复jsonl-与查看方式)
+
+## 文档与参与
+
+[完整使用手册](docs/next/usage-guide.zh-CN.md) ·
+[架构与代码导读](docs/next/engineering-walkthrough.md) ·
+[开发指南](docs/next/development-and-release.md) ·
+[反馈问题](https://github.com/K529529/Nexus/issues)
+
+Nexus 使用 [MIT License](LICENSE)。

{nexus_coding_agent-0.2.0 → nexus_coding_agent-0.3.0}/docs/next/00-review-and-decisions.md

@@ -1,3 +1,11 @@
+# 当前入口（2026-11-05）
+
+以下原文保留历史审核与阶段状态，不覆盖 V0.3 当前实现。
+当前说明见 [0.3.0 Release Notes](release-0.3.0.md)、
+[发布准备与验收](release-0.3.0-validation.md) 及 [使用手册](usage-guide.zh-CN.md)。
+
+---
+
 # Nexus Next V0：审核入口与迁移说明
 
 状态：**Phase 2 Implementation 已获用户授权（2026-10-01）**

{nexus_coding_agent-0.2.0 → nexus_coding_agent-0.3.0}/docs/next/01-development-design.md

@@ -1,8 +1,30 @@
 # Nexus Next V0：开发设计
 
+> 历史设计基线及增量索引。当前 V0.3 已包含连续 Session/独立压缩、长期记忆、
+> 平台沙箱、Durable Execution、模型/遥测和 [Stage8 原生协作](stage8-multi-agent-plan.md)。
+> Stage8 复用同一个手写 Loop，不沿用旧“禁止所有多 Agent”的范围判断。
+> 当前行为见 [代码导读](engineering-walkthrough.md) 和 [0.3.0 待发布说明](release-0.3.0.md)；
+> 下文原始 V0 设计保留，不代表所有旧上下文/恢复规则仍生效。
+
 状态：**APPROVED v0.2 / Phase 2 Implementation** · 2026-10-01
 用户在本次实施任务中明确批准本开发设计与验收清单（IMPLEMENTATION AUTHORIZATION: APPROVED）。按 S1 → S5 连续实施；实际验收状态见 acceptance-evidence.md。不另设 ADR 审批层。
 
+2026-10-21 用户授权的 [V0.3 第一阶段增量](v0.3-stage1.md) 已在阶段分支实施：
+§6.3 的跨 Run 完全隔离最初改为有界会话摘录；新增预算交接与可见输出语言配置，
+补丁仅缺 Begin 标记时允许有证据的窄规范化。其余单 Loop、配对、恢复和私有协议边界保留。
+2026-10-22 用户重新确定 [Context Runtime V0.3](context-runtime-v0.3.md)：
+连续 Session 历史、独立双模型、唯一 Session Compaction，替代有界摘录、Run 压缩与 HOT/COLD。
+以下 V0.2 设计原文用于基线对照，发生上述差异时以最新增量说明为准。
+
+2026-10-29 用户授权的 [Stage 6 Durable Execution V1](stage6-durable-execution.md)
+为工具调用增加 PREPARED / DISPATCHING / COMPLETED 持久化边界。只有明确未分发的
+PREPARED 调用允许经当前 Gateway 续跑；未知副作用不重放，Patch 恢复核对实际哈希。
+本增量取代下文对应的原始 resume 限制，保留单 Loop 与现有 HITL。
+
+2026-10-31 用户批准 [Stage 7 Web 工具增量](stage7-web-tools.md)：在 Local 模式按配置
+提供内置 `web_search` / `web_fetch`，使用独立 DeepSeek Search/Fetch API。此增量扩展
+原 V0 的内置工具范围，复用 ToolResult、Gateway 和 Stage 6 checkpoint，不改变 Agent Loop。
+
 ## 1. 最小运行结构
 
 ```mermaid
@@ -295,7 +317,7 @@ TUI 渲染错误降级为纯文本，不修改模型消息；future OTel sink
 
 无 MCP 配置时完全跳过。调用阶段断连/错误形成 tool observation，副作用未知时如实记录；可停用该失联服务器并更新可用工具清单，但不自动重连、不重试带副作用调用、不重新发现新增工具。用户重新进入会话时重新连接。没有健康检查平台或 failover 子系统。
 
-文本块与 structured JSON 按统一预算回传；`is_error` 保持失败语义。非文本内容返回类型与不可用说明，明确标记 unsupported content，不把 base64 塞入上下文，也不自动抓 resource URL。timeout 使用同一调用预算，取消关闭本地客户端/子进程并说明远端副作用可能未知。
+文本块、内嵌文本资源（`resource.text`）与 structured JSON 按统一预算回传；内嵌文本资源提取正文到现有 `data.text`，不另建结果格式，`is_error` 保持失败语义。二进制资源、资源链接及其他非文本内容返回类型与不可用说明，明确标记 unsupported content，不把 base64 塞入上下文，也不自动抓 resource URL。timeout 使用同一调用预算，取消关闭本地客户端/子进程并说明远端副作用可能未知。
 
 不注册 sampling/elicitation handler，不开启 SDK 的自动多轮 input-required 处理；服务请求这些能力时返回不支持的工具结果。MCP server instructions、工具描述或结果不会成为高优先级项目指令。配置只读用户目录，不自动读取仓库 MCP 配置，不从 repo 安装服务器。
 

{nexus_coding_agent-0.2.0 → nexus_coding_agent-0.3.0}/docs/next/02-delivery-and-acceptance.md

@@ -1,5 +1,8 @@
 # Nexus Next V0：实施与验收清单
 
+> 以下为原始 V0 历史清单，保留原始判定。V0.3 当前验收与合入状态见
+> [0.3.0 发布准备记录](release-0.3.0-validation.md)，不是下文旧阶段状态的机械延续。
+
 状态：**APPROVED v0.2 / Phase 2 Implementation** · 2026-10-01
 
 依据：[冻结架构契约](architecture-contract-v0.md)、[开发设计](01-development-design.md)。用户已在本次实施任务明确授权连续推进 S1 → S5。以下表格定义验收标准，实际执行结果另记 acceptance-evidence.md；历史 S0 说明不再表示本次实施未授权。