# Rudder Docs

> Canonical public documentation for Rudder, generated from scripts/docs-content-map.yml.

Rudder gives humans and agents a shared place to move work from a goal to a reviewable result. Choose Chat for conversational work or an Issue when named ownership, dependencies, or a review path help.

## English

### Start here
- [Rudder](https://docs.rudderhq.dev): Rudder is open-source software for assigning, running, reviewing, and improving AI agent work across goals, Chat, issues, runs, budgets, workspaces, and feedback loops.

### Get started
- [Quick Start](https://docs.rudderhq.dev/get-started/installation): Install Rudder Desktop, confirm a healthy local startup, and recover safely when startup fails.
- [Create Your First Organization](https://docs.rudderhq.dev/get-started/first-organization): Complete one inspectable task with a request, Agent Run, output, and review decision.

### Concepts
- [Core Concepts](https://docs.rudderhq.dev/concepts/overview): Follow one task through Rudder's organizations, work surfaces, agents, runs, outputs, and review.
- [Goals, projects, and issues](https://docs.rudderhq.dev/concepts/goals-projects-issues): Connect an organization's durable direction to grouped work and executable Issue records.
- [Agents and Agent Runs](https://docs.rudderhq.dev/concepts/agents): Distinguish a durable Rudder Agent from its runtime, bounded runs, outputs, cost, and raw evidence.
- [Built-in Browser](https://docs.rudderhq.dev/concepts/built-in-browser): Use Rudder Desktop's shared website session while keeping Agent tab control isolated to one active run.
- [Issues](https://docs.rudderhq.dev/concepts/issues): Use an Issue when agent work needs durable status, ownership, dependencies, or a review path.
- [Chat and Messenger](https://docs.rudderhq.dev/concepts/chat-messenger): Choose Chat or an Issue for the task, then use Messenger to recover work needing human attention.
- [Calendar](https://docs.rudderhq.dev/concepts/calendar): Inspect scheduled and dated work without turning Calendar into a second task state machine.
- [Workspaces and Library](https://docs.rudderhq.dev/concepts/workspaces): Place project context, durable files, execution changes, and backups in their intended filesystem boundaries.
- [Skills](https://docs.rudderhq.dev/concepts/skills): Package reusable operating procedures while keeping discovery, enablement, loading, and proven use distinct.
- [Automations](https://docs.rudderhq.dev/concepts/automations): Run repeatable agent work from a defined trigger, owner, context, concurrency policy, and output route.
- [Reviews, feedback, and learning](https://docs.rudderhq.dev/concepts/reviews-feedback-learning): Judge an output, record durable feedback, and promote lessons without confusing Review with Approval.
- [Plugins](https://docs.rudderhq.dev/concepts/plugins): Extend Rudder with host-managed capabilities, jobs, webhooks, and UI slots without redefining core workflows.

### How-to guides
- [Run an Issue from intake to done](https://docs.rudderhq.dev/how-to/issue-lifecycle): Move one Issue through assignment, Agent execution, evidence, structured review, and accepted close-out.
- [Create an Agent](https://docs.rudderhq.dev/how-to/create-agent): Create a durable Agent with a clear role, capability boundary, runtime, reporting line, and low-risk verification run.
- [Create an Automation](https://docs.rudderhq.dev/how-to/create-automation): Create a repeatable Automation with a stable owner, trigger, context, concurrency policy, and verified output route.
- [Review Agent work](https://docs.rudderhq.dev/how-to/review-agent-work): Inspect one Agent output and its Run evidence, record a structured decision, and route the next action.
- [Manage Workspaces and Library](https://docs.rudderhq.dev/how-to/manage-workspaces-and-library): Curate project context, keep durable output in Library, inspect files safely, and recover organization workspace state.
- [Export and import an organization](https://docs.rudderhq.dev/how-to/export-import-organization): Build a portable organization bundle, preview dependencies and collisions, then import it with fresh identity and no secret leakage.
- [Manage Plugins](https://docs.rudderhq.dev/how-to/manage-plugins): Install, activate, inspect, update, and uninstall a trusted Plugin while preserving core Rudder records.
- [Configure Feishu or Lark](https://docs.rudderhq.dev/how-to/configure-feishu-integration): Connect one Rudder Agent to a Feishu or Lark bot and verify the provider-bound conversation and Run evidence.
- [Configure an Agent runtime](https://docs.rudderhq.dev/how-to/configure-agent-runtime): Configure and test the execution environment Rudder invokes for one Agent, then verify a real low-risk run.

### Reference
- [Issue statuses](https://docs.rudderhq.dev/reference/issue-statuses): Look up each Rudder Issue status, its review rules, and the boundary between state and ownership.
- [Runtime types](https://docs.rudderhq.dev/reference/runtime-types): Compare Rudder runtime choices, readiness states, managed tools, and adapter boundaries.
- [Workspace boundaries](https://docs.rudderhq.dev/reference/workspace-boundaries): Look up where project work, durable files, Agent state, run workspaces, backups, and scratch files belong.
- [Automation output routing](https://docs.rudderhq.dev/reference/automation-output-routing): Compare tracked Issue and Chat output for Rudder Automations, including run records and failure evidence.
- [Permissions and platforms](https://docs.rudderhq.dev/reference/permissions-and-platforms): Look up deployment trust, local runtime homes, credentials, Browser access, and platform-specific file behavior.
- [Approvals, budgets, and activity](https://docs.rudderhq.dev/reference/approvals-budgets-activity): Look up governed decisions, spend limits, audit activity, run evidence, and operator attention in Rudder.

### Project
- [GDPval Harness Benchmark](https://docs.rudderhq.dev/benchmarks/gdpval-harness): Methodology, results, and evidence limits for Rudder's 10-task GDPval-based harness pilot.
- [About Rudder](https://docs.rudderhq.dev/about): Learn what Rudder is, why it exists, and where its responsibilities stop.
- [Contact](https://docs.rudderhq.dev/contact): Choose the public Rudder channel for a question, bug report, feature request, or release update.
- [Changelog](https://docs.rudderhq.dev/releases): Stable Rudder release notes, upgrade notes, and GitHub Release links.

## 简体中文

### Start here
- [Rudder](https://docs.rudderhq.dev/zh): Rudder 为人和 Agent 提供一个共同工作空间，让任务从目标走向可检查、可评审的结果。

### Get started
- [快速开始](https://docs.rudderhq.dev/zh/get-started/installation): 安装 Rudder Desktop，确认本地启动成功，并在启动失败时安全恢复。
- [创建第一个组织](https://docs.rudderhq.dev/zh/get-started/first-organization): 完成一件可检查的任务，让请求、Agent 运行、产物和评审结论留在一起。

### Concepts
- [核心概念](https://docs.rudderhq.dev/zh/concepts/overview): 通过一件任务理解 Rudder 中的组织、任务入口、Agent、运行记录、产物和评审。
- [目标、项目和 Issue](https://docs.rudderhq.dev/zh/concepts/goals-projects-issues): 把组织的长期方向连接到成组工作和可以直接执行的 Issue。
- [Agent 和运行记录](https://docs.rudderhq.dev/zh/concepts/agents): 分清长期存在的 Agent、运行环境、单次运行、产物、成本和原始证据。
- [内置 Browser](https://docs.rudderhq.dev/zh/concepts/built-in-browser): 在 Rudder Desktop 中共享网站登录状态，同时把 Agent 的标签页控制限制在单次运行内。
- [Issue（任务单）](https://docs.rudderhq.dev/zh/concepts/issues): 任务需要持久状态、明确负责人、依赖关系或评审流程时，使用 Issue。
- [Chat 和 Messenger](https://docs.rudderhq.dev/zh/concepts/chat-messenger): 先为任务选择 Chat 或 Issue，再用 Messenger 找回需要人处理的工作。
- [Calendar](https://docs.rudderhq.dev/zh/concepts/calendar): 按时间检查定时和已发生的工作，同时避免建立第二套任务状态。
- [工作区和 Library](https://docs.rudderhq.dev/zh/concepts/workspaces): 把项目上下文、持久文件、运行改动和备份放在各自的文件边界内。
- [技能](https://docs.rudderhq.dev/zh/concepts/skills): 把可复用的操作步骤整理成技能，同时分清发现、启用、加载和确实使用。
- [自动化](https://docs.rudderhq.dev/zh/concepts/automations): 用明确的触发方式、负责人、上下文、并发策略和结果去向重复运行 Agent 工作。
- [评审、反馈和学习](https://docs.rudderhq.dev/zh/concepts/reviews-feedback-learning): 判断产物、保存反馈并谨慎沉淀经验，同时分清评审与审批。
- [插件](https://docs.rudderhq.dev/zh/concepts/plugins): 通过宿主管理的能力、任务、webhook 和界面插槽扩展 Rudder，同时保留核心工作流规则。

### How-to guides
- [让 Issue 从接收走到完成](https://docs.rudderhq.dev/zh/how-to/issue-lifecycle): 把一条 Issue 依次推进到分配、Agent 执行、证据、结构化评审和接受完成。
- [创建 Agent](https://docs.rudderhq.dev/zh/how-to/create-agent): 创建一名职责、能力边界、运行环境和汇报关系都清楚的长期 Agent，并完成低风险验证。
- [创建自动化](https://docs.rudderhq.dev/zh/how-to/create-automation): 创建一项负责人、触发方式、上下文、并发策略和结果去向都稳定的自动化，并验证真实结果。
- [评审 Agent 工作](https://docs.rudderhq.dev/zh/how-to/review-agent-work): 检查一项 Agent 产物和运行证据，记录结构化结论，并把下一步交给正确负责人。
- [管理工作区和 Library](https://docs.rudderhq.dev/zh/how-to/manage-workspaces-and-library): 筛选项目上下文，把持久产物留在 Library，安全检查文件，并恢复组织工作区。
- [导出和导入组织](https://docs.rudderhq.dev/zh/how-to/export-import-organization): 构建可移植组织包，先预览依赖和冲突，再以新身份导入，同时避免泄露机密。
- [管理插件](https://docs.rudderhq.dev/zh/how-to/manage-plugins): 安装、启用、检查、更新和卸载可信插件，同时保留 Rudder 核心记录。
- [配置飞书或 Lark](https://docs.rudderhq.dev/zh/how-to/configure-feishu-integration): 把一个 Rudder Agent 连接到飞书或 Lark 机器人，并验证外部对话和运行证据。
- [配置 Agent 运行环境](https://docs.rudderhq.dev/zh/how-to/configure-agent-runtime): 配置并测试 Rudder 为一个 Agent 调用的执行环境，再用真实的低风险运行确认结果。

### Reference
- [Issue 状态](https://docs.rudderhq.dev/zh/reference/issue-statuses): 查询 Rudder 中每种 Issue 状态、评审规则，以及状态和负责人之间的边界。
- [运行环境类型](https://docs.rudderhq.dev/zh/reference/runtime-types): 比较 Rudder 的运行环境选项、就绪状态、托管工具和适配器边界。
- [工作区边界](https://docs.rudderhq.dev/zh/reference/workspace-boundaries): 查询项目工作、持久文件、Agent 状态、运行工作区、备份和临时文件应该放在哪里。
- [自动化输出路由](https://docs.rudderhq.dev/zh/reference/automation-output-routing): 比较 Rudder 自动化的 Issue 和 Chat 输出，包括运行记录与失败证据。
- [权限与平台](https://docs.rudderhq.dev/zh/reference/permissions-and-platforms): 查询部署信任、本地运行环境目录、凭据、内置浏览器访问和各平台文件行为。
- [审批、预算和活动记录](https://docs.rudderhq.dev/zh/reference/approvals-budgets-activity): 查询 Rudder 中受治理的决策、费用限制、审计活动、运行证据和需要人处理的事项。

### Project
- [GDPval 执行框架基准测试](https://docs.rudderhq.dev/zh/benchmarks/gdpval-harness): Rudder 基于 GDPval 的 10 项任务试验：方法、结果与证据边界。
- [关于 Rudder](https://docs.rudderhq.dev/zh/about): 了解 Rudder 是什么、为什么存在，以及它的职责边界。
- [联系 Rudder](https://docs.rudderhq.dev/zh/contact): 根据问题、缺陷、功能建议或版本更新，选择合适的 Rudder 公开渠道。
- [更新日志](https://docs.rudderhq.dev/zh/releases): Rudder 稳定版本的发布记录、升级说明和 GitHub Release 链接。