GitHub Agentic Workflows 正式全面可用（GA）：在 GitHub Actions 中原生运行 AI 编码 Agent

来自 GitHub Next 的开源项目 GitHub Agentic Workflows 现已正式全面可用（GA）。借助 Agentic 工作流，你可以在 GitHub Actions 内部使用编码 Agent，自动化处理 Issue 分类分发、CI 失败分析、文档更新以及测试编写等基于推理的任务。

只需在自然语言 Markdown 文件中定义你的自动化流程，GitHub Agentic Workflows 就会将其编译为标准的 Actions YAML。因为它们本质上就是 GitHub Actions，所以能够复用你现有的 Runner 组和策略约束。你可以使用 GitHub Copilot，也可以自带编码 Agent 来驱动每个工作流。

我们一直在逐步推行基于 Agent 的开发者工作流，其成效是实打实可衡量的。仅我们的一位主管，每周就在开发工作上节省了约 29 个小时。在整个团队中，测算出的研发速度平均提高了一倍以上，目前团队 91% 的 commit 均由 AI 编写。借助这些工作流，团队将生产环境报错从每天约 11,000 次削减到不足 1,000 次，并端到端地推进了一次技术架构迁移。

其中一个工作流能将功能需求直接转化为带有验证凭据的落地实现。其他工作流则用于保持依赖框架为最新版本，并持续提升代码质量与测试覆盖率。由于 AI 能够端到端推动流程落地，我们的工程师可以并行推进四到五个项目，并专注于更宏大的目标。
Ben Alton – 埃森哲策略、架构与创新团队高级经理

GitHub Agentic Workflows 为你的自动化流程整合了分层安全防护。Agent 访问 GitHub 内容时需遵循完整性过滤器（integrity filter）规则，默认以只读权限运行，并在位于 Agent Workflow Firewall 防护之后的沙箱容器中执行。所有输出都会经过安全输出（safe outputs）流程的校验，并且专属的威胁检测（threat detection）任务会在所有提议的变更生效前进行全面扫描。

自公开预览版以来的重点更新包括：

• Actions Runner Controller (ARC) 与 Docker-in-Docker 支持： 支持在配置了 Docker-in-Docker 的自托管 ARC Runner 上运行 Agentic 工作流。
• 改进的预算预测： gh aw forecast 不再是实验性功能。它现在同时包含威胁检测额度与月度预测区间，能让你更全面地掌握 AI 点数消耗情况。
• 全新模型、引擎选择与工作流： 刷新了支持的模型目录，支持按工作流单独选取模型供应商，并提供了可通过 gh aw add-wizard 添加的新工作流模板。
• 模型路由（Model routing）： 工作流可接入模型路由机制，根据任务类型与难度进行针对性优化。
• 强化的 Agent 隔离： Agent 可以在 Cloud Hypervisor 或 NVX 微虚拟机（micro-VM）中运行，并通过 Enclave 飞地机制防止跨仓库工作流中的数据泄露。

请参阅 GitHub Agentic Workflows 发布说明查看完整更新列表。

按照快速入门指南安装 CLI 扩展，数分钟内即可触发你的首个工作流。也可以在 GitHub Next 的 agentics 仓库中探索预构建的工作流，获取涵盖分类分发、依赖管理、代码审查和报告生成等开箱即用的示例。

GitHub Agentic Workflows 现已在 github.com 上向 GitHub Free、Team 和 Enterprise 用户全面可用。使用默认的 GitHub Copilot CLI 引擎驱动工作流需要具备 Copilot 许可证。

欢迎查阅 GitHub Agentic Workflows 文档了解更多详情，并加入社区讨论参与交流。

GitHub Agentic Workflows, an open source project from GitHub Next, is now generally available. With agentic workflows, you can automate reasoning-based tasks like issue triage, CI failure analysis, documentation updates, and test writing by using coding agents inside GitHub Actions.

Define your automation in natural language Markdown files, and GitHub Agentic Workflows compiles them into standard Actions YAML. Because these are GitHub Actions, they reuse your existing runner groups and policy constraints. You can use GitHub Copilot or bring your own coding agent to power each workflow.

We’ve been rolling out our agentic developer workflows, and the results are measurable. One of our leads saved an estimated 29 hours a week on his development work alone. Across his team, measured velocity has more than doubled on average, and AI now authors 91% of the team’s commits. With these workflows, the team cut production errors from about 11,000 a day to under 1,000 and drove a technology migration end to end.

One workflow takes a feature from requirements to implementation with validation evidence. Others keep our frameworks current and steadily raise code quality and test coverage. Because AI drives the process to completion, our engineers can run four or five efforts in parallel and stay focused on the bigger goals.
Ben Alton – Senior manager, Strategy, Architecture & Innovation team at Accenture

GitHub Agentic Workflows incorporates layered safeguards for your automation. Agents access GitHub content respecting the integrity filter rules, run with read-only permissions by default, and execute inside a sandboxed container behind the Agent Workflow Firewall. The outputs are validated through the safe outputs process, and a dedicated threat detection job scans all proposed changes before they are applied.

Highlights since the public preview include:

• Actions Runner Controller (ARC) and Docker-in-Docker support: Run agentic workflows on self-hosted ARC runners with Docker-in-Docker.
• Improved budget forecasting: gh aw forecast is no longer experimental. It now includes both threat detection credits and monthly ranges for a fuller picture of your AI credit spend.
• New models, engine choices, and workflows: The model inventory has been refreshed, model providers can be picked per workflow, and there are new workflow templates that can be added with gh aw add-wizard.
• Model routing: Workflows can use a model router that optimizes for task type and difficulty.
• Improved agent isolation: Agents can run in a Cloud Hypervisor or NVX micro-VM and prevent data leaks in cross-repository workflows with enclaves.

See the GitHub Agentic Workflows release notes for the full list of changes.

Follow the quickstart guide to install the CLI extension and trigger your first workflow in minutes. Explore prebuilt workflows in GitHub Next’s agentics repository for ready-to-use examples covering triage, dependency management, code review, and reporting.

GitHub Agentic Workflows is now generally available on github.com for GitHub Free, Team, and Enterprise. Using the default GitHub Copilot CLI engine to power a workflow requires a Copilot license.

Learn more in the GitHub Agentic Workflows documentation, and join the conversation in the community discussion.