This release covers AI agents as governed, first-class repository participants: how to govern them, and how to write conventions for them.
Highlights
- Added the AI Agent Governance guide: three agent identity models, permission scoping, PR approval rules, CI triggers with secrets isolation, and traceability, with hosted agent restrictions grounded in the official GitHub documentation
- Added the AGENTS.md template, copyable into team repositories, with setup commands, required verification, boundary rules, commit conventions, and a mapping to tool-specific instruction files
- This repository now follows its own recommendations: a root AGENTS.md, AI-assisted contribution rules in the contributing guides, and documentation checks that include AGENTS.md
- Expanded Multi-Agent Branch Strategy with executable workflows: comparing experiment branches, integrating the winning solution, handling same-file collisions, and cleaning up losing branches
- Expanded Stacked PR for AI-Generated Changes with native Git stack operations (
--update-refs,rebase --onto,--force-with-lease) and Graphite CLI core commands - All content above is maintained as Chinese and English pairs
Why This Release
v2.0 set the direction toward AI-native development workflows, but agent governance and collaboration conventions were still missing.
AI agents are already creating branches, opening PRs, and triggering CI. v2.1.0 upgrades them from diff generators to governed participants: maintainers get a governance configuration checklist, teams get a copyable AGENTS.md template, and developers get executable commands for multi-agent work and PR stacks.
这个版本把 AI agent 当作仓库里的正式参与者来覆盖：怎么治理它们，怎么给它们写约定。
主要内容
- 新增 AI Agent 治理指南：agent 的三种身份形态、权限收紧、PR 审批规则、CI 触发与 secrets 隔离、可追溯性，托管 agent 的内置限制以 GitHub 官方文档为依据
- 新增 AGENTS.md 模板，可直接复制到团队仓库，含环境命令、验证要求、边界禁令、提交约定，并附各工具配置文件对照表
- 本仓库开始遵循自己倡导的约定：根目录新增 AGENTS.md，贡献指南补充 AI 辅助贡献规则，文档检查覆盖 AGENTS.md
- Multi-Agent Branch Strategy 补齐可执行流程：对比实验分支、整合胜出方案、同文件冲突处理、清理落选分支
- Stacked PR for AI-Generated Changes 补齐原生 git 栈操作（
--update-refs、rebase --onto、--force-with-lease）和 Graphite CLI 核心命令 - 以上内容均为中英双语成对维护
为什么发这个版本
v2.0 确立了面向 AI 编程时代的方向，但 agent 治理和协作约定还是空白。
AI agent 已经在创建分支、发起 PR、触发 CI，v2.1.0 把它们从"diff 生成器"升级为"被治理的参与者"：管理者有了治理配置清单，团队有了可复制的 AGENTS.md 模板，开发者有了多 Agent 并行和栈式 PR 的实操命令。