ai-agent-workflowGroup
AGENTS.md18 starsChanged 5 months agoMITView on GitHub
平台中立的 agent 协作入口。Codex CLI 与其他 harness 读本文件;Claude Code 读 CLAUDE.md。 两端共享主源:agents/ / skills/ / docs/ / scripts/。 aiGroup is a dual-harness AI team collaboration framework. The main agent decomposes requirements, assembles an Agent Team, and delegates to specialist agents. It does NOT directly write code, design UI, or run tests. 源池 agents/<name>.md 共 52 个,由 aigroup init 按 manifest 模块选择安装: 详细派遣矩阵见 docs/rules/agents.md。 agent 体系两端共享一份——不存在"Claude 专项 agent vs Codex role adapter"对照表。Codex 端在主对话中加载对应 skill 即可达成同样效果。 .codex/agents/ 下的 3 个 persona TOML 是…
Files
AGENTS.md
平台中立的 agent 协作入口。Codex CLI 与其他 harness 读本文件;Claude Code 读
CLAUDE.md。 两端共享主源:agents//skills//docs//scripts/。
Project Overview
aiGroup is a dual-harness AI team collaboration framework. The main agent decomposes requirements, assembles an Agent Team, and delegates to specialist agents. It does NOT directly write code, design UI, or run tests.
Architecture
| 目录 | 内容 |
agents/ | agent 源池(52 个,开发态分发位置) |
skills/ | skill 源池(扁平 skills/<name>/SKILL.md) |
commands/ | 斜杠命令源 |
.claude/agents/ | Claude Code 实际启用的 agent(由 aigroup init 按需复制) |
.codex/agents/ | Codex 原生 persona TOML(3 个:explorer / reviewer / docs_researcher) |
.codex/config.toml | Runtime + MCP + persona 配置 |
docs/ | 知识库主源(架构、规则、流程) |
docs/rules/ | 强制规则集(agents / coding-style / git-workflow / testing / security / performance / hooks) |
manifests/ | 安装清单(agent / skill 分组) |
.orchestration/<session>/ | Agent 协作产物工作区 |
scripts/orchestration/ | 协调 CLI(session.cjs) |
scripts/hooks/ | hook dispatcher + checks(Codex 端需手动调用) |
Available Agents
源池 agents/<name>.md 共 52 个,由 aigroup init 按 manifest 模块选择安装:
| 模块 | 默认 | 数量 | 说明 |
agents-core | ✅ | 12 | 派遣规则定义的 10 个 + init-architect / get-current-datetime |
agents-quality | ⏤ | 9 | 代码质量分析 |
agents-language | ⏤ | 16 | 各语言 reviewer / build-resolver |
agents-ops | ⏤ | 8 | 运维与工作流操作 |
agents-domain | ⏤ | 9 | 特殊领域(a11y / healthcare / opensource / gan-* / seo) |
详细派遣矩阵见 docs/rules/agents.md。
agent 体系两端共享一份——不存在"Claude 专项 agent vs Codex role adapter"对照表。Codex 端在主对话中加载对应 skill 即可达成同样效果。
Codex 端独有
.codex/agents/ 下的 3 个 persona TOML 是 Codex 原生 multi-agent 机制,仅在需要切换 sandbox / reasoning_effort 时使用:
| Persona | 用途 |
explorer | 只读 sandbox,证据收集 |
reviewer | 高 reasoning,代码 + 安全审查 |
docs_researcher | 带网络访问,API / 文档查证 |
切换:/agent <persona>。完整 Codex 端说明见 .codex/AGENTS.md。
知识库地图
| 需要了解 | 查阅 |
| Agent 派遣矩阵 | docs/rules/agents.md |
| 强制规则(铁律 / 编码 / Git / 测试 / 安全 / 性能 / Hooks) | docs/rules/ |
| Codex 端差异(模型 / MCP / persona / 无 hook 兜底) | .codex/AGENTS.md |
| 工作流 phase 心智模型 | docs/workflow-pipeline.md |
| 危险信号 | docs/red-flags.md |
| 项目实例上下文 | docs/PROJECT_CONTEXT.md |
Codex 限制(必须知道)
- No hooks — Codex 不支持 hook 事件;必须主动运行
node scripts/hooks/dispatcher.cjs stop - No slash commands — 用主对话 +
/agent <persona>,不识别.claude/commands/ - No
Agent({ subagent_type })语法 — 派遣 = 主对话加载对应 skill;persona 切换 =/agent
详见 .codex/AGENTS.md。
Claude Code 兼容
Claude Code 用户使用 CLAUDE.md 作为入口,由主会话按 Agent Team 协议派遣安装在 .claude/agents/ 下的 project-level agents。两端的派遣规则都以 docs/rules/agents.md 为唯一事实源。
In this repository83 other files it gives its agents.
- .codex/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- source-command-fix-buildSkill
- fine-tuning-expertSkill
- ml-pipelineSkill
- rag-architectSkill
- spark-engineerSkill
- api-designerSkill
- architecture-designerSkill
- code-documenterSkill
- debugging-wizardSkill
- feature-forgeSkill
- fullstack-guardianSkill
- graphql-architectSkill
- legacy-modernizerSkill
- microservices-architectSkill
- database-optimizerSkill
- postgres-proSkill
- sql-proSkill
- csharp-developerSkill
- dotnet-core-expertSkill
- angular-architectSkill
- flutter-expertSkill
- nextjs-developerSkill
- react-expertSkill
- react-native-expertSkill
- senior-frontendSkill
- ui-ux-pro-maxSkill
- vue-expert-jsSkill
- vue-expertSkill
- cpp-proSkill
- golang-proSkill
- rust-engineerSkill
- cli-developerSkill
- cloud-architectSkill
- devops-engineerSkill
- kubernetes-specialistSkill
- mcp-developerSkill
- monitoring-expertSkill
- sre-engineerSkill
- terraform-engineerSkill
- websocket-engineerSkill
- java-architectSkill
- kotlin-specialistSkill
- javascript-proSkill
- nestjs-expertSkill
- typescript-proSkill
- spring-boot-engineerSkill
- spring-boot-engineer_znSkill
- laravel-specialistSkill
- php-proSkill
- competitive-analysisSkill
- meeting-notesSkill
- prd-templateSkill
- stakeholder-updateSkill
- user-research-synthesisSkill
- django-expertSkill
- fastapi-expertSkill
- pandas-proSkill
- python-proSkill
- The first 60 shown.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

