AI Skill Hub 强烈推荐:Brigade 是一款优质的Agent工作流。AI 综合评分 8.0 分,在同类工具中表现稳健。如果你正在寻找可靠的Agent工作流解决方案,这是一个值得深入了解的选择。
Brigade 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
Brigade 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:pip 安装(推荐)
pip install brigade
# 方式二:虚拟环境安装(推荐生产环境)
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install brigade
# 方式三:从源码安装(获取最新功能)
git clone https://github.com/escoffier-labs/brigade
cd brigade
pip install -e .
# 验证安装
python -c "import brigade; print('安装成功')"
# 命令行使用
brigade --help
# 基本用法
brigade input_file -o output_file
# Python 代码中调用
import brigade
# 示例
result = brigade.process("input")
print(result)
# brigade 配置文件示例(config.yml) app: name: "brigade" debug: false log_level: "INFO" # 运行时指定配置文件 brigade --config config.yml # 或通过环境变量配置 export BRIGADE_API_KEY="your-key" export BRIGADE_OUTPUT_DIR="./output"
<p align="center"> <img src="docs/assets/brigade-wordmark.svg" alt="Brigade (by Escoffier Labs)" width="640"> </p>
<p align="center"> <strong>Run agents anywhere. Keep one record of the work.</strong><br> Brigade tracks what's ready, who claimed it, and what ran across Claude Code, Codex, cloud agents, and your machines. Your local harnesses stay in sync on MCP servers, skills, and memory. </p>
<p align="center"> <a href="https://brigade.tools/docs/getting-started/install">Install</a> · <a href="https://brigade.tools/docs">Docs</a> · <a href="docs/comparison.md">Compare</a> · <a href="QUICKSTART.md">Quickstart</a> </p>
<p align="center"> <img src="https://shieldcn.dev/github/ci/escoffier-labs/brigade.svg?workflow=ci.yml&branch=main&label=ci&size=xs" alt="CI status"> <img src="https://shieldcn.dev/pypi/v/brigade-cli.svg?label=pypi&size=xs" alt="PyPI version"> <img src="https://shieldcn.dev/https/raw.githubusercontent.com/escoffier-labs/brigade/download-metrics/downloads.json.svg?size=xs" alt="Combined PyPI and npm downloads per month"> <img src="https://shieldcn.dev/badge/python-3.10+-blue.svg?logo=python&logoColor=white&size=xs" alt="Python 3.10+"> <img src="https://shieldcn.dev/badge/rust-code_graph-b7410e.svg?logo=rust&logoColor=white&size=xs" alt="Rust: code graph engine"> <img src="https://shieldcn.dev/badge/go-evidence_log-00add8.svg?logo=go&logoColor=white&size=xs" alt="Go: evidence log engine"> <img src="https://shieldcn.dev/badge/license-MIT-4e7247.svg?size=xs" alt="MIT license"> </p>
Coding work now runs in several places at once. You drive Claude Code or Codex in a terminal, a background run works another repo, a cloud agent opens a pull request in another tab, and CI checks the result somewhere else. Each tool keeps its own history. When one agent hands work to the next, the handoff is usually a sentence like "tests pass."
Agents did not create this split. Terminals, background jobs, and remote machines have always kept separate records. Running agents in parallel makes each gap cost more.
Brigade keeps one record around the work itself, across harnesses, repos, machines, and cloud seats: which tasks are ready, who claimed them, which commands ran and how they exited, and what the next session should know. It also keeps MCP servers, skills, and memory in sync, so an agent picking up the work in another harness starts with the same setup and context. The record is plain local files, with an optional hub for claims across machines. Agents read and write it through the CLI and MCP. You set policy, budgets, scopes, and gates, and can stop the work at any point.
For example, a Claude Code agent can hand a verify receipt to Codex in an evidence package, alongside a handoff that explains the change. Codex checks the package and the receipt's real exit code instead of taking the agent's word for it.
The Python CLI may send a passive version request. Python CLI telemetry lists its fields, retention, and opt-outs.
Stable: v0.30.0 (brigade-cli==0.30.0 on PyPI, @escoffier-labs/brigade@0.30.0 on npm). Beta: main is the 0.31 preview line (0.31.0.devYYYYMMDD wheels). See the changelog for what each release changed.
The downloads badge sums brigade-cli on PyPI (excluding mirrors) and @escoffier-labs/brigade on npm over the same 30 completed UTC days. Downloads include automation and do not count unique people. The metric JSON records each source, the period, and the last successful refresh. npm platform companions are excluded. Daily refreshes retain the previous metric if a source fails validation or its latest day is more than three UTC days old.
If Node.js 22 or newer and npm are installed on glibc Linux x64/ARM64, macOS x64/ARM64, or Windows x64, install the npm package. This route does not need Python or pipx first:
npm install --global @escoffier-labs/brigade@0.30.0
brigade setup
To try it without a global install, run each command through npx instead, for example npx --yes @escoffier-labs/brigade@0.30.0 --version. The package brings its own uv, through platform packages that npm installs as optional dependencies, and the published 0.30.0 wheel. Leave optional dependencies enabled. The first launch downloads a managed Python, so it needs network access. Install Git yourself for repository features. npm targets are Linux (glibc) x64 and ARM64, macOS x64 and ARM64, and Windows x64. Upgrade with npm, because brigade update refuses npm-managed installs. Platform details and known limits: npm package notes.
On other platforms, or to use Python 3.10 or newer directly:
pipx install brigade-cli # or: uv tool install brigade-cli
brigade setup
Then follow the install guide or QUICKSTART.md. To let your coding agent do the setup, hand it first 10 minutes and the agents guide.
If setup fails, run brigade bug report --summary "What happened and what you expected" --dry-run in the affected workspace. Review the complete redacted preview before sharing. See bug reports.
For preview builds on a pipx install, run brigade update --channel beta. Details: update channels.
| Project | MCP | Rules and skills | Projection | Source of truth | Audit and rollback |
|---|---|---|---|---|---|
| [Brigade](https://github.com/escoffier-labs/brigade) | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> MCP + tools catalog | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Skills + harness adapters | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Dry run first, then write | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Brigade catalog | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Preserve, gate, roll back |
| [Rulesync](https://github.com/dyoshikawa/rulesync) | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Supported | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Rules, skills, hooks, subagents | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Per-target generation | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Rulesync project | <img src="docs/assets/status-partial.svg" width="10" height="10" alt="Partial"> Generated outputs |
| [Ruler](https://github.com/intellectronica/ruler) | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Supported | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Rules, skills, subagents | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Per-agent files | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> .ruler/ directory | <img src="docs/assets/status-partial.svg" width="10" height="10" alt="Partial"> Gitignore automation, no rollback |
| [add-mcp](https://github.com/neon-solutions/add-mcp) | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> MCP servers | <img src="docs/assets/status-undocumented.svg" width="10" height="10" alt="Not documented"> Not documented | <img src="docs/assets/status-partial.svg" width="10" height="10" alt="Partial"> Maps fields into native files | <img src="docs/assets/status-partial.svg" width="10" height="10" alt="Partial"> Native files + sync | <img src="docs/assets/status-partial.svg" width="10" height="10" alt="Partial"> Warns on dropped fields |
| [agentsync](https://agentsync.cc/) | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Many agent targets | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Skills, hooks, memory, plugins | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Native/lossy/skipped report | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Canonical config | <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Native/lossy/skipped report |
Most neighbors go deeper on one layer. Brigade connects work, memory, tool config, and evidence around the same coding-agent task. Where a neighbor is stronger, its cell says so. Sources, dated notes, and projects that left these tables: docs/comparison.md (checked 2026-10-18).
Key: <img src="docs/assets/status-supported.svg" width="10" height="10" alt="Supported"> Supported · <img src="docs/assets/status-partial.svg" width="10" height="10" alt="Partial"> Partial · <img src="docs/assets/status-undocumented.svg" width="10" height="10" alt="Not documented"> Not documented
高质量的开源AI工作流项目
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ MIT 协议 — 最宽松的开源协议之一,可自由商用、修改、分发,仅需保留版权声明。
总体来看,Brigade 是一款质量优秀的Agent工作流,在同类工具中具备一定竞争力。AI Skill Hub 将持续追踪其更新动态,建议收藏备用,结合自身场景选择合适时机引入使用。
| 原始名称 | brigade |
| 原始描述 | 开源AI工作流:Brigade CLI: AI agent memory, handoffs, and local guardrails for Codex, Claude C。⭐26 · Python |
| Topics | AI工作流智能代理 |
| GitHub | https://github.com/escoffier-labs/brigade |
| License | MIT |
| 语言 | Python |
收录时间:2026-06-09 · 更新时间:2026-06-11 · License:MIT · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端