项目介绍
Agent 开发框架 · 2026-11-09 更新 · 本站尚未实测
Cortex 的定位是“给你的 Agent 配一位幕僚长”。它由两部分组成:一套用纯 Markdown 文件(含 YAML frontmatter)协调多个 AI Agent 的开放协议,以及一个随仓库提供的 Claude Code 插件。README 说明该协议与运行时无关,Codex、Cursor、Gemini CLI 或任何能读写文件的 Agent 都可参与。
其协作模型为:用户 → 幕僚长(总协调 Agent)→ 团队目录(Markdown 文件)→ 工作 Agent。幕僚长接收指令、把任务写入各 Agent 的任务文件,并监控进度;工作 Agent 在各自项目中轮询任务文件、执行任务并回报状态。README 明确指出 Agent 之间不做直接通信,所有协调都经由共享 Markdown 文件完成,因此各 Agent 不需要同时在线。
团队目录可以是一个 Obsidian 库、一个 git 仓库或任意目录。README 建议将团队目录做成 git 仓库,以便每日复盘清理旧数据后仍能通过 git log 追溯历史。
核心能力
点击能力名称查看输入、输出和原文依据。
- 任务派发与轮询:幕僚长把任务写入
agents/<slug>/tasks.md,工作 Agent 按心跳周期读取状态为ready的任务,改为in-progress并附开始时间,完成后置为done并附完成时间与小结。 - 状态回流:Agent 在任务文件中更新
<!-- cortex:last-tick -->存活时间戳,幕僚长据此识别疑似离线的 Agent;进度通过 Agent 笔记中的## Session Log汇总。 - 插件技能:提供
/setup(首次初始化)、/register <name>(注册新 Agent)、/join <name>(接入并生成.cortex.md协议文件)、/leave <name>(退出)四个技能。 - 每日简报与复盘:支持配置每日简报与复盘时间,清理超过 7 天的已完成任务,并裁剪会话日志。
- 多运行时混编:README 举例称 Claude Code 幕僚长协调 Codex 与 Gemini 工作 Agent 可开箱工作,并附有 Codex、Cursor、Gemini CLI、OpenCode 及通用适配示例文档。
实际应用场景
适用场景
- 个人或小团队希望用一个“总控”Agent 跨多个项目分派编码、文档类任务,并保留可读、可版本化的任务记录。
- 已使用 Claude Code、Codex、Cursor、Gemini CLI 等终端或编辑器 Agent,希望以文件为中介做轻量编排。
- 偏好纯文本、无服务器的协作方式,不要求实时双向通信,任务可延迟数小时被领取。
安装与接入
前置条件根据 README 的快速开始,需要已安装 Claude Code 的插件机制。仓库发布日期、版本号等未在给定资料中说明,待核验。
从本地克隆安装:
claude plugins marketplace add /path/to/cortex --scope user
claude plugins install cortex@agentweave
从 GitHub 安装:
claude plugins marketplace add https://github.com/agentweave/cortex --scope user
claude plugins install cortex@agentweave
安装后启动 Claude Code 运行 /setup,按提示填写团队目录路径(默认 ~/cortex-team)、Telegram chat ID(可选)、心跳间隔与每日时间。随后在幕僚长项目目录运行 /join Chief of Staff,再用 /register <名称> 注册工作 Agent,并在其项目目录运行对应的 /join。
配置文件为 ~/.cortex/config.yaml,包含 team_dir、heartbeat_minutes、daily_briefing、daily_review、telegram_chat_id、chief_of_staff_project 等字段。非 Claude Code 运行时需参考仓库内协议规范与适配示例,具体步骤未在给定 README 中展开,待核验。
使用条件与限制
- 是否需要部署
- 需自行准备运行环境,具体要求见正文
- 是否需要注册
- 账号和认证要求见正文;未明确说明的项目仍待核验
- 是否需要 API Key
- 待核验
- 费用与额度
- 费用边界见正文,项目许可不代表模型或托管服务免费
- 所需权限
- 待核验
- 许可证与服务条款
- 待核验
权限费用与限制
- Cortex 本轮资料未给出任何价格信息,是否收费待向官方核验;公开仓库不等于其所依赖的 Claude Code、Codex、Cursor、Gemini CLI 等服务免费。
- 各 Agent 运行时所需的模型或 API Key、账号权限、调用额度由对应平台决定,Cortex 仓库资料未说明,待核验。
- Telegram 通知依赖单独的 Claude Code Telegram 插件,Cortex 只保存 chat ID;README 说明不使用 Telegram 时,幕僚长可通过终端或远程控制正常工作。
- 协议本身不提供鉴权、加密或隔离机制,团队目录的访问权限取决于所在文件系统与 git 仓库设置,相关资料未涉及安全模型,待核验。
- 任务文件格式参考 Shuttle Protocol,两者兼容性程度未在资料中说明。
国内使用情况
网友核验 用户经验分享
与站长实测独立,不代表官方认证或本站已实测。
最近核验:暂无有效核验。参与人数按当前项目有效身份去重;不同核验项分别统计,不合并成用户总结果。仅统计审核通过且未过期的记录。
分项统计、有效期与样本量
| 核验项目 | 有效人数 | 成功 / 部分 / 失败 / 无法判断 | 成功率 | 最近核验 / 有效期 |
|---|---|---|---|---|
| 官方网站访问 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 7 天 |
| GitHub 仓库访问 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 7 天 |
| 安装与运行 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 30 天 |
| 官方服务国内连接 / MCP 连接 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 30 天 |
| 国内网络情况 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 7 天 |
| 国内模型兼容性 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 30 天 |
| 自定义 API | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 30 天 |
| 海外服务依赖 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 30 天 |
| 海外账号要求 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 30 天 |
| 中文界面 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 30 天 |
| 客户端兼容性 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 30 天 |
| 账号及费用信息 | 0 | 0 / 0 / 0 / 0 | 样本不足 | 暂无 / 30 天 |
成功率 = 成功 ÷(成功 + 部分成功 + 失败);无法判断不计入分母。至少 5 份可判断结果才显示成功率。账号、费用、海外依赖与语言等事实项,成功表示核查结论与资料吻合,具体观察另行统计;不表示免费或没有海外依赖。不同网络、客户端和版本的结果不能直接等同。
最近有效核验
暂无审核通过且未过期的网友核验。欢迎提交真实使用记录。
查看全部有效核验 →国内实测状态:国内可用性待核验。协议兼容和官方资料说明不等于国内网络可用。
- 官方网站访问
- 尚未实测 待核验
- GitHub 仓库访问
- 尚未实测 待核验
- 官方服务国内连接
- 尚未实测 待核验
- 国内网络情况
- 尚未实测 待核验
- 国内模型兼容性
- 尚未实测 待核验
- 自定义 API
- 尚未实测 待核验
- 海外服务依赖
- 尚未实测 待核验
- 海外账号要求
- 尚未实测 待核验
- 中文界面
- 尚未实测 待核验
- 国内最后核验时间
- 尚未实测
许可与国内使用
仓库 LICENSE 文件为 MIT 许可证,覆盖仓库代码。README 未提及商业模块、托管服务或第三方模型的授权安排,这些部分需单独核验。
资料中未说明国内网络环境下的可用性、是否支持国内模型或国内客户端,也未说明服务部署地域;相关结论均待核验,不能据此推断兼容性。原文仅列出 Codex、Cursor、Gemini CLI、OpenCode 等运行时示例。
项目资料与来源
字段依据与待核验事项
access:来源确认 · claude plugins marketplace add https://github.com/agentweave/cortex --scope user claude plugins install cortex@agentweave
project_type:来源确认 · Cortex is an open protocol for coordinating AI agents through plain markdown files.
原始 README(折叠技术参考)
# Cortex
Give your agents a chief of staff.
Cortex is an open protocol for coordinating AI agents through plain markdown files. It ships with a [Claude Code](https://docs.anthropic.com/en/docs/claude-code) plugin, but the protocol works with any agent runtime — Codex, Cursor, Gemini CLI, or anything that can read and write files.
A **chief of staff** agent manages **worker agents** across projects. The chief of staff receives your instructions, dispatches work to agents, and monitors progress — all through plain markdown with YAML frontmatter.
**Not using Claude Code?** See the [protocol spec](docs/protocol.md) and [adapter examples](docs/adapters/) for your runtime.
## How It Works
```
You --> Chief of Staff (Claude Code) --> Team Directory (markdown files) <-- Worker Agents (Claude Code)
```
- **Team directory** — a shared folder of markdown files. Agent notes, project notes, and task files. Can be an Obsidian vault, a git repo, or any directory.
- **Chief of staff** — a coordinator agent that receives your instructions, dispatches tasks to agent task files, and sends daily briefings. Talk to it via Telegram (recommended), terminal, or remote control.
- **Worker agents** — project-specific agents that poll their task file for work, execute tasks, and report status.
- **No direct agent-to-agent communication.** All coordination happens through the shared markdown files.
## Quick Start
**1. Install the plugin**
From a local clone:
```bash
claude plugins marketplace add /path/to/cortex --scope user
claude plugins install cortex@agentweave
```
From GitHub:
```bash
claude plugins marketplace add https://github.com/agentweave/cortex --scope user
claude plugins install cortex@agentweave
```
The marketplace name (`agentweave`) comes from `.claude-plugin/marketplace.json` in the repo. This file is required for the plugin system to discover and install the plugin.
**2. Run setup**
In your terminal, start Claude Code and run:
```
/setup
```
This creates your team directory, config, and chief of staff agent note. It will ask for:
- Team directory path (default: `~/cortex-team`)
- Telegram chat ID (optional — for mobile access via the Telegram plugin)
- Heartbeat interval and daily schedule times
**3. Join as chief of staff**
In the chief of staff's project directory:
```
/join Chief of Staff
```
This generates `.cortex.md` (the protocol file), updates `CLAUDE.local.md`, and automatically starts the heartbeat (polls for work every 15 min). The heartbeat restarts on each session when `/join` runs.
**4. Register a worker agent**
As the chief of staff (or in any session):
```
/register Billing Dev
```
Then in the worker's project directory:
```
/join Billing Dev
```
## What You Get
When an agent runs `/join`, two files are created in the project:
**`.cortex.md`** — the full protocol that tells the agent how to operate:
```markdown
# Cortex Protocol
You are **Billing Dev**, a member of Cortex — a coordinated team of AI agents.
## Your Identity
- Agent note: agents/billing-dev/billing-dev.md in the team directory
- Task file: agents/billing-dev/tasks.md
- Project: ~/Projects/billing-service
- Team directory: ~/cortex-team
## Protocol
### Checking for Work
- Read your task file for tasks with **Status:** ready
- Mark in-progress with **Started:** timestamp, do the work, mark done with **Completed:** timestamp
### Heartbeat
- Poll every 15 minutes for new work
...
```
**`CLAUDE.local.md`** — a one-line reference so the agent auto-syncs on session start (untracked, agent-specific):
```markdown
## Cortex
On session start, run /join Billing Dev to sync with Cortex. See .cortex.md for full protocol.
```
## How Agents Communicate
Agents never talk to each other directly. All coordination flows through the team directory — a shared folder of markdown files that acts as a message bus.
- **Chief of staff writes tasks** to agent task files (`agents/<slug>/tasks.md`)
- **Workers poll** their task file on a heartbeat, pick up `ready` tasks, and report `done`
- **Liveness timestamps** — agents update `<!-- cortex:last-tick -->` in their task file. Chief detects stale agents with active tasks as likely down.
- **Status flows back** through agent notes' `## Session Log` — the chief of staff reads these to monitor progress
This means agents don't need to run simultaneously. A worker can pick up tasks hours after the chief of staff wrote them.
### Housekeeping
The team directory should be a git repo. During the daily review, the chief of staff prunes old data:
- **Done tasks** older than 7 days are deleted from task files
- **Session logs** are trimmed to only the latest entry per agent
- **Git preserves history** — pruned content lives in `git log` if you ever need it
## Task Files
Each agent has a task file at `agents/<slug>/tasks.md`. The format follows the [Shuttle Protocol](https://github.com/agentweave/shuttle/blob/main/docs/PROTOCOL.md).
**Status lifecycle:** `ready` → `in-progress` → `done`
A ready task:
```markdown
## Add Stripe webhook endpoint for payment failures
**Status:** ready
Create POST /webhooks/stripe, verify signature, handle payment_intent.payment_failed events.
```
When a worker picks it up:
```markdown
## Add Stripe webhook endpoint for payment failures
**Status:** in-progress
**Started:** 2026-03-28T10:15
Create POST /webhooks/stripe, verify signature, handle payment_intent.payment_failed events.
```
When complete:
```markdown
## Add Stripe webhook endpoint for payment failures
**Status:** done
**Started:** 2026-03-28T10:15
**Completed:** 2026-03-28T10:42
Create POST /webhooks/stripe, verify signature, handle payment_intent.payment_failed events.
### Summary
Implemented POST /webhooks/stripe with signature verification. Handles payment_intent.payment_failed by updating subscription status to past_due. Added tests.
```
## Skills Reference
### /setup
First-time Cortex setup. Creates the team directory structure, writes `~/.cortex/config.yaml`, and registers the chief of staff.
### /register \<name\>
Register a new agent. Creates an agent subfolder in `<team_dir>/agents/<slug>/` with the agent note and an empty task file. Asks for the agent's project path, role, and capabilities.
### /join \<name\>
Onboard an agent. Reads the agent note from the team directory, generates `.cortex.md` in the current project with the full protocol, and updates `CLAUDE.local.md`. Idempotent — safe to re-run as a sync.
For the chief of staff, generates an extended protocol with coordinator duties (dispatch, monitoring, daily briefing/review).
### /leave \<name\>
Offboard an agent. Removes `.cortex.md`, cleans `CLAUDE.local.md`, and sets agent status to inactive in the team directory.
## Configuration
`~/.cortex/config.yaml`:
```yaml
team_dir: ~/cortex-team
heartbeat_minutes: 15
daily_briefing: "09:00"
daily_review: "18:00"
telegram_chat_id: "your-chat-id"
chief_of_staff_project: "~/Projects/chief-of-staff"
```
Telegram is handled by the separate [Claude Code Telegram plugin](https://docs.anthropic.com/en/docs/claude-code). Cortex only needs the `telegram_chat_id` to know where to send messages. If you skip Telegram, the chief of staff works fine via terminal or remote control.
## Team Directory Structure
```
~/cortex-team/
agents/
chief-of-staff/
chief-of-staff.md
tasks.md
billing-dev/
billing-dev.md
tasks.md
dashboard-dev/
dashboard-dev.md
tasks.md
projects/
billing-service.md
dashboard.md
templates/
agent-template.md
project-template.md
```
## Slug Derivation
Agent names are converted to slugs for file naming: lowercase, spaces replaced with hyphens, special characters stripped.
| Name | Slug | Files |
|------|------|-------|
| Billing Dev | billing-dev | `agents/billing-dev/billing-dev.md` + `tasks.md` |
| Chief of Staff | chief-of-staff | `agents/chief-of-staff/chief-of-staff.md` + `tasks.md` |
| Dashboard Dev | dashboard-dev | `agents/dashboard-dev/dashboard-dev.md` + `tasks.md` |
## Plugin Structure
```
cortex/
.claude-plugin/
plugin.json — plugin metadata
marketplace.json — marketplace definition (required for installation)
skills/
setup/ — first-time setup
join/ — agent onboarding
leave/ — agent offboarding
register/ — agent registration
README.md
LICENSE
```
## Other Runtimes
Cortex is runtime-agnostic. The [protocol spec](docs/protocol.md) defines the team directory format, agent notes, task files, and coordination rules. Any agent that can read and write files can participate.
Adapter examples:
- [Codex](docs/adapters/codex.md)
- [Cursor](docs/adapters/cursor.md)
- [Gemini CLI](docs/adapters/gemini-cli.md)
- [OpenCode](docs/adapters/opencode.md)
- [Generic (any agent)](docs/adapters/generic.md)
A mixed team — Claude Code chief of staff coordinating Codex and Gemini workers — works out of the box. The team directory is the universal interface.
## License
MIT
历史前置条件说明(补充参考)
需已安装 Claude Code 及其插件机制;团队目录建议为 git 仓库;Telegram 通知需另装 Claude Code Telegram 插件;各 Agent 所需模型、API Key、账号与额度待核验;是否收费待核验;国内可用性待核验。
项目来源与清单版权许可
https://github.com/agentweave/cortex
整理来源:https://github.com/slavakurilyak/awesome-ai-agents
清单版权与许可:
MIT License
Copyright (c) 2024 Slava Kurilyak
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.