Notifications & Integrations

feishu-local-agent-bridge-windows

zenhatescoding/feishu-local-agent-bridge-windows · ★ 0

zenhatescoding

Personal vibecoding fork note: this is a Windows personal-use adaptation built together with Codex, not an attempt to present itself as the original project. It is based on…

License
MIT
Added
2026-08-23
Maintenance
active
Install kind
github

No manifest check is available; listing does not imply compatibility or security review.

安装

dsh plugin --profile web add github:zenhatescoding/feishu-local-agent-bridge-windows

GitHub 源码安装:首次需按提示配置 allowBuilds 构建授权后重试

For install and environment setup, plus plugin development tutorials, see DSH community docs ↗

README

Source README · body: en · Read Chinese README

Table of contents
  1. One Recommended Install
  2. Branch Roles
  3. Supported Agents
  4. Independent Or Collaborative
  5. Documentation
  6. Build Once
  7. Start A Collaborative Group
  8. Runtime Reference
  9. Safety
  10. Development

Connect Claude Code, Codex, Google Antigravity, DeepSeek Harness and Hermes to Feishu/Lark, either as independent bots or as a team collaborating in one topic.

中文说明

For a Hub deployment, clone release/hub. It contains contains all maintained bridge adapters, DeepSeek Harness setup scripts, the collaboration Hub, shared-file support and Windows process management.

git clone --branch release/hub --single-branch https://github.com/ZenHatesCoding/feishu-local-agent-bridge-windows.git `
  C:\feishu-local-agent-bridge

Branch Roles

The release/* branches are deployment baselines and develop/* branches are their corresponding work lines. All other branches are explicit archives for rollback or historical reference.

Branch Role Policy
release/hub Stable central Hub deployment Use for a Hub installation
develop/hub Hub protocol and conversation-semantics development Merge into release/hub after protocol tests pass
release/worker Stable second-PC Worker deployment; the Worker does not start a Hub Use after the real second-PC Feishu check passes
develop/worker Second-PC Worker deployment development Merge into release/worker after that acceptance check
archive/* Frozen milestones and earlier standalone packages Rollback/history only

The same checkout can build all bridge runtimes. Each bot still needs its own Feishu app/profile and local agent login. Hermes stays in its existing install and connects through the removable project Hook.

Exact setup for each agent: Agent bridge guide.

Supported Agents

Agent Bridge mode in this branch Local prerequisite
Claude Code Native claude adapter Installed and logged-in claude CLI
Codex Native codex adapter Installed and logged-in Codex CLI
Google Antigravity antigravity adapter in agy mode Interactive agy login
DeepSeek Harness Independent deepseek-harness adapter Node.js 22+ and built Harness CLI
Hermes Isolated collaboration Hook Existing Hermes installation; never reinstalled

Independent Or Collaborative

Independent bridge: one Feishu app talks to one local agent and keeps its own profile, sessions, workspaces and credentials.

Collaborative group: several bots join one group. One Feishu topic is one conversation. Agents and people can reply to one another in that topic; when a message transfers or requests work, it carries explicit work semantics. The next agent receives authorized conclusions and durable shared files without receiving private reasoning or unrelated history.

The design soul is share task state, not model mind-state. A real Feishu @ is the visible wake-up signal; a Hub attention grant (dispatch) authorizes the corresponding run. Both must exist for an agent-to-agent reply, preventing accidental fanout and wake-up loops. Ownership changes only for explicit work actions such as handoff and ask.

Read collaboration design and the product north star before changing this protocol.

Documentation

Question English 中文
What are Hub, Pilot and dispatch? Concepts 概念入门
How do I configure each agent bridge? Agent bridges Agent 桥接
What experience must the project preserve? Product vision 产品目标
How do context, routing and files work? Design 设计原理
How do I deploy and operate the group? Windows operations Windows 运维
How do computers connect securely, and what is Tailscale? Networking 多电脑联网
Can Bots run on different computers? Distributed roadmap 跨电脑路线图
What are the common Pilot commands? Pilot script summary Pilot 脚本速查

This README is the human entry point; coding agents start at AGENTS.md. Every maintained detailed document links back here and to its language counterpart.

Build Once

Set-Location C:\feishu-local-agent-bridge
corepack enable
pnpm install
pnpm build

Start one independent Claude, Codex or Antigravity bridge with a dedicated profile:

node .\dist\cli.js run --profile codex --agent codex --workspace C:\workspaces\codex
node .\dist\cli.js run --profile claude --agent claude --workspace C:\workspaces\claude
node .\dist\cli.js run --profile antigravity --agent antigravity --workspace C:\workspaces\antigravity

Prepare and bind DeepSeek Harness from the same checkout:

.\scripts\bootstrap-deepseek-bridge.ps1
.\scripts\setup-deepseek-feishu.ps1
.\scripts\start-deepseek-bridge-service.ps1

Start A Collaborative Group

.\scripts\collab-pilot\Setup-CollabPilot.ps1
notepad .\.runtime\pilot.local.json
.\scripts\collab-pilot\Test-CollabPilotConfig.ps1
.\scripts\collab-pilot\Start-CollabPilot.ps1

The repository deploys the Hub, protocol, bridge code and process management. You provide each agent installation/login, Feishu app/profile, launch command, workspace and model settings. See Windows operations before editing the manifest.

Runtime Reference

Each profile can run as a per-profile service. Windows uses Task Scheduler and a .cmd launcher. Useful commands:

lark-channel-bridge start --profile <name>
lark-channel-bridge status --profile <name>
lark-channel-bridge stop --profile <name>
lark-channel-bridge profile export <name>
lark-channel-bridge profile export <name> --include-secrets --yes
lark-channel-bridge profile remove <name>
lark-channel-bridge profile remove <name> --purge --yes

Feishu commands include /status, /config, /cd, /ws, /resume, /stop, /doctor, /invite user, /remove user, /invite group, /remove group and /invite all group.

Cloud-doc comments are document-scoped and follow document permissions. Chat access is private by default. The profile-local lark-cli directory keeps each bot's authorization separate; the lark-cli identity policy defaults to bot-only.

Workspaces use workspaces.default. Canonical permission configuration is:

{
  "permissions": {
    "defaultAccess": "full",
    "maxAccess": "full"
  }
}

The legacy sandbox setting is read for migration only.

Safety

  • Profile state and App Secrets remain local and Git-ignored.
  • The Hub listens on 127.0.0.1 by default.
  • Pilot preserves one-PC Hub-plus-all-Bots operation and also supports remote workers with per-Agent authentication over a private network. Automatic artifact retrieval is tracked in the distributed roadmap.
  • Collaboration visibility is protocol isolation, not OS isolation.
  • Hermes is not reinstalled or upgraded; only the named Hook is added/removed.
  • Stop-CollabPilot.ps1 -RestoreOriginals restores configured independent bridge launchers without deleting shared task artifacts.

Development

pnpm test
pnpm typecheck
pnpm build

Based on zarazhangrui/lark-coding-agent-bridge and distributed under the original MIT license.