← Back to Skills Marketplace
hanshojin

Subagent Bridge / 子会话传话

by HANSHOJIN · GitHub ↗ · v1.1.0 · MIT-0
cross-platform ✓ Security Clean
58
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install subagent-bridge
Description
Bridge communication between subagents via sessions_send — pipeline, debate, broadcast, and aggregate patterns.
README (SKILL.md)

Subagent Bridge 🌉 — 子会话传话

Let different subagents (child sessions) chat with each other under the same OpenClaw agent. Tell your main session what to do in plain language — no complex commands needed.

让同一个 OpenClaw 下的不同子会话(subagent)之间互相传话。 你在聊天框里像跟人说话一样发号施令就行。


🏷️ First: Give Them Names / 先说最重要的:起名字

Every subagent that you want to bridge needs a memorable name when spawned. 每个要长期用、要互相传话的子会话,spawn 时就给它起个容易记住的名字。

You → Main:  spawn a search expert, call it 搜搜 (Soso)
  Main → You: 搜搜 (agent:main:subagent:xxx) is ready!

You → Main:  spawn an analysis expert, call it 想想 (Xiangxiang)
  Main → You: 想想 (agent:main:subagent:yyy) is ready!

The main session remembers the name ↔ session-key mapping. Later you just refer to them by name.

主会话会记住这个映射关系。后面传话你直接叫名字就行。


🗣️ How to Use / 怎么用

Say these to your main session naturally. 对主会话说人话就行。

What you want Say this 你可以说
Talk to one subagent ask 搜搜:xxx tell 想想:xxx send message to 搜搜:xxx 问 搜搜:xxx 传话给 想想:xxx 告诉 搜搜:xxx
Pipe one's output to another let 搜搜 pass its result to 想想 让 搜搜 查完结果给 想想
Two subagents debate let 搜搜 and 想想 debate about xxx 让 搜搜 和 想想 讨论一下 xxx
Broadcast to many ask 搜搜, 想想, coder: xxx 问 搜搜、想想、编程小子:xxx
Gather then summarize let 搜搜, 想想 each work on this, then send results to 汇总大师 让 搜搜、想想 各做各的,结果给 汇总大师

⚙️ How the Main Session Does It / 主会话内部做的工作

Step 1: Find the subagent / 找人

sessions_list(kinds: ["subagent"])

Match by friendly name → get sessionKey.

Step 2: Send the message / 传话

sessions_send(sessionKey="agent:main:subagent:\x3Cuuid>", message="...", timeoutSeconds=N)
  • timeoutSeconds: 0 — fire-and-forget, no reply expected
  • timeoutSeconds: >0 — wait for reply. Search tasks: 60-90s, analysis: 30-60s

Step 3: Multi-turn dialogue (A ↔ B debate / 多轮讨论)

Both sides can reply. Send REPLY_SKIP to end your turn early. Max rounds: 5 (configurable up to 20 via session.agentToAgent.maxPingPongTurns).


📐 Bridge Patterns / 四种传话模式

📊 Pipeline(流水线)

User → Main spawns 搜搜 → 搜搜's result sent to 想想 → 想想 analyzes and returns

💬 Debate(辩论)

Main: "搜搜, do analysis → pass result to 想想"
Main: "想想, critique 搜搜's result → pass critique back"
Loop until consensus or max rounds.

📢 Broadcast(广播)

Main sends the same message to 搜搜, 想想, coder simultaneously.
Aggregate all replies.

🤝 Aggregate(汇总)

搜搜, 想想, coder each work independently.
Main collects results → sends them all to 汇总大师 for a summary.

⚠️ Limits / 限制

  • sessions_send cannot target thread-scoped chat sessions.
  • Cross-agent bridging needs tools.agentToAgent.enabled: true in config.
  • Subagents are ephemeral — bridge them while they're alive.
  • sessions_send 不能发到 thread 类型的聊天会话。
  • 跨不同 agentId 的传话需要配置 tools.agentToAgent.enabled: true
  • 子会话用完就过期,要传话趁它活着的时候。
Usage Guidance
Install this only if you want your main agent to coordinate active subagents. Use explicit subagent names, avoid sending sensitive content through broadcast or debate flows, and confirm ambiguous routing requests before allowing messages or histories to be shared across sessions.
Capability Tags
crypto
Capability Assessment
Purpose & Capability
The artifacts consistently describe routing messages between active subagents using session-listing, status/history, and send-style workflows for pipeline, debate, broadcast, and aggregation patterns.
Instruction Scope
The skill encourages broad natural-language commands such as ask, tell, debate, and broadcast, which is convenient but can make message-routing intent ambiguous unless the user is explicit.
Install Mechanism
The package contains only markdown documentation files, no executable scripts, no declared dependencies, and local static scan metadata reports no suspicious findings.
Credentials
The requested session tools fit the stated bridge purpose; there is no artifact evidence of filesystem access, credential handling, network exfiltration, destructive commands, or unrelated data collection.
Persistence & Privilege
Reference examples use active subagents and cleanup="keep" so sessions remain available during coordination, but this is disclosed as part of the workflow and is bounded to user-directed subagent communication.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install subagent-bridge
  3. After installation, invoke the skill by name or use /subagent-bridge
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.1.0
中英双语 README,方便国内外用户搜索和阅读
v1.0.0
初始版本:支持 Pipeline、Debate、Broadcast、Aggregate 四种 subagent 传话模式
Metadata
Slug subagent-bridge
Version 1.1.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 2
Frequently Asked Questions

What is Subagent Bridge / 子会话传话?

Bridge communication between subagents via sessions_send — pipeline, debate, broadcast, and aggregate patterns. It is an AI Agent Skill for Claude Code / OpenClaw, with 58 downloads so far.

How do I install Subagent Bridge / 子会话传话?

Run "/install subagent-bridge" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Subagent Bridge / 子会话传话 free?

Yes, Subagent Bridge / 子会话传话 is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Subagent Bridge / 子会话传话 support?

Subagent Bridge / 子会话传话 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Subagent Bridge / 子会话传话?

It is built and maintained by HANSHOJIN (@hanshojin); the current version is v1.1.0.

💬 Comments