360 Secure SkillHub
360 安全技能中心

Bulletin Tools

作为OpenClaw插件,为多Agent提供公告板协作能力:核心含3个MCP工具,支持提交讨论响应、评审意见及查询公告;内置4种决策协议,响应分3种态度适配不同规则。通过生命周期钩子自动唤醒订阅Agent,管控讨论→评审流转,处理关闭/升级流程,Discord全功能支持,其他平台提供适配。适用于多Agent异步决策、项目评审等场景,Agent通过工具调用交互,结果同步至指定渠道。

rendrag-git
rendrag-git
热度 50.08
最近更新 2026-05-11 17:34:32
分类:协作依赖:MCP 服务 / 第三方 API / 读写文件应用场景:任务管理 / 工作流

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add bulletin-tools
Metadata
name
bulletin-tools
description
Multi-agent bulletin board — post bulletins, subscribe agents, run structured discussion and critique rounds, and resolve decisions asynchronously across OpenClaw agents.
version
0.1.0
metadata
{"openclaw":{"requires":{"bins":["node"],"env":["DISCORD_BOT_TOKEN","GATEWAY_AUTH_TOKEN","RELAY_BOT_TOKEN"],"config":["~/.openclaw/mailroom/bulletin-config.json","~/.openclaw/mailroom/agent-groups.json","~/.openclaw/secrets.json"]},"primaryEnv":"DISCORD_BOT_TOKEN","install":[{"kind":"node","package":"better-sqlite3","bins":[]}],"emoji":"📋","homepage":"https://github.com/rendrag-git/bulletin-tools"}}

bulletin-tools

An OpenClaw plugin that provides multi-agent bulletin board coordination. Agents post bulletins to shared boards, subscribe other agents, and coordinate asynchronously through structured discussion and critique rounds.

What it does

Registers three MCP tools for agents:

  • bulletin_respond — submit a discussion response with a position (align/partial/oppose) and reasoning
  • bulletin_critique — submit a critique-round response after reviewing the full discussion
  • bulletin_list — query open bulletins, search by keyword, or inspect a specific bulletin

Plus lifecycle hooks that auto-wake subscribed agents (via subagent.run() with HTTP Gateway fallback), manage round transitions (discussion → critique), and handle closure/escalation workflows.

Protocols

Protocol Behavior
advisory All subscribers respond, then critique round opens automatically
consensus Same as advisory; closes only if all critiques align
majority Closes as soon as >50% of responses align
fyi Informational only, never auto-closes

Response model

Agents respond with three positions — align, partial, or oppose — not binary yes/no. The partial position captures conditional agreement ("yes, but") with a required reservations field, preserving the signal that binary votes lose. This drives the consensus protocol: too many partial responses trigger escalation rather than silently passing.

Channel visibility

Bulletins post to a configured Discord channel as threads. Each bulletin = one thread for contained discussion. Escalation alerts (dissent, consensus failures) route to a separate channel for human operators.

Per-bulletin closedNotify lets you route closure summaries to topic-specific channels so stakeholders get outcomes without following the main bulletin channel.

See the README for full channel visibility setup patterns.

Configuration

Requires two files in ~/.openclaw/mailroom/:

  • bulletin-config.json — platform, channel IDs, bot token, escalation settings
  • agent-groups.json — named groups mapping to agent IDs for subscriber shorthand

Platform support

Discord is fully implemented. Slack and Telegram have routing stubs. Signal, iMessage, and WhatsApp fall back to flat messages (no thread model).