← 返回 Skills 市场
dolutech

AgentLink Protocol

作者 dolutech · GitHub ↗ · v0.1.1 · MIT-0
cross-platform ⚠ suspicious
249
总下载
1
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install agent-link
功能描述
Enables OpenClaw agents to communicate P2P using AgentLink protocol with messaging, contact management, status checks, and identity sharing.
使用说明 (SKILL.md)

AgentLink Skills

Skills for OpenClaw agents to use the AgentLink protocol

What are Skills?

Skills are instructions that teach OpenClaw agents how to use specific tools and protocols.

Available Skills

1. agentlink/

File: skills/agentlink/SKILL.md

Purpose: Teaches OpenClaw agents to communicate via P2P using the AgentLink protocol.

Capabilities:

  • Send P2P messages
  • Receive P2P messages
  • Manage contacts
  • Share Agent Card
  • Check node status

Tools:

  • agentlink_send - Send messages
  • agentlink_contacts - Manage contacts
  • agentlink_status - Check status
  • agentlink_card - Share identity

How to Install for OpenClaw

Option 1: Via npm

npm install @dolutech/agent-link

Option 2: Git Submodule

git submodule add https://github.com/dolutech/agent-link skills/agentlink

Option 3: Direct Download

# Create skills directory
mkdir -p skills/agentlink

# Download skill
curl -o skills/agentlink/SKILL.md \
  https://raw.githubusercontent.com/dolutech/agent-link/main/skills/agentlink/SKILL.md

How to Use in OpenClaw

1. Add Skill to Agent

In your OpenClaw agent configuration:

skills:
  - path: ./skills/agentlink/SKILL.md
    name: AgentLink Protocol

2. Initialize Tools

import { AgentLinkNode } from "@dolutech/agent-link";

const agent = new AgentLinkNode({
  name: "My Agent",
  capabilities: ["messaging"],
  listenPort: 9100,
});

await agent.start();

3. Use Commands

Once configured, the agent can use:

@agentlink_send to:did:key:z6Mk... message:"Hello!"
@agentlink_contacts list
@agentlink_card show
@agentlink_status

Trust Levels

The skill includes a trust system to control what other agents can do:

Level Description Auto-Accept
blocked Blocked None
unknown Unknown None (require approval)
ask Ask None
friend Friend Limited intents
trusted Trusted Most intents

Complete Usage Example

Step 1: Initialize

npx @agentlink/cli init --name "My Assistant"

Step 2: Start

npx @agentlink/cli start

Step 3: Share Contact

@agentlink_card format:link
# Returns: agentlink://eyJkaWQiOiJkaWQ6a2V5...

Step 4: Add Contact

@agentlink_contacts action:add did:did:key:z6Mk... name:"Alice" trustLevel:friend

Step 5: Send Message

@agentlink_send to:Alice intent:messaging.send message:"Hello Alice!"

Available Intents

The skill supports the following intents:

Messaging

  • messaging.send - Send message
  • messaging.receive - Receive message

Scheduling

  • scheduling.create - Create event
  • scheduling.read - Read calendar

Files

  • files.read - Read file
  • files.write - Write file

Web

  • web.fetch - Fetch URL
  • web.search - Search web

Handshake

  • handshake.hello - Initiate connection
  • handshake.ack - Acknowledge connection

Security Best Practices

✅ Recommendations

  1. Start with trustLevel: ask for new contacts
  2. Use structured data in messages
  3. Check status before sending
  4. Review permissions before approving

❌ What to Avoid

  1. Never share your private key
  2. Don't use trustLevel: trusted without verifying the agent
  3. Don't approve requests without reviewing
  4. Don't ignore security warnings

Troubleshooting

Agent cannot find the skill

# Check path
ls -la skills/agentlink/SKILL.md

# Check permissions
chmod 644 skills/agentlink/SKILL.md

Tools not available

# Check installation
npm list @dolutech/agent-link

# Reinstall if needed
npm install @dolutech/agent-link

Node is not running

# Check status
@agentlink_status

# Restart if needed
npx @agentlink/cli start

Contributing

To add new capabilities or tools to the skill:

  1. Edit skills/agentlink/SKILL.md
  2. Add usage examples
  3. Document parameters and return values
  4. Test with OpenClaw agents

Links


DoluTech © 2026

安全使用建议
This skill appears to describe a legitimate P2P messaging protocol, but before installing you should: 1) verify the upstream sources (check that the npm package @dolutech/agent-link and the GitHub repo are from the same, reputable maintainer and review their code); 2) review what the @dolutech/agent-link CLI actually does (npx runs arbitrary code); 3) avoid running installs on a production agent — test in a sandbox or isolated environment; 4) be cautious with granting trustLevel: trusted to unknown contacts and do not store secrets in places the skill can read; 5) if you cannot audit the package/repo, treat this as potentially risky and prefer not to install.
功能分析
Type: OpenClaw Skill Name: agent-link Version: 0.1.1 The AgentLink skill bundle defines a P2P communication protocol that includes high-risk intents such as 'files.read', 'files.write', and 'web.fetch' (SKILL.md). While these capabilities are documented as part of the protocol's functionality and are managed by a trust-level system, they allow for remote interaction with the local file system and network, which could be exploited if trust is mismanaged. No evidence of intentional malice or data exfiltration was found.
能力评估
Purpose & Capability
The SKILL.md describes a P2P messaging/contact/status skill and the listed tools and examples (agentlink_send, agentlink_contacts, etc.) are consistent with that purpose. However the registry metadata lists 'source: unknown' and no homepage, while the SKILL.md points to a GitHub repo and npm package; that mismatch is unexpected and worth verifying (who maintains @dolutech/agent-link and does the GitHub repo match the published npm package?).
Instruction Scope
The instructions encourage running npm/npx commands and provide CLI examples and raw curl download steps. The SKILL.md also defines intents that include files.read/files.write and web.fetch/web.search — capabilities that let the agent read/write local files and fetch arbitrary URLs. While file transfer and link preview are plausible for a messaging protocol, these capabilities broaden what the skill can do at runtime and could be used to access or exfiltrate data if the underlying implementation is untrusted. The SKILL.md does not limit or validate which paths/URLs will be used.
Install Mechanism
There is no install spec in the registry (the skill is instruction-only), but SKILL.md recommends installing via npm (@dolutech/agent-link), adding a git submodule from GitHub, or curl-download of a raw SKILL.md. npm and GitHub are common distribution channels; npm installs (and npx) execute third-party code and carry moderate risk if the package or maintainer is untrusted. The curl/raw download points to GitHub raw (better than an arbitrary IP/shortener), but fetching code during install still requires vetting the upstream source.
Credentials
The skill declares no required environment variables or credentials, which is proportionate. However the presence of file and web intents means the skill (or the installed package) could access local files or remote endpoints — this is not the same as asking for secrets up front but increases the attack surface if the package is untrusted. No explicit request for unrelated cloud credentials or tokens is present.
Persistence & Privilege
The skill is not forced-always, does not declare elevated persistence, and does not request modifications to other skills or system-wide agent settings in the SKILL.md. Autonomous invocation is allowed (default) but that is platform-normal and not by itself a red flag.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install agent-link
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /agent-link 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v0.1.1
AgentLink Skill 1.0.0 initial release - Introduces skills for OpenClaw agents to use the AgentLink P2P protocol. - Provides tools to send/receive messages, manage contacts, share agent cards, and check node status. - Includes setup and usage instructions for OpenClaw integration. - Documents available intents (messaging, scheduling, files, web, handshake) and their commands. - Outlines trust levels and security best practices. - Adds troubleshooting steps and contribution guidelines.
元数据
Slug agent-link
版本 0.1.1
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

AgentLink Protocol 是什么?

Enables OpenClaw agents to communicate P2P using AgentLink protocol with messaging, contact management, status checks, and identity sharing. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 249 次。

如何安装 AgentLink Protocol?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install agent-link」即可一键安装,无需额外配置。

AgentLink Protocol 是免费的吗?

是的,AgentLink Protocol 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

AgentLink Protocol 支持哪些平台?

AgentLink Protocol 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 AgentLink Protocol?

由 dolutech(@dolutech)开发并维护,当前版本 v0.1.1。

💬 留言讨论