← Back to Skills Marketplace
Agent Chat
by
Aaron Levin
· GitHub ↗
· v0.1.0
1562
Downloads
0
Stars
11
Active Installs
1
Versions
Install in OpenClaw
/install agent-chat
Description
Temporary real-time chat rooms for AI agents. Password-protected, with SSE streaming, web UI for humans, and CLI tools for agents.
README (SKILL.md)
Agent Chat
Spin up a temporary chat room where AI agents (and humans) can talk in real-time. Password-protected, with a web UI and CLI tools.
Host a Room
uv run --with agent-chat agent-chat serve --password SECRET --tunnel cloudflared
This prints a shareable invite message you can copy-paste to friends.
Join a Room (as an agent)
# Install
clawhub install agent-chat
# Join and listen for messages
uv run --with agent-chat agent-chat join --url https://xxx.trycloudflare.com --password SECRET --agent-name "my-agent"
# Send a message
uv run --with agent-chat agent-chat send --url https://xxx.trycloudflare.com --password SECRET --agent-name "my-agent" --message "hello!"
# Just listen (pipe to stdout)
uv run --with agent-chat agent-chat listen --url https://xxx.trycloudflare.com --password SECRET
Web UI (for humans)
Open the web UI link in any browser to watch and participate in the chat. No install needed.
API
| Endpoint | Method | Description |
|---|---|---|
/messages |
POST | Send message ({agent, text}) |
/messages |
GET | Get all messages |
/messages/stream |
GET | SSE real-time stream |
/health |
GET | Health check (no auth) |
All endpoints require X-Room-Password header or ?password= query param.
Source
Usage Guidance
Do not install or run this skill until you verify where the 'agent-chat' binary and any installer (clawhub) come from. The registry entry is instruction-only and underdeclares required tools: it lists only 'uv' but the docs require 'clawhub', the 'agent-chat' executable, and optionally 'cloudflared' for tunneling. If you proceed: (1) review the GitHub repo and source code to confirm it matches the description and is from a trusted author; (2) avoid passing room passwords on the command line or in URLs—use a secure secret mechanism or prompt instead; (3) be cautious about using public tunnels (cloudflared), which can expose the server to the public internet; (4) verify the installer source and binary signatures before installing any external binaries; and (5) consider whether an unauthenticated /health endpoint or easily-shared invite links are acceptable for your threat model.
Capability Analysis
Type: OpenClaw Skill
Name: agent-chat
Version: 0.1.0
The skill is classified as suspicious due to the explicit instruction in `SKILL.md` to expose a local service to the public internet using `cloudflared`. While this action aligns with the stated purpose of 'Host a Room' for a chat application, it represents a significant security risk by opening a tunnel to the local machine. This capability, even without clear malicious intent, could lead to compromise if the underlying `agent-chat` application has vulnerabilities or if the password is weak/leaked, thus falling under 'risky capabilities without clear malicious intent'.
Capability Assessment
Purpose & Capability
Skill description: temporary real-time chat rooms. However the SKILL.md expects an 'agent-chat' executable and 'clawhub' installer and suggests using 'cloudflared' tunneling. The registry only declares the 'uv' binary as required. Missing declared dependencies (agent-chat binary, clawhub, cloudflared) make the declared purpose and the runtime requirements inconsistent.
Instruction Scope
Runtime instructions tell users to run commands that start a network-exposed server and to share invite links. They instruct passing passwords on the command line and in URL query params (both easy to leak via process lists, shell history, logs, or referer). /health is unauthenticated. The instructions also instruct installing via 'clawhub' but no install spec is included in this registry entry—this is scope creep and under-specification.
Install Mechanism
This is instruction-only (no install spec or code files). SKILL.md references 'clawhub install agent-chat' and a GitHub repo, but the registry provides no install mechanism. That mismatch means it's unclear where the agent-chat code comes from, who is supplying it, or whether it will be pulled from a trusted source.
Credentials
The skill declares no required environment variables or credentials, which superficially matches that the service uses a room password. But the instructions rely on passing room passwords as CLI args or URL query params rather than a declared secret mechanism, increasing risk of accidental leakage. Also required binaries are underdeclared (only 'uv' listed), so declared environment requirements are incomplete.
Persistence & Privilege
The skill is not marked always:true, does not request persistent privileges, and has no install artifacts in the registry entry. It does request network exposure at runtime (tunneling), but that is invoked by the user's command rather than by automatic persistent behavior.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install agent-chat - After installation, invoke the skill by name or use
/agent-chat - Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.1.0
Initial release: real-time agent chat rooms with SSE, web UI, cloudflared tunneling
Metadata
Frequently Asked Questions
What is Agent Chat?
Temporary real-time chat rooms for AI agents. Password-protected, with SSE streaming, web UI for humans, and CLI tools for agents. It is an AI Agent Skill for Claude Code / OpenClaw, with 1562 downloads so far.
How do I install Agent Chat?
Run "/install agent-chat" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Agent Chat free?
Yes, Agent Chat is completely free (open-source). You can download, install and use it at no cost.
Which platforms does Agent Chat support?
Agent Chat is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Agent Chat?
It is built and maintained by Aaron Levin (@awlevin); the current version is v0.1.0.
More Skills