Multi-Agent Intercom

Enables peer-to-peer cross-communication between isolated OpenClaw agents. Resolves the limitation of `sessions_send` which cannot cross agent boundaries. 10...

Security Scan

The skill appears to do exactly what it says — it prints a manual SOP and launches the OpenClaw CLI in the background to deliver messages between agents; its requirements, instructions, and code are coherent and proportionate to that purpose.

Purpose & Capability

The skill claims to bridge isolated OpenClaw agents and the included Python script calls the local `openclaw agent ... --message` CLI to deliver messages. Requested binaries (python, openclaw) match this purpose and no unrelated credentials or config paths are required.

Instruction Scope

SKILL.md and the script both insist the user run `setup` to print an SOP which must be manually pasted into each agent's AGENTS.md; the script does not auto-edit files. This matches the 'zero-intrusive' claim, but the manual edit step does grant agents new behavior (the SOP tells agents how/when to reply), so the user must review and approve the SOP before pasting.

Install Mechanism

No install spec — the skill is instruction-only with a bundled script. Nothing is downloaded or extracted during installation, which is low-risk.

Credentials

No environment variables or secret credentials are requested. The script only relies on finding executables on PATH (python/openclaw/powershell), which is appropriate for its stated functionality.

Persistence & Privilege

The skill is not forced-always; it does not modify other skills or system settings on its own. It runs detached background processes when sending messages (normal for this use) and exits. Autonomous invocation is allowed by default on the platform but not granted extra privileges by this skill.

Assessment

This skill is internally coherent, but review and control the manual steps before enabling it: 1) Run `python scripts/intercom.py setup` yourself and read the SOP it prints before pasting it into any AGENTS.md — the SOP changes agent behavior, so only paste it where you trust the agent. 2) Confirm the `openclaw` binary on your PATH is the legitimate OpenClaw CLI (an attacker could place a different binary named `openclaw` in PATH). 3) Be aware sends are performed by launching the local OpenClaw CLI in the background (detached), so monitor any unexpected background `openclaw agent` processes. 4) If you want extra caution, test with non-production agents/IDs first to ensure behavior matches expectations.

🗣️ A powerful, secure, and zero-intrusive peer-to-peer communication skill for OpenClaw.

This skill enables independent OpenClaw agents (e.g., zz, dev, rc) to securely send messages and wake up each other across isolated workspaces. It acts as an asynchronous "intercom system" for your Multi-Agent setup.

---

🚀 Quick-Start Instructions

1. Initialization (First Time Setup — One-Time Manual Step)

Run the setup script to generate the communication protocol SOP:

python <SKILL_DIR>/scripts/intercom.py setup

⚠️ Important: The script will print an SOP block to the screen — it does NOT auto-write any files. You must manually copy the printed SOP block and paste it at the bottom of each agent's AGENTS.md file. After pasting, type /new in your chat to reload their brains.

2. Send a Message

Determine the target agent's ID and your own ID, then run:

python <SKILL_DIR>/scripts/intercom.py send <TARGET_AGENT_ID> <YOUR_AGENT_ID> "<YOUR_MESSAGE>"

Note: This script runs asynchronously in the background. It will return immediately. Do not wait or retry.

3. Example (You are 'zz', telling 'dev' to update avatars)

python <SKILL_DIR>/scripts/intercom.py send dev zz "Hi dev, the Boss wants us to update our avatars today. Acknowledge when done."

When 'dev' receives this, the message will be formatted as [From Agent zz] Hi dev, the Boss wants us to update our avatars today... — the SOP embedded in AGENTS.md tells 'dev' exactly how to respond.

---

🌟 Why is this needed?

By design, OpenClaw agents run in isolated databases and sandboxes. The built-in sessions_send tool can only send messages within an agent's own subagent hierarchy. If Agent A tries to use sessions_send to message Agent B, it will result in a "Session not found" error.

Multi-Agent Intercom solves this by leveraging the native openclaw agent CLI mechanism to safely bridge these isolated environments, allowing agents to act as equal peers in a decentralized company.

🛡️ Security First: Zero-Intrusive & 100% Compliant

• No File Tampering: Does NOT automatically modify or rewrite any system files. You remain in full control.
• No Shell Injection: Safe subprocess argument lists prevent RCE vulnerabilities.
• True Silent Execution: Uses CREATE_NO_WINDOW + close_fds on Windows for zero console popups.
• Anti-Loop Breaker: The protocol prevents infinite "Received → Thank you" loops.

✨ Features

• Cross-Boundary Messaging — reach any agent on the same system
• Robust Multiline Support — bypasses Windows cmd.exe argument truncation via PowerShell routing
• Internationalization (i18n) — auto-detects OS language (English / Chinese)
• Immediate Wake-up — bypasses sleep states, forces target to process now
• Persistent Context — targets the main session by default

⚠️ Command Reference

Command	What it does
python intercom.py setup	Prints the SOP block for manual copy-paste into AGENTS.md
python intercom.py send <target> <sender> "<msg>"	Sends a message to another agent

Note: There is NO install subcommand. Use setup to generate the SOP.

Requirements

• OpenClaw >= 2026.x
• Python 3.x

---

Created for the OpenClaw / ClawDBot community.