Back to skill

Security audit

ClawOrchestrate

Security checks for vulnerabilities and agentic risk

Overview

This skill has a coherent orchestration purpose, but it exposes remote agent execution through an under-secured HTTP dispatcher with weak deployment defaults and persistent service setup.

Review before installing. Only run this on trusted private networks, require a strong non-default API key, avoid exposing port 9876 broadly, prefer TLS or identity-aware private transport, and fix the service configuration before enabling persistence. Treat any reachable dispatcher as an interface that can start local OpenClaw agents under that user's permissions.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (6)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/dispatch-all.sh:7
Finding

Predictable fallback API credential

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/dispatch-all.sh:27
Finding

Bearer credentials and agent instructions transmitted over plaintext HTTP

Content
View full analysis
&1) ``` The same plaintext request pattern is used for PC3: ```bash result=$(curl -s -X POST "http://$PC3:9876/dispatch" \ -H 'Content-Type: application/json' \ -H "Authorization: Bearer $API_KEY" \ -d "{\"agent\":\"$agent\",\"message\":\"$MESSAGE\"}" 2>&1) ``` ### Technical Analysis The script transmits a reusable bearer credential and the complete agent instruction through HTTP without application-layer TLS. Bearer credentials provide access to anyone who possesses them and do not protect against interception or replay. The hard-coded addresses appear to be Tailscale addresses, which ordinarily provide encrypted transport within a correctly configured tailnet. Nevertheless, the application does not enforce that the destination is reached through a trusted encrypted overlay. Reuse on another network, routing mistakes, proxies, or future address changes could expose both credentials and task contents. ### Attack Path 1. An operator runs the broadcast script over a network path that is not protected by a correctly configured encrypted overlay. 2. An attacker able to observe or intercept that path captures the HTTP `Authorization` header. 3. The attacker replays the recovered bearer credential against port 9876. 4. The attacker submits arbitrary messages to available agent identifiers. 5. The dispatcher invokes the selected local agent with the supplied instruction. A man-in-the-middle attacker could also modify the unprotected request body in transit. ### Impact Assessment Successful interception permits ...[truncated 320 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
scripts/dispatcher.py:26
Finding

Dispatcher listens on every network interface by default

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/dispatcher.py:175
Finding

API secret supplied through process command-line arguments

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/dispatcher.py:156
Finding

Predictable log files in shared temporary directory permit symlink attacks and data exposure

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Note
Location
scripts/setup-service.sh:21
Finding

Persistent service installer omits mandatory authentication configuration and enters a restart loop

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (18)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The description presents a broader orchestration system: dispatching tasks, monitoring progress, and coordinating teams across machines via multiple network methods. The supplied code chunk implements only a narrow subset: broadcasting one message to predefined agents on two predefined IPs using HTTP requests to a /dispatch endpoint with bearer auth. That supports the dispatch portion, but not the claimed monitoring or broader coordination capabilities. The transport is also materially narrower than described, since the script itself uses HTTP to hardcoded addresses and does not implement SSH or generic multi-network orchestration. Therefore the description overstates the actual behavior of this code chunk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The README explicitly instructs users to expose a lightweight HTTP dispatcher and send agent commands over plain unauthenticated HTTP. In this skill's context, those requests trigger remote agent task execution, so any reachable party on the network could inject commands, tamper with tasks, or abuse the orchestrator for unauthorized remote actions.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
65% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · README.md (reported line 20)May include surrounding context.

bash
# 1. Install on remote machines
scp scripts/dispatcher.py user@remote:~/.openclaw/
ssh user@remote "nohup python3 ~/.openclaw/dispatcher.py &"

# 2. Send tasks from gateway
curl -X POST http://remote-ip:9876/dispatch \

Session Persistence

Medium
Category
Rogue Agent
Confidence
65% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 32)May include surrounding context.

bash
# 1. Install on remote machines
scp scripts/dispatcher.py user@remote:~/.openclaw/
ssh user@remote "nohup python3 ~/.openclaw/dispatcher.py &"

# 2. Send tasks from gateway
curl -X POST http://remote-ip:9876/dispatch \

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

The example shows sending task data to http://remote-ip:9876/dispatch via POST, which implies commands are transmitted without encryption or authenticated integrity checks. Because the payload contains instructions for remote agents, an attacker on the path or any party with access to the endpoint could observe, modify, or submit arbitrary tasks.

Content

Scanner excerpt · README.md (reported line 23)May include surrounding context.

md
ssh user@remote "nohup python3 ~/.openclaw/dispatcher.py &"

# 2. Send tasks from gateway
curl -X POST http://remote-ip:9876/dispatch \
  -d '{"agent":"my-agent","message":"Start building."}'

# 3. Broadcast to all machines

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill advertises and documents capabilities that involve network access, shell execution, and file operations, but it declares no tool scope or permission boundaries. In an orchestration skill that can dispatch work to remote machines, missing explicit permission constraints increases the chance of over-broad execution and unintended access if the skill is installed or invoked in a permissive environment.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The quick-start instructs users to expose a /dispatch endpoint that accepts agent and message input and triggers remote agent execution, while explicitly noting there is no authentication. Even on a private overlay network, any compromised peer or misconfigured routing could abuse the endpoint to send arbitrary tasks to remote agents, making this especially dangerous in a tool designed for cross-machine control.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 39)May include surrounding context.

bash
# Dispatch to an agent
curl -X POST http://remote-ip:9876/dispatch \
  -H 'Content-Type: application/json' \
  -d '{"agent":"byondedu-ceo","message":"Read TASKS.md and start building."}'

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The script advertises itself as 'Secure' but sends both the bearer token and message body over plain HTTP to remote hosts. Anyone able to observe traffic on the path to those agents can recover credentials and task contents, then replay or forge dispatch requests.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
83% confidence
Finding

External transmission by itself is expected for an orchestration script, so the network call is not inherently malicious. However, in this implementation the transmission occurs over insecure HTTP and includes an authentication token, which makes the call security-relevant and exploitable if the network path is not fully trusted.

Content

Scanner excerpt · scripts/dispatch-all.sh (reported line 27)May include surrounding context.

sh
for agent in "${PC2_AGENTS[@]}"; do
    echo -n "  PC2 → $agent: "
    result=$(curl -s -X POST "http://$PC2:9876/dispatch" \
        -H 'Content-Type: application/json' \
        -H "Authorization: Bearer $API_KEY" \
        -d "{\"agent\":\"$agent\",\"message\":\"$MESSAGE\"}" 2>&1)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

The broadcast loop transmits sensitive operational data and the Authorization bearer token in cleartext without warning or safeguards. In an orchestration tool, intercepted messages could reveal internal coordination details while a stolen token could let an attacker issue commands to multiple agents across machines.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/dispatcher.py (reported line 170)May include surrounding context.

python
def trigger_agent(self, agent_id, message):
        """Run openclaw agent in background."""
        try:
            proc = subprocess.Popen(
                [OPENCLAW, "agent", "--agent", agent_id, "--message", message, "--timeout", "120"],
                stdout=open(f"/tmp/dispatch-{agent_id}.log", "a"),
                stderr=subprocess.STDOUT

Session Persistence

Medium
Category
Rogue Agent
Confidence
80% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 114)May include surrounding context.

md
#!/bin/bash
# ClawOrchestrate — Systemd service installer
# Run this on each remote machine to set up persistent dispatcher

set -e

Session Persistence

Medium
Category
Rogue Agent
Confidence
80% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · scripts/setup-service.sh (reported line 2)May include surrounding context.

sh
#!/bin/bash
# ClawOrchestrate — Systemd service installer
# Run this on each remote machine to set up persistent dispatcher

set -e

Session Persistence

Medium
Category
Rogue Agent
Confidence
80% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · scripts/setup-service.sh (reported line 14)May include surrounding context.

sh
echo "Installing ClawOrchestrate dispatcher as systemd service..."

# Create service file
cat > ~/.config/systemd/user/${SERVICE_NAME}.service << EOF
[Unit]
Description=ClawOrchestrate Agent Dispatcher

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · scripts/setup-service.sh (reported line 14)May include surrounding context.

sh
echo "Installing ClawOrchestrate dispatcher as systemd service..."

# Create service file
cat > ~/.config/systemd/user/${SERVICE_NAME}.service << EOF
[Unit]
Description=ClawOrchestrate Agent Dispatcher

Session Persistence

Medium
Category
Rogue Agent
Confidence
80% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 93)May include surrounding context.

md
# Reload and enable
systemctl --user daemon-reload
systemctl --user enable ${SERVICE_NAME}
systemctl --user start ${SERVICE_NAME}

echo "✅ Service installed and started."

Session Persistence

Medium
Category
Rogue Agent
Confidence
80% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · scripts/setup-service.sh (reported line 33)May include surrounding context.

sh
# Reload and enable
systemctl --user daemon-reload
systemctl --user enable ${SERVICE_NAME}
systemctl --user start ${SERVICE_NAME}

echo "✅ Service installed and started."