ClawHub
Back to skill

Security audit

agos claw chat

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed chat connector, but it can let remote chat requests drive a user-configured local shell command and stores a connector token plainly on disk.

Install only if you intentionally want AITalk remote chat requests to reach a local OpenClaw command. Use a fixed trusted command, run it with least privilege, avoid shell wrappers where possible, assume chat payloads may contain sensitive data, and protect or rotate the stored connector token.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

subprocess module call

Medium
Category
Dangerous Code Execution
Content
env = os.environ.copy()
        env["OPENCLAW_MESSAGE"] = message
        env["OPENCLAW_PAYLOAD"] = json.dumps(request_payload)
        proc = subprocess.run(
            self.agent_cmd,
            shell=True,
            env=env,
Confidence
99% confidence
Finding
The connector executes a configurable command via subprocess.run with shell=True, which turns the skill into a remote-triggered local code execution bridge once requests are claimed from the server. In this context, the claimed network payload is passed into the execution environment and the connector's stated role does not require unrestricted shell execution, so the primitive is materially dangerous.

Context-Inappropriate Capability

High
Confidence
98% confidence
Finding
This code gives the connector the ability to execute arbitrary local shell commands supplied through configuration, despite being presented as a polling/connector runtime. Because requests are fetched from a remote API and then handed to the local command path, the skill becomes a conduit for remote influence over local execution behavior.

Missing User Warnings

Medium
Confidence
85% confidence
Finding
The session token is stored on disk in plaintext in a predictable location under the user's home directory. If local attackers, other users, malware, or backups can access that file, they may reuse the bearer token to impersonate the connector session and interact with the remote service.

Missing User Warnings

Medium
Confidence
88% confidence
Finding
The connector injects claimed request content into environment variables before launching a subprocess. Environment variables are commonly inherited, exposed through diagnostics, crash reports, or process inspection, so sensitive request data may leak beyond the intended component boundary.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal