Back to skill

Security audit

Claw Relay

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent and not deceptive, but it asks users to route all AI-agent server traffic through a home laptop and uses risky installation commands without enough warnings.

Review this carefully before installing. Only use it if you understand that your home connection may become the visible source for the VPS agent's traffic, with possible bandwidth, account, reputation, and legal consequences. Prefer installing Tailscale through trusted package-manager instructions, use ACLs or tags to restrict which machines can use the exit node, and document how to disable the exit node and restore normal routing.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:62
Finding

Unverified Remote Installer Executed Directly by a Shell

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 62 and 99
Vulnerability Type: Remote payload retrieval and execution
Risk Level: Critical

Vulnerable code at line 62:

bash
curl -fsSL https://tailscale.com/install.sh | sh

Vulnerable code at line 99:

bash
curl -fsSL https://tailscale.com/install.sh | sh

Technical Analysis

The Skill instructs users to retrieve a mutable installation script from an external URL and stream it directly into sh. The downloaded content is not pinned to a specific version, saved for inspection, or validated using a cryptographic signature or checksum.

HTTPS protects the connection in transit under normal conditions, and the URL appears to use Tailscale's official domain. However, it does not establish that the returned script is immutable or independently verified. The effective code can change after the Skill has been reviewed. Compromise of the remote hosting infrastructure, domain, TLS trust chain, or installer publication process could therefore turn these instructions into an arbitrary code-execution channel.

Installing Tailscale is relevant to the declared functionality, but executing an unverified remote script immediately is not the least-risk installation method. The installer may invoke privileged package-management operations or ask the user for authorization, potentially modifying repositories, packages, services, firewall settings, and network configuration.

Attack Path

  1. The user or AI agent follows either the residential-node or cloud-node setup instructions.
  2. curl requests the current contents of https://tailscale.com/install.sh.
  3. An attacker who has compromised the publication infrastructure, domain, delivery path, or installer supplies modified shell code.
  4. The pipeline passes the response directly to sh without inspection or integrity verification.
  5. The malicious commands execute with the invoking user' ...[truncated 978 chars]
Remediation
View remediation

Remediation Suggestions

  • Remove both direct curl | sh pipelines.
  • Prefer installation through an authenticated operating-system package repository using the platform's native package manager.
  • Pin the package or installer to an explicitly reviewed version where operationally feasible.
  • If a standalone installer must be used, download it to disk first, verify a vendor-published cryptographic signature or checksum obtained through a trusted channel, inspect it, and execute it only after verification succeeds.
  • Require explicit human confirmation before installation, privilege elevation, service activation, firewall changes, or system-wide route changes.
  • Document which packages, repositories, services, firewall rules, and networking settings the installation is expected to modify.
  • Separate download, verification, and execution into distinct commands so that retrieval failure or verification failure cannot proceed to execution.
  • Apply the safer installation procedure consistently to both the residential node and the cloud node.
Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (8)

External Script Fetching

High
Category
Supply Chain
Confidence
97% confidence
Finding

Piping a remotely fetched install script directly into the shell executes unverified code from the network with minimal opportunity for inspection. If the upstream site, transport, DNS, or distribution path is compromised, the user could run attacker-controlled code during installation.

Content

Scanner excerpt · SKILL.md (reported line 62)May include surrounding context.

Linux:

bash
curl -fsSL https://tailscale.com/install.sh | sh

2. Enable as exit node

Chaining Abuse

High
Category
Tool Misuse
Confidence
96% confidence
Finding

The '| sh' construct is the dangerous execution step that turns a network fetch into immediate command execution. In a security-sensitive networking skill, this is especially risky because it conditions users to run unreviewed code as part of privileged setup.

Content

Scanner excerpt · SKILL.md (reported line 62)May include surrounding context.

Linux:

bash
curl -fsSL https://tailscale.com/install.sh | sh

2. Enable as exit node

External Script Fetching

High
Category
Supply Chain
Confidence
97% confidence
Finding

This repeats the same unsafe pattern of executing a remote script directly via shell on the VPS. Because the target is a cloud server running an AI agent, compromise could expose credentials, agent state, and future traffic routed through the machine.

Content

Scanner excerpt · SKILL.md (reported line 99)May include surrounding context.

1. Install Tailscale

bash
curl -fsSL https://tailscale.com/install.sh | sh

2. Join the tailnet and set exit node

Chaining Abuse

High
Category
Tool Misuse
Confidence
96% confidence
Finding

The same chaining pattern appears on the VPS and creates immediate remote-code-execution risk if the fetched content is malicious or tampered with. Since this host may carry agent credentials and route all traffic, compromise could have broad downstream effects.

Content

Scanner excerpt · SKILL.md (reported line 99)May include surrounding context.

1. Install Tailscale

bash
curl -fsSL https://tailscale.com/install.sh | sh

2. Join the tailnet and set exit node

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 68)May include surrounding context.

2. Enable as exit node

bash
sudo tailscale up --advertise-exit-node

3. Approve the exit node

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 113)May include surrounding context.

2. Enable as exit node

bash
sudo tailscale up --advertise-exit-node

3. Approve the exit node

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly states that all VPS traffic will be routed through the human's residential laptop but does not warn about the privacy, legal, bandwidth, and attribution consequences of doing so. This can expose the human's home IP to third-party services, make their connection the source of agent actions, and create risk if the agent performs abusive or policy-violating activity.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 195)May include surrounding context.

md
## Troubleshooting

- **Exit node not showing**: Make sure you approved it in the admin console
- **VPS still shows datacenter IP**: Run `tailscale status` to verify the exit node is connected, then `sudo tailscale up --exit-node=<laptop>` again
- **Connection drops**: Check that the laptop has internet access and Tailscale is running
- **Laptop went to sleep**: Tailscale reconnects automatically when the laptop wakes up, but the VPS will lose internet access while the laptop is offline