Back to skill

Security audit

Notebooklm Content

Security checks for vulnerabilities and agentic risk

Overview

The skill appears purpose-aligned for NotebookLM automation, but it deserves Review because it controls an authenticated Chrome session and its setup guide tells users to print a gateway token in the terminal.

Install only if you are comfortable letting the agent operate a NotebookLM tab in your logged-in Chrome session. Enable the relay only for the specific tab you intend to automate, avoid taking screenshots or snapshots of unrelated sensitive content, and do not paste or share terminal output containing gateway tokens. Rotate the gateway token if it may have been exposed.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
references/chrome-relay-setup.md:8
Finding
Gateway Authentication Token Exposed in Terminal Output<![CDATA[ ## Vulnerability Details **File Location**: `references/chrome-relay-setup.md`, lines 8–14 **Vulnerability Type**: Plaintext sensitive-data exposure **Risk Level**: Medium ### Vulnerable Code ```markdown 3. Configure: - Port: 18792 - Gateway token: from ~/.openclaw/openclaw.json -> gateway.auth.token ## Get Token ```bash cat ~/.openclaw/openclaw.json | grep -A5 auth ``` ``` ### Technical Analysis The setup guide instructs users to read the OpenClaw configuration file and print the authentication section, including five subsequent lines, to standard output. This can disclose the gateway authentication token and potentially adjacent sensitive configuration. Although the guide warns users not to paste the token into chat, the command itself still exposes sensitive values in the terminal. The output may be captured by terminal logging, shell-session recording, screen sharing, screenshots, CI logs, support transcripts, or clipboard-based troubleshooting. The broad `grep -A5 auth` selection may also reveal more data than the single required token. This is a sensitive-data handling weakness rather than automatic exfiltration: exploitation requires an attacker to obtain the exposed output or otherwise observe the terminal session. ### Attack Path 1. A user follows the documented setup procedure. 2. The user executes `cat ~/.openclaw/openclaw.json | grep -A5 auth`. 3. The gateway token and neighboring configuration are printed in plaintext. 4. The output is captured through terminal logging, screen sharing, screenshots, session recording, or troubleshooting material. 5. An attacker with access to that output extracts the gateway token. 6. If the attacker can reach the associated OpenClaw Gateway, the attacker attempts to authenticate with the disclosed token. 7. Successful authentication may permit use of gateway capabilities, including control of browser tabs that the user has exposed through the relay. ### Impact Assessment Disclosure of t ...[truncated 640 chars]
Remediation
<![CDATA[ ## Remediation Suggestions - Remove the command that prints the authentication section to standard output. - Direct users to a protected configuration interface that masks credentials and supports secure copy operations. - If command-line extraction is unavoidable, retrieve only the exact token field rather than surrounding lines, and avoid displaying it in terminal output. - Prefer a secure configuration command that reads the token directly from the protected file or standard input without echoing it. - Ensure the configuration file has restrictive permissions, such as owner-only read and write access. - Warn users not to include terminal output containing credentials in logs, screenshots, support tickets, chat messages, or screen-sharing sessions. - Redact gateway tokens automatically from diagnostic output and logging systems. - Provide documented token-rotation and revocation procedures, and instruct users to rotate any token that may have been exposed. - Restrict the gateway to loopback or otherwise trusted network interfaces unless remote access is explicitly required, and apply least-privilege controls to gateway capabilities. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Vague Triggers

Medium
Confidence
92% confidence
Finding
The manifest description uses broad activation phrases such as creating slides, generating audio explanations, and building learning materials from text/URLs, which can match many common user requests outside a narrowly scoped NotebookLM task. This increases the chance the skill is invoked in situations where users did not intend browser automation against a logged-in Google session, expanding exposure of notebook contents, URLs, and account context.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The skill description does not warn users that browser snapshots, tab inspection, screenshots, and automated interaction with a logged-in NotebookLM session can expose sensitive notebook contents, URLs, page text, and other on-screen data. In this context, the risk is elevated because the workflow explicitly depends on a live authenticated browser and includes commands for snapshots and screenshots, making accidental data disclosure more likely.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The guide instructs users to run a shell command against ~/.openclaw/openclaw.json specifically to retrieve an authentication token from a local credential-bearing file. Although it says not to paste the token in chat, the instructions still normalize direct secret extraction and display in the terminal without safer handling guidance, increasing the risk of accidental disclosure through logs, screenshots, shell history, or copy/paste into untrusted contexts.

Static analysis

No suspicious patterns detected.