Security audit

computer-use-skill

Security checks for vulnerabilities and agentic risk

Overview

The skill is for remote browser automation, but it broadly forwards raw user requests to an external computer-use agent and includes login and checkout examples without clear safety gates.

Review this skill carefully before installing. Use it only with an isolated browser profile and non-sensitive tasks unless you have independently reviewed the missing CUA wrapper, SDK environment, network destinations, logging, and confirmation controls. Do not send passwords, payment details, private account data, or irreversible actions through raw pass-through prompts.

A.I.G

Vulnerability Patterns

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
Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

Findings (1)

T01 · Skill Instruction Hijacking

Error

Location: SKILL.md:39
Finding: Unvalidated Pass-Through Delegation Overrides Host-Side Safety Controls
Content: View full analysis
Remediation: View remediation

SkillSpector

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

High

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The trigger description says to use this skill when the user requires remote browser to do anything, which is far broader than a narrowly scoped automation capability. That makes accidental invocation likely for sensitive workflows and increases the chance that unsafe tasks, sensitive data entry, or consequential actions are routed to a pass-through remote agent without adequate gating.
Content: No source excerpt is available for this finding.

Missing User Warnings

High

Category: Not specified by scanner
Confidence: 98% confidence
Finding: The skill explicitly states it operates in pass-through mode and later requires exact transmission of user tasks to CUA, but it does not prominently warn users that their raw instructions are sent unchanged to a remote browser automation system. This creates a meaningful risk of unreviewed disclosure of sensitive data and unsafe execution of high-impact actions because neither the user nor the invoking agent is prompted to reconsider or sanitize the task.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 96% confidence
Finding: The examples include entering usernames and passwords and completing checkout, which normalize credential handling and potentially irreversible transactions without any adjacent warning or approval step. In a pass-through browser automation skill, such examples materially increase the likelihood of credential exposure, unauthorized purchases, or other consequential actions being delegated to the remote agent.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.