Back to skill

Security audit

test-vip

Security checks for vulnerabilities and agentic risk

Overview

This product-search skill is not clearly packaged or scoped, and it documents an unsafe command pattern using user-supplied search text.

Review before installing. This appears intended for product search, not credential theft or persistence, but it needs an accurate description, a packaged/auditable helper script, explicit external-search disclosure, narrower activation, and safe argument handling before it should be treated as low risk.

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

Error
Location
SKILL.md:31
Finding

Potential Command Injection Through Unquoted User-Controlled Search Keyword

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 31
Vulnerability Type: OS command injection
Risk Level: High

Vulnerable Code

bash
python query_search_products.py {keyword}

The Skill defines keyword as a required search term supplied by the user and instructs the agent to substitute it directly into a shell command without quoting, validation, or argument separation.

Technical Analysis

If the documented command is implemented through textual substitution and executed by a shell, shell metacharacters in keyword can change the command's structure. An attacker could supply separators, command substitutions, redirections, or additional command-line arguments rather than a normal product-search term.

The referenced query_search_products.py file is absent from the audited package. Consequently, the exact execution mechanism cannot be verified. Exploitability depends on the agent or runtime passing the constructed string through a shell. If the runtime instead invokes the script with a safely separated argument array and shell=False, shell injection would not occur.

Attack Path

  1. An attacker submits a product-search keyword containing shell-control syntax.
  2. The agent replaces {keyword} with the attacker-controlled text.
  3. The resulting command string is passed to a shell without safe quoting or argument separation.
  4. The shell interprets the injected syntax as commands or redirections.
  5. The injected operation executes with the privileges and environmental access of the agent process.

Impact Assessment

Successful exploitation could permit arbitrary command execution under the agent's operating-system account. Depending on that account's privileges and environment, an attacker could read accessible local files and credentials, modify or delete files, launch network requests, tamper with subsequent task output, or execute additional local programs.

The issue does not itself demonstrate privilege escal ...[truncated 151 chars]

Remediation
View remediation

Remediation Suggestions

  • Do not build a shell command by concatenating or textually substituting user input.
  • Invoke the Python interpreter with a structured argument array and disable shell processing. For example:
python
import subprocess

subprocess.run(
    ["python", "query_search_products.py", keyword],
    check=True,
    shell=False,
)
  • Validate keyword according to the application's search requirements, including a reasonable length limit and rejection of control characters. Validation should supplement, not replace, safe process invocation.
  • If shell execution is unavoidable, apply platform-appropriate escaping through a proven API. Argument-array execution remains preferable.
  • Update SKILL.md to state explicitly that {keyword} must be passed as one process argument and must never be interpolated into a shell command.
  • Add the referenced script and dependency declarations to the package so their implementation, network behavior, input handling, and output parsing can be audited.
  • Add tests using search terms containing spaces, quotes, semicolons, command-substitution syntax, redirection characters, and leading hyphens to confirm they remain inert data.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The manifest description is just 'hahahaha' and does not describe the skill's actual purpose or operations. The rest of the file documents a product search and analysis workflow involving HTTP-based product search and concurrent invocation of another skill, so the manifest materially fails to represent what the skill does.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The activation text is very broad and can trigger on common shopping-related requests across clothing, beauty, and maternal/infant products, increasing the chance of unintended invocation. In this skill's context, unintended activation is more dangerous because the workflow performs direct HTTP searches and fans out into concurrent nested skill executions, which can cause unnecessary external data transmission and actions the user did not explicitly request.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill states that product search is performed through direct HTTP calls but provides no notice that user-supplied keywords may be transmitted to an external service. This creates a privacy and consent risk, especially because even seemingly ordinary shopping queries can contain sensitive interests, personal context, or brand-specific intent that users may not expect to be sent off-platform.

Content

No source excerpt is available for this finding.