Web Performance Engine

• Outcome: Review
• Latest audit: Sep 10, 2026 · 8:59 PM UTC
• Version: 1.0.0

Overview

This is a legitimate website performance skill, but it includes unsafe command examples that could run unintended shell commands if a hostile URL is audited.

Review before installing if the agent may audit URLs supplied by other people. Use the skill only with trusted targets or require the agent to validate URLs and pass them as separate quoted arguments, preferably avoiding shell interpolation entirely. Run any live audits in a constrained environment with limited filesystem and network access.

A.I.G

Vulnerability Patterns

• Insecure Skill Coding Practices: Finds exploitable flaws such as hardcoded secrets or command injection
• Skill Instruction Hijacking: Alters the agent's session goals or safety constraints when the skill loads
• Agent Memory Poisoning: Writes attacker-controlled rules into memory that affect later sessions
• Remote Payload Retrieval and Execution: Fetches external code whose behavior can change after review
• Embedded Malicious Code: Ships malicious scripts inside the skill and executes them locally

Findings (1)

T09 · Insecure Skill Coding Practices

Error

• Location: SKILL.md:124

Finding

Shell Command Injection Through an Unquoted User-Controlled URL

Content

View full analysis

Vulnerability Details

File Location: SKILL.md, lines 124–130
Vulnerability Type: Shell command injection
Risk Level: High

Vulnerable Code

bash

# Check: curl -sI {url} | grep -i 'x-cache\|cf-cache\|x-cdn'

# Compression: Must have brotli or gzip
# Check: curl -sI -H "Accept-Encoding: br,gzip" {url} | grep -i content-encoding

# HTTP/2 or HTTP/3
# Check: curl -sI --http2 {url} | head -1

Technical Analysis

The {url} placeholder is inserted into shell command examples without quoting, URL validation, or an end-of-options delimiter. The skill accepts natural-language commands containing user-supplied audit URLs and directs an agent to perform these checks.

If an agent replaces {url} verbatim and executes the resulting command through a shell, shell metacharacters in the supplied value—such as command separators, substitutions, or redirections—will be interpreted as shell syntax rather than as part of the URL. Merely prefixing a malicious value with an HTTP URL does not prevent exploitation.

The vulnerability is present in documentation rather than an executable script, so exploitation depends on an agent operationalizing the examples through a shell. Nevertheless, executing recommended commands is consistent with the skill's audit workflow.

Attack Path

1. An attacker provides an audit request containing a crafted URL with shell metacharacters and an appended command.
2. The agent follows the server-optimization playbook in SKILL.md.
3. The agent substitutes the supplied value directly for {url} in one of the documented curl commands.
4. The agent invokes the generated command through a shell.
5. The shell interprets the injected syntax and executes the attacker's appended command.
6. The injected command runs with the same operating-system identity, environment access, filesystem permissions, and network access as the agent process.

Impact Assessment

Successful exploitation permits arbitrary command executi ...[truncated 542 chars]

Remediation

View remediation

Remediation Suggestions

1. Parse the supplied target with a standards-compliant URL parser and allow only explicitly supported schemes, preferably https and, where necessary, http.
2. Reject malformed targets, embedded credentials, control characters, and unexpected URL components.
3. Execute curl without a shell by passing the URL as a distinct argument in an argument array.
4. If shell execution is unavoidable, place the validated URL in a variable, quote every expansion, and use an end-of-options delimiter:

bash

curl -sI -- "$url" | grep -iE 'x-cache|cf-cache|x-cdn'
curl -sI -H 'Accept-Encoding: br,gzip' -- "$url" | grep -i content-encoding
curl -sI --http2 -- "$url" | head -n 1

5. Do not rely on quoting alone as the validation boundary. Add network-target controls to reject loopback, link-local, private, metadata-service, and otherwise prohibited destinations if untrusted users can supply URLs.
6. Update the skill instructions to explicitly prohibit direct interpolation of natural-language input into shell commands.
7. Run audit tooling under a dedicated, least-privileged account with restricted filesystem access, minimal environment secrets, outbound-network controls, and execution timeouts.

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (2)

Hidden Instructions

High

• Category: Prompt Injection
• Confidence: 70% confidence

Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 158)May include surrounding context.

CSS optimization:

html

<!-- BEFORE: Render-blocking -->
<link rel="stylesheet" href="styles.css">

<!-- AFTER: Critical CSS inline + async load -->

Vague Triggers

Medium

• Category: Not specified by scanner
• Confidence: 88% confidence

Finding

The README provides invocation examples such as "What's slowing down my site?" and "Set a performance budget for my project" without clarifying whether these are exact triggers, examples only, or what contexts should activate the skill. Because the file does not define trigger scope, exclusions, or negative examples, the activation conditions are somewhat ambiguous for a markdown skill description.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.