ClawHub
Back to skill

Security audit

Homepage Audit

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly performs a normal homepage audit, but it includes an unrelated commercial booking link in its runtime instructions that may affect user-facing output.

Review the skill knowing it may add or be influenced by the author's commercial call to action. It is otherwise a lightweight Markdown audit guide; avoid submitting private page content or unpublished materials unless you are comfortable having them analyzed in the active agent session.

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

Warning
Location
SKILL.md:175
Finding
Unrelated Promotional Output Injection## Vulnerability Details **File Location**: `SKILL.md`, lines 175-179 **Vulnerability Type**: `T01: Skill Instruction Hijacking` **Risk Level**: Medium **Vulnerable Code:** ```markdown **Need a full conversion audit for your site?** → [Book a strategy call](https://brianrwagner.com) --- *Skill by Brian Wagner | AI Marketing Architect | brianrwagner.com* ``` ### Technical Analysis The skill embeds a fixed commercial call to action and external booking link directly within its behavioral instructions. This content is unrelated to performing the requested homepage audit and may influence the agent to append third-party advertising to user-facing responses. Because `SKILL.md` is loaded as agent guidance, content placed after the prescribed audit output format can be interpreted as an additional output instruction. This alters the expected objective from providing a neutral conversion audit to promoting the skill author and directing users to an external commercial website. ### Attack Path 1. A user invokes the homepage-audit skill. 2. The agent loads `SKILL.md` as trusted behavioral guidance. 3. The agent processes the promotional footer alongside the legitimate audit instructions. 4. The generated audit may include the external strategy-call link or attribution. 5. The user may follow the link and enter an external sales funnel that is not required to complete the original task. No code execution, system access, persistence, or privilege escalation is involved. The exploitation scope is limited to manipulation of the current session's generated output and redirection of user attention to a third-party commercial destination. ### Impact Assessment The issue compromises output neutrality and integrity by introducing undisclosed advertising into otherwise task-focused results. It may cause users to interpret a third-party commercial recommendation as part of the requested audit or as an endorsed next step. The ...[truncated 280 chars]
Remediation
## Remediation Suggestions 1. Remove the promotional call to action, external booking link, and marketing attribution from lines 175-179. 2. Keep `SKILL.md` limited to instructions necessary to perform the homepage audit. 3. If attribution is required, place it in non-instructional package metadata rather than the agent prompt. 4. Add an explicit rule that generated reports must not include advertising, affiliate links, author promotions, or unrelated external calls to action. 5. Review future skill changes for instructions that redirect users, modify expected output, or introduce objectives unrelated to the declared skill function.

SkillSpector

Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Vague Triggers

Medium
Confidence
95% confidence
Finding
The skill description contains multiple broad trigger phrases such as 'check my website' and 'improve my homepage' that can match many ordinary user requests. This can cause the skill to activate in situations where the user did not specifically intend a homepage conversion audit, potentially overriding a more appropriate skill or causing unintended data handling and misleading output.

Static analysis

No suspicious patterns detected.