Back to skill

Security audit

gmira

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent web-design skill with disclosed project-editing behavior and no evidence of hidden persistence, exfiltration, or destructive actions.

Installers should expect this skill to be opinionated and to modify frontend project files when used for builds. Review proposed component installs, package additions, copy tone, locale, and any deployment step before approving them.

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
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The routing rule uses very broad natural-language triggers such as "build me a site" and "make this better," which are common user requests and can cause this skill to capture work outside a narrowly intended scope. In an agentic system, that increases the chance of over-activation, unexpected behavior chaining, and bypass of a more appropriate specialized skill, especially because the skill then prescribes mandatory follow-on steps like verification and screenshot review.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill mandates a specific prose style for all user-facing writing, including page copy, without indicating that user preferences or project requirements can override it. This can override caller intent, brand guidelines, accessibility needs, or localization requirements, creating policy drift and reducing user control over outputs.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The doctrine imposes a mandatory writing style and denylist for user-facing text without indicating that user preferences or project requirements can override it. In an agent skill, this can cause the agent to ignore explicit user instructions about tone, branding, or accessibility-sensitive phrasing, creating specification drift and potentially misleading outputs presented as if they reflect user intent.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This is a direct intent-code divergence captured in the file itself: the narrative says the source comments describe a visible mouse interaction, but the shown implementation resets uMouse to (0.5, 0.5) and fixes intensity every frame. That means the documented interactive intent contradicts the actual non-interactive behavior.

Content

No source excerpt is available for this finding.

Behavior Manipulation

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Subtle instructions detected that may alter agent decision-making or introduce hidden biases.

Content

Scanner excerpt · references/finding-01-tame-the-arsenal.md (reported line 105)May include surrounding context.

md
Gradient text, which impeccable's craft floor bans outright (`skill-ban-gradient-text`, emphasis
comes from weight or size). The component's `children` slot is the escape hatch and the library
should always use it rather than the `title1` / `title2` props.

Conclusion: the arsenal is worth having, and every component in it needs a taming pass before it
touches a client surface. That pass is a skill.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
77% confidence
Finding

The playbook prescribes copy in German and then provides English equivalents as the accepted register, but it does not state that language selection should be based on user preference or market choice. Because SQP-3 applies to all file types, this can be treated as a locale-policy issue when the skill's guidance effectively narrows output language without explicit opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The copy guidance hard-codes EUR amounts for pricing and test spend examples, which can amount to a locale-specific default in natural-language guidance. The file does not state that the playbook is intended only for a Eurozone audience or offer a user/market choice, so this may conflict with language/locale neutrality expectations.

Content

No source excerpt is available for this finding.