Back to skill

Security audit

Skill Design Guide

Security checks for vulnerabilities and agentic risk

Overview

The scanners report a design guidance skill with no AIG findings and three contextual findings; the verdict is benign with medium confidence, though its install-related behavior and activation scope need attention.

The verdict is benign, but confidence is medium and scanner coverage is incomplete: Skillspector marked analysis partial because some references could not be resolved, and it could not verify the installed package's contents. Before installing, consider whether the reported package installation and broad trigger fit your needs, and review the package separately if you proceed.

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
  • Rogue AgentSelf-Modification, Session Persistence
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (3)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied shell command performs package installation rather than providing the described design guidance. Although the package name aligns with the declared topic, its contents are not supplied, so its behavior cannot be verified. The observable installation action is an undeclared capability, not merely an implementation detail of the described guidance.

Content

No source excerpt is available for this finding.

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · references/agent-design-research.md (reported line 150)May include surrounding context.

md
- **Tools/scripts = Hands** (file I/O, search, specific operations)
- **Knowledge base/directories = Session** (persistent state and data)

Separation means: modify Skill logic without touching the knowledge base; update knowledge base without rewriting the Skill; change directory structure without redesigning the Skill.

---

The auto-trigger phrase "which pattern should I use" is not restricted to skill architecture.

Low
Category
Trigger Abuse
Confidence
96% confidence
Finding

The README states that mentioning this phrase automatically activates the skill, although it can occur in unrelated discussions about software design, visual design, or other patterns. No activation constraint requires the surrounding request to concern AI skill architecture.

Content

No source excerpt is available for this finding.