Security audit

skill-optimizer

Security checks for vulnerabilities and agentic risk

Overview

The skill is a simple Markdown workflow for analyzing skill performance and applying user-approved edits, with no hidden code or install-time behavior.

Install only if you want a skill that can read chosen logs and, after confirmation, edit another skill's SKILL.md. Review the exact target file and requested changes before approving edits, and keep backups or version control for skills you optimize.

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

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: Allowing a user to provide a file path for the skill to read, without defining scope or warning about local file access, can expose unintended local or sensitive files. In a skill that analyzes conversation logs, this context makes the risk credible because operators may paste arbitrary paths to logs, configs, or other private data.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The skill’s advertised purpose is analysis and recommendation, but the workflow escalates into directly modifying another skill’s SKILL.md. That expands authority from advisory behavior into persistent configuration changes, creating a mismatch that can cause users or downstream systems to approve a skill without understanding it can alter files.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The workflow says the skill will directly modify the target skill’s SKILL.md after confirmation, but it does not clearly warn that this is a persistent file change. In this context, that is especially risky because the skill is framed as an optimizer, so users may treat it as advisory while it actually rewrites operational prompt/configuration files.

Intent-Code Divergence

Low

Confidence: 91% confidence
Finding: The Chinese metadata says the skill will generate suggestions and execute modifications, while the main description only promises analysis and suggestions. This inconsistent disclosure can mislead users about the skill’s true write capabilities, weakening informed consent and review.

Static analysis

No suspicious patterns detected.