Security audit

api-scaffold-gen

Security checks for vulnerabilities and agentic risk

Overview

This skill is a plausible API project generator, but it asks for broad read/write/command authority and deployment-related actions without enough user control or data-handling clarity.

Review this skill before installing if you use private repositories or production infrastructure. Use it only in a bounded workspace, confirm every file write and command before execution, avoid giving it production credentials, and treat claims about local-only processing cautiously unless your agent environment guarantees no repository content is sent to an external LLM or service.

SkillSpector

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Behavioral ASTexec() Call, eval() Call, Dynamic Import
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 (4)

Intent-Code Divergence

Medium

Confidence: 84% confidence
Finding: The skill claims code is generated locally without uploading code, yet elsewhere requires LLM/API connectivity and references network-dependent behavior. This inconsistency can mislead users into exposing proprietary source code or secrets to remote services under the false assumption that processing is strictly local.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The skill exposes read/write/exec capabilities and encourages project generation and deployment actions, but does not prominently warn about filesystem modification and shell-command side effects. In an agent setting, vague authorization boundaries around exec/write can lead to unintended command execution, destructive file changes, or deployment actions triggered from natural-language requests.

Vague Triggers

Medium

Confidence: 78% confidence
Finding: The invocation instructions are too vague about how the skill should be triggered and what precise inputs are required. In agent environments, ambiguous activation broadens the chance that unrelated user requests are interpreted as permission to scaffold projects, execute commands, or modify files.

Vague Triggers

Medium

Confidence: 80% confidence
Finding: Telling users to 'speak to the agent directly' without a constrained trigger phrase or scope increases the risk of overbroad activation. Because the skill has exec/write capabilities, unclear invocation semantics make accidental or unauthorized side effects more dangerous than in a read-only skill.

Static analysis

No suspicious patterns detected.