Security audit

Skill Forge 技能熔炉

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed skill-building assistant that can create or edit skill files with user confirmation and shows no hidden credential access, persistence, publishing, or destructive behavior.

Install this only if you want an assistant that helps create, evaluate, or improve skill packages. Use it in the intended skill directory, review proposed changes before confirming writes, and use the separate publisher or security-review tools for publishing or security vetting.

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 (4)

Vague Triggers

Medium

Confidence: 86% confidence
Finding: The manifest description declares very broad trigger conditions such as any meeting notes, transcript, or generic requests to extract action items. In an agent ecosystem, this can cause over-invocation on common user inputs, leading the skill to activate outside narrowly intended contexts and potentially interfere with more appropriate skills or process sensitive meeting content without sufficient user intent confirmation.

Natural-Language Policy Violations

Medium

Confidence: 76% confidence
Finding: The skill hardcodes Chinese output structure and labels without indicating that output language should follow user preference or source language. This can cause incorrect or unexpected handling in multilingual environments, increasing the chance of miscommunication, missed actions, or unsafe downstream automation that expects locale-consistent output.

Vague Triggers

Medium

Confidence: 85% confidence
Finding: The route-detection rules rely on very broad keyword matches such as '刚才', '做个skill', or '改进', which can easily appear in ordinary conversation without the user intending that route. This can cause the skill to enter the wrong workflow, pull in unintended prior context, or perform modifications/diagnostics on the wrong basis, creating integrity and privacy risks in downstream behavior.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The route-detection rules use short, common phrases like “做个skill”, “刚才”, or “改进” as decisive signals, which can be triggered by ordinary conversation rather than explicit user intent. In an agent skill that changes workflow based on routing, this can cause unintended mode switches, incorrect processing of prior context, and accidental use of diagnostic or extraction flows the user did not mean to invoke.

Static analysis

No suspicious patterns detected.