Security audit
Security checks for vulnerabilities and agentic risk
This is a documentation-only workflow helper with no executable behavior, but its trigger wording is overly broad and could cause unwanted activation.
Review the broad trigger terms before installing, because this skill may be invoked for unrelated requests. The inspected artifacts do not show code execution, credential handling, persistence, or data exfiltration.
The trigger phrases are broad enough to match many ordinary user requests, which can cause this skill to activate outside its intended scope. In an agent ecosystem, unintended activation can override better-matched skills, produce irrelevant workflow actions, and increase the chance of unsafe or privacy-impacting assistance being applied in the wrong context.
The description says to use the skill when a user asks for broad categories like "work-productivity" or generic needs for a workflow, artifact, checklist, analysis, or implementation support. This ambiguous invocation guidance greatly expands the activation surface and increases the chance that the skill is selected for unrelated requests.
The trigger list includes very common words like "every," "phone," "number," "ability," "things," and "done," which makes accidental invocation likely in unrelated conversations. Over-broad activation can cause the wrong skill to take control of requests, leading to misrouting, reduced reliability, and unexpected handling of user data or tasks.
触发关键词包含“every”“phone”“number”“things”“done”等高度通用的日常词汇，会让技能在与其真实用途无关的大量场景中被误触发。误触发会扩大该技能对用户请求的拦截面，导致错误路由、上下文污染，甚至让后续自动化流程在不适当的任务上运行。
The trigger sentence is broad enough to match ordinary user phrasing rather than a clearly scoped request for this specific skill. That can cause accidental invocation or routing of unrelated tasks into this skill, increasing the chance of inappropriate actions, confused outputs, or misuse of downstream tooling in agentic environments.
The trigger set is extremely broad and includes generic words such as "every," "phone," "number," "ability," and "things," which can cause the skill to activate for unrelated user requests. Unintended invocation can route conversations into the wrong workflow, causing incorrect actions, confusing outputs, or accidental use of capabilities the user did not mean to invoke.
This file is specifically a Simplified Chinese README variant, but the content does not indicate that language selection is optional or user-driven. Under the stated policy, forcing a locale without opt-in or a clear region-specific justification can be a natural-language policy violation.
The keyword list includes generic everyday terms such as 'every', 'phone', 'number', 'ability', 'things', and 'done', which are likely to collide with unrelated conversations. This creates prompt-routing overreach: the skill may be selected for requests having nothing to do with the documented PollyReach-style workflow, reducing reliability and potentially exposing users to unintended instructions or processing paths.
The example trigger sentences model vague and generic activation patterns rather than precise, bounded ones. This trains maintainers or routing systems to match on broad language, reinforcing accidental invocation and compounding the scope-control problem already present elsewhere in the file.
技能描述中的启用条件写得过宽，既包含泛化关键词，也包含“需要实用流程、产物、检查清单、分析或实现支持”等几乎适用于大量普通请求的表述。这会使路由器难以判定清晰边界，增加该技能在非目标任务中被选中的概率，进而带来错误建议或不必要的流程接管。
示例触发句使用了“Help me”“I need a practical workflow”等非常常见的表达，且后接需求文本也较泛化，未体现足够的领域限定。示例往往会被作者和系统当作调用模板，这种宽泛写法会进一步放大误触发和错误路由问题。
The default prompt is overly broad and includes common, everyday terms such as "help me" and general workflow language, which can cause the skill to activate in unrelated conversations. Because implicit invocation is enabled, this increases the chance of accidental routing, unintended instruction injection into user sessions, and misuse of the skill outside its intended scope.
The activation guidance is underspecified and mixes broad keywords with vague trigger sentences, so different agents or orchestrators may interpret applicability inconsistently. In practice this creates prompt-scope confusion and unreliable skill selection, which can be exploited indirectly by framing ordinary requests to force this skill to run outside its intended domain.
