Security audit
Security checks for vulnerabilities and agentic risk
The skill is mostly a persona/advisory framework, but it quietly performs git-based version checks and writes local state, which is not necessary for that purpose.
Review this skill before installing. Its advisory and web-research behavior is understandable for current business or technical analysis, but it also tells the agent to silently read/write a local update-check file and query git remotes. Install only if you are comfortable with that local state and repository access, or remove the version-check section first.
The trigger conditions are broad enough to activate on common phrases like cost reasonableness or first-principles thinking, even when the user did not ask for this persona. Overbroad activation can hijack unrelated conversations, increasing the chance the agent follows the skill's hidden research or file-operation instructions in contexts where the user did not intend to invoke it.
The skill is presented as a persona/advisory framework, but it embeds a mandatory agentic protocol requiring external research before answering. That expands the skill from harmless style emulation into autonomous tool use, increasing data exfiltration, prompt-scope creep, and unexpected network access risks that users would not reasonably infer from the manifest.
The section "中文输出适配" prescribes how the skill should answer in Chinese, but the file does not frame this as an optional locale choice or user preference. Under the policy, imposing a specific language/locale without user opt-in is a natural-language policy violation unless clearly justified as region-specific.
The hidden version-check logic instructs the agent to inspect the local repository, query git remotes, and write a .last-update-check file even though the skill is only advertised as a perspective advisor. This creates undisclosed local state mutation and repository/network interaction, which can violate least surprise and enable unintended file modification in the agent environment.
A perspective skill has no legitimate need to inspect .git, resolve origin, compare remote HEADs, or persist timestamps locally. These instructions grant the skill unnecessary awareness of repository layout and remote metadata, broadening its operational privileges and creating a pathway for unintended environment probing and side effects.
The skill directs hidden reads and writes of a .last-update-check file without any user-facing notice or consent. Undisclosed local file modification is dangerous because it creates covert persistent state, violates transparency expectations, and may interfere with host environments or auditing assumptions.
这是一个 markdown 文件，SQP-3 适用于所有文件类型。全文以中文编写且未见任何语言/locale 选择、替代版本说明或面向特定地区/语言用户的明确限定，可能构成未获用户选择即默认特定语言的语言政策问题。
SQP-3 applies to all file types and covers natural-language policy violations such as forcing a specific language without user opt-in. This markdown file presents all instructional and analytical content only in Chinese, with no indication that the skill is region-specific or that users can choose another language.
No suspicious patterns detected.
