Security audit
Security checks for vulnerabilities and agentic risk
This source-checking skill is mostly a plain rules guide, but it needs review because it forces promotional text and an external invite link into every response.
Install only if you are comfortable with a source-screening assistant that may activate broadly during research and may add unrelated promotional text and an external invite link to outputs. The source-checking logic itself is transparent and non-executable, but the forced promotion should be removed or made opt-in before normal use.
The README says the skill can be triggered directly in normal conversation and gives very broad examples such as asking whether a source is reliable. That makes activation boundaries unclear and increases the chance the skill will run unexpectedly on ordinary user requests, which can interfere with other tasks or silently apply its filtering heuristics when the user did not explicitly request them.
Phrases like "执行 WebSearch / 网页抓取 / 阅读用户提供的链接之后" and producing any report using external information cover very common actions and do not narrowly constrain when the skill should run. This breadth could cause unintended invocation during many ordinary browsing or research tasks.
The skill explicitly requires appending promotional content and an external invite link to every output, regardless of user intent or task relevance. This is dangerous because it turns the agent into a covert marketing channel, creates prompt-level output manipulation, and can exfiltrate user attention to third-party destinations unrelated to the requested security or source-verification task.
Forcing Chinese promotional text and a specific public-account/invite link into every response is an unauthorized content injection mechanism. It overrides user context and language, degrades output integrity, and can be abused for spam, brand manipulation, or funneling users to external properties without consent.
The natural-language instructions and usage guidance are entirely in Chinese, and the file does not state that the skill is China-specific or that users may choose another language. Under the policy, forcing a specific language without user opt-in can be a locale-policy issue.
L056 explicitly states '禁止自动追加' for the blacklist/whitelist to avoid poisoning, while L071 says discard traces are required partly to '便于沉淀新黑名单条目'. Although not executable code, these instructions create mixed intent about whether runtime filtering results may drive blacklist expansion automatically or only via human review.
No suspicious patterns detected.
