Security audit
Security checks for vulnerabilities and agentic risk
This is a documentation-style helper for Tavily/web-search workflows; its main issue is overly broad trigger wording, not hidden code or harmful behavior.
Before installing, consider narrowing or disabling implicit triggers if you only want this skill used for explicit Tavily or web-search workflow requests. There is no evidence of hidden execution, credential handling, persistence, or destructive behavior in the inspected artifacts.
The skill description and activation guidance are broad enough to match many ordinary requests involving work, web, search, or APIs, even when the user did not intend to invoke this specific skill. Over-broad routing can cause unintended skill activation, producing irrelevant or unsafe behavior and increasing the chance that downstream actions are taken under the wrong assumptions.
The keyword list includes very generic terms like "web," "search," "api," "user," and "bug fix," which are common across unrelated tasks. In a skill-routing environment, this can lead to frequent false activation, hijacking normal requests and potentially steering users into workflows or outputs that do not fit their intent.
The manifest description says to use the skill whenever a user mentions broad concepts like work-productivity, tavily, web, search, or via, which makes the invocation boundary ambiguous. In an agent environment, this kind of overbroad routing increases the chance that the skill is selected for unrelated tasks, causing prompt-scope confusion and reducing control over which instructions govern the interaction.
The trigger keyword list includes very generic terms such as "web", "search", "api", and "user", which are likely to appear in many unrelated prompts. This can cause unintended skill activation, pulling the agent into workflows the user did not request and potentially overriding more appropriate skills or injecting irrelevant instructions into the response path.
The invocation guidance defines only positive examples and gives no boundaries for when the skill should not run, which increases the chance of overmatching during routing. Because this skill targets common concepts like work productivity, web, search, and bug fixing, the lack of scope constraints makes accidental or inappropriate activation more likely in normal conversations.
The trigger sentence is phrased as a natural-language request ('I need a practical workflow...') that overlaps with ordinary user speech, making accidental invocation plausible. In agent systems, broad triggers can cause the wrong skill to activate, leading to unintended tool use, irrelevant workflow execution, or unsafe routing when a user did not explicitly request this skill.
The trigger phrases are generic natural-language requests such as 'Help me' and 'I need a practical workflow', which can overlap with ordinary user conversations and cause the skill to activate unintentionally. In an agent ecosystem, accidental activation can route tasks to the wrong workflow, leading to confusing behavior, inappropriate tool use, or processing of user data outside the intended context.
The example trigger phrases are conversational and vague, so they normalize activation on partial or common language rather than precise intent. This makes the routing surface noisier and can cause the system to select this skill for unrelated productivity or troubleshooting requests.
The example trigger phrases use generic lead-ins like "Help me" and "I need a practical workflow", paired with broad requirement text rather than realistic bounded requests. These examples encourage loose matching behavior and reinforce accidental activation patterns, though the impact is somewhat lower because examples are secondary to the manifest and keyword definitions.
The default prompt is broad and includes many generic terms, which can cause the skill to be invoked in situations the user did not clearly intend. In an agent environment, unintended invocation can leak context into the wrong workflow, produce irrelevant actions, or cause the system to apply this skill too aggressively across unrelated requests.
The trigger examples are so broad and natural-language-like that they can cause the skill to activate for ordinary user requests unrelated to the intended Tavily workflow. In an agent routing context, overbroad activation can misapply this skill, leak irrelevant internal context into responses, and reduce safety by bypassing more appropriate specialized handling.
This Chinese README contains substantial English-only instructional content such as the demand and workflow descriptions, but does not tell users they can choose their preferred language. That can amount to an implicit language-policy constraint without user opt-in.
The natural-language description and prompt include Chinese text ('搜索-style') embedded in otherwise English instructions, but there is no indication that the user can choose their preferred language or locale. This may impose a language assumption without explicit opt-in.
The file mixes English with Chinese text such as "搜索" and "各位懂设计的老哥进来帮个忙——手机端首页三选一," but does not explain whether the skill supports multiple languages or how language selection should occur. This can be a natural-language policy concern if the skill implicitly assumes or forces locale behavior without user opt-in.
