Security audit
Security checks for vulnerabilities and agentic risk
This skill is a documentation-only helper for improving error messages, with broad invocation wording but no hidden code, persistence, credential use, or destructive behavior.
Before installing, be aware that this skill may be invoked for general debugging or support prompts because its triggers are broad and implicit invocation is enabled. It is best used deliberately for rewriting or designing clearer error messages; it does not appear to install code, access credentials, persist state, or perform actions on its own.
The skill description is broad enough to match many ordinary support, debugging, and productivity requests, which can cause the skill to activate outside its intended niche. Over-broad routing increases the chance of inappropriate invocation, prompt-context pollution, and accidental interception of user tasks that should be handled by a more specific or safer skill.
The trigger keywords and example phrases are extremely generic, including terms like 'debugging,' 'support,' and 'troubleshooting,' which are common across many unrelated requests. This makes accidental or overly aggressive activation likely, potentially causing the skill to engage on broad classes of user input and override better-scoped behavior.
The trigger sentence is so broad and awkwardly templated that it can match ordinary user requests about help, workflows, support, or error handling that were not intended to invoke this specific skill. In agent systems that auto-route by keyword or phrase similarity, this increases the chance of unintended activation, causing the skill to intercept unrelated conversations and shape outputs without clear user intent.
The invocation guidance lacks precise criteria for when the skill should run, and the listed examples are broad enough to overlap with many normal support conversations. This ambiguity can cause unauthorized or unintended skill execution, weakening least-astonishment and making prompt-injection or routing abuse easier because an attacker can phrase a normal-looking request to force the skill into scope.
The trigger sentence is generic enough to match ordinary user requests about help, workflows, or error handling, which can cause the skill to activate unintentionally. Over-broad activation increases the chance that the agent applies this skill in contexts where it was not explicitly requested, leading to prompt-routing mistakes and potentially exposing unrelated user content to unnecessary processing.
The trigger phrases are broad natural-language requests such as 'Help me' and 'I need a practical workflow', which can match ordinary user prompts unrelated to explicit skill invocation. That creates a prompt-squatting or unintended-activation risk where this skill may intercept requests unexpectedly, potentially overriding more appropriate routing or inserting its workflow into normal conversations.
The top-level description defines activation using a wide set of productivity and support-related categories without clear scope limits. Because the boundary is unclear, the skill may be selected for requests that are only loosely related, increasing the chance of misclassification, poor guidance, or interference with more appropriate skills.
The trigger section uses broad generic keywords such as 'debugging', 'support', and 'user feedback', plus very general example invocations. This can cause the skill to activate in many unrelated contexts, leading to inappropriate routing, over-application of the skill, and reduced reliability of downstream agent behavior.
The default prompt uses a broad, natural-language invocation phrase about helping with clearer error messages, which overlaps heavily with ordinary user requests in the same domain. Because implicit invocation is enabled, the agent may trigger unexpectedly during routine conversations about debugging or support, causing unintentional routing, prompt injection exposure through delegated context, or confusing behavior for users and downstream agents.
Despite being a zh-CN README, key user-facing content such as the demand description remains in English, and the trigger phrases later are also English-only. This can amount to an implicit language constraint without opt-in or justification, which conflicts with locale-choice expectations.
No suspicious patterns detected.
