Security audit
Security checks for vulnerabilities and agentic risk
This is a documentation-only helper for unit-test coverage work; its activation wording is broad, but I found no hidden execution, credential handling, persistence, or destructive behavior.
Installers should be aware that this skill may be invoked broadly for testing-related requests because implicit invocation is enabled and the trigger terms are general. It appears safe for unit-test and coverage workflows, but its publisher should narrow the trigger wording and improve the Chinese localization for clearer routing.
The invocation guidance does not define clear boundaries for when the skill should or should not run, and it lacks negative examples. In an agent environment, ambiguous routing guidance increases the chance of misfire, where unrelated user requests are handled by this skill, potentially causing unintended actions or exposing internal context to an unnecessary workflow.
The trigger sentence uses a very broad natural-language phrase ('Help me...') that can cause the skill to activate in situations far beyond its intended scope. Overly permissive triggering can lead to inappropriate routing, accidental invocation, and prompt-context collisions with other skills, which is a security and reliability concern in agent systems.
The trigger phrases are extremely broad and closely mirror common user requests about testing and software help, which can cause the skill to activate outside its intended scope. Over-broad activation increases the chance of prompt-routing mistakes, unintended handling of unrelated requests, and expansion of the skill's effective authority in multi-skill environments.
The description and usage guidance cast a wide net over generic software-help requests such as 'software-and-data,' 'testing,' and 'implementation support,' which can cause the skill to activate outside its intended niche. Over-broad routing is dangerous because it can preempt more appropriate, specialized skills and lead to incorrect or lower-safety guidance being applied in unrelated contexts.
The keyword list contains very generic terms like 'testing,' 'regression,' and 'quality' that are common across many software requests and do not uniquely identify this skill. This increases the chance of unintended invocation, misrouting user requests, and bypassing better-matched workflows or controls from other skills.
The example trigger sentences use broad everyday requests like 'Help me' and 'I need a practical workflow,' which are not sufficiently scoped to unit-test coverage work. Ambiguous examples can bias routing systems toward selecting this skill for unrelated tasks, causing confusion, poor task handling, or inappropriate generation paths.
The skill description defines eligibility using a very broad set of categories and artifact types, without clear constraints on scope or exclusions. This makes the skill more likely to be selected outside its intended domain, which can lead to irrelevant guidance, accidental override of a better-matched skill, or expanded attack surface through unnecessary prompt exposure.
The trigger list includes broad generic terms such as testing, regression, and quality, which can cause the skill to activate in many unrelated software conversations. Over-broad activation increases the chance of inappropriate context injection, workflow hijacking, or user confusion, especially in agent systems that auto-select skills from keywords.
The default prompt uses a very generic trigger phrase around common concepts like software, unit tests, coverage, and testing, which can overlap with ordinary user requests and cause unintended implicit invocation. Because implicit invocation is enabled, a normal request about testing could automatically activate this skill when the user did not explicitly ask for it, increasing the chance of prompt injection exposure or unexpected behavior.
The trigger sentence uses very broad natural language ('Help me ...', 'I need a practical workflow ...') tied to a common software task, which can cause the skill to activate on ordinary conversation rather than explicit invocation. In an agent system, unintended invocation can misroute user requests, override more appropriate skills, or inject this skill's workflow into unrelated contexts, reducing reliability and potentially affecting downstream actions.
This is a Chinese README, but key user-facing descriptive and trigger content remains in English, which can impose a language expectation without explicit opt-in. The document does not explain that English content is intentional or offer users a language/locale choice.
