Security audit
Security checks for vulnerabilities and agentic risk
This is a documentation-help skill for OpenAPI/Swagger work, with broad trigger wording but no hidden code, persistence, credential access, or destructive behavior.
Reasonable to install if you want help with OpenAPI or Swagger documentation. Be aware that its trigger phrases are loose, so it may activate for some general API or developer-experience requests unless your runtime lets you restrict implicit invocation or choose skills explicitly.
The trigger phrases are broad, repetitive, and partially malformed, making it easier for the skill to activate on loosely related requests rather than clear user intent. In an agent environment, unintended invocation can route users into the wrong workflow, cause confusing outputs, or expose downstream tooling and prompts to requests that were not meant for this skill.
The trigger sentence is so broad and malformed that it can match ordinary user phrasing unrelated to an explicit skill invocation, increasing the chance of accidental activation. In an agent setting, overly permissive routing can cause the wrong skill to run, leading to unintended access to context, misleading outputs, or workflow hijacking by prompts crafted to resemble common speech.
The trigger phrases are broad and include generic wording like 'Help me' and 'I need a practical workflow', which can cause the skill to activate in contexts beyond explicit OpenAPI documentation requests. In an agent environment, ambiguous invocation increases the risk of unintended execution, context hijacking, or the skill handling requests it was not meant to process.
This is a zh-CN README, but the keywords and trigger phrases presented to users are in English only. That can effectively impose a language preference on users of the Chinese-localized file without any opt-in or explanation.
The skill description and usage scope are broad enough to match many routine API- or software-related requests without strong boundaries. Overly broad activation can cause the wrong skill to engage, leading to inappropriate handling of user tasks, accidental context capture, or reduced reliability in downstream agent behavior.
The keyword list contains broad technical phrases like 'software-and-data' and 'developer experience' that are common across many unrelated requests. This increases the chance of unintended activation and scope confusion, which can misroute user interactions and expose the agent to prompt-surface expansion from irrelevant contexts.
The skill description and activation conditions are broad enough to overlap with ordinary technical conversations about APIs, Swagger, or REST documentation. This can cause unintended routing to the skill, which may override a better-matched skill or inject irrelevant workflow assumptions into normal user interactions.
The keyword list is too unconstrained and contains common technical terms like 'openapi', 'swagger', 'api documentation', and 'rest api' without qualifiers. In a multi-skill environment, such generic keywords can cause frequent false activations and misroute user intent, reducing reliability and potentially exposing the user to incorrect or overly scoped automation.
The example trigger phrases use very broad natural-language constructions like 'Help me' and 'I need a practical workflow', which are common in normal conversation and do not clearly distinguish an explicit skill invocation. This increases accidental invocation risk and makes routing behavior ambiguous, especially when multiple skills could plausibly match the same request.
The default prompt is broad and generic enough that normal user requests about APIs, software, or documentation could unintentionally trigger this skill. That increases the chance of silent or unexpected invocation, which can cause prompt-context injection into unrelated conversations and reduce user control over when the skill is used.
Enabling implicit invocation without strong activation constraints allows the skill to be auto-selected in contexts where the user did not clearly request it. In a documentation-generation skill, this is risky because the model may inject specialized behavior or transform outputs unexpectedly, creating integrity and consent issues even if the skill is not overtly malicious.
Although the sentence adds framing, it remains ambiguous and still resembles natural-language help requests rather than a precise routing condition. This can let benign user messages or adversarially phrased prompts spuriously trigger the skill, reducing routing integrity and making prompt-injection style misdirection easier.
The file lists separate English and Chinese instruction and guide files, but does not state how language selection occurs or that the user can choose their preferred language. In the absence of an explicit opt-in or selection mechanism, this can be read as a locale-handling policy gap.
The example trigger sentences are malformed and do not clearly communicate safe activation boundaries, making it harder for routing systems or maintainers to understand intended scope. Ambiguous or broken examples can contribute to accidental invocation and inconsistent behavior, though the direct security impact is lower than broad trigger metadata.
