Security audit
Security checks for vulnerabilities and agentic risk
This Markdown-only skill does not show malicious behavior, but it needs review because it can be triggered too broadly and instructs the agent to create a project PRD file despite declaring only read access.
Install only if you are comfortable with the skill searching project markdown and web sources for startup validation and potentially creating docs/prd.md. Prefer invoking it explicitly, and require the agent to ask before writing or overwriting any project file.
The trigger keywords include very broad terms such as 'through', 'validate', 'score', 'startup', and 'idea', which are common in ordinary conversation. Overbroad activation increases the chance the skill runs unintentionally and, in this case, may lead to unexpected research steps or file-writing behavior.
A second repeated trigger list reinforces the same vague activation pattern, compounding the likelihood of accidental routing. Repetition of broad triggers makes unintended execution more probable across agent implementations that rely on keyword matching.
The summary and description are presented partly in Chinese while the rest of the skill is mixed-language, but there is no statement that the user can choose their preferred language or locale. This can violate language/locale policy when a specific language is imposed without opt-in.
The skill goes beyond validating an idea by selecting implementation stacks and steering toward scaffolding/build planning. That broader operational behavior can push an agent from advisory analysis into action-oriented development decisions that the user did not clearly request from the skill’s description.
The skill is presented as an idea-validation tool, but it also instructs the agent to create and write a PRD file in the project directory. This expands scope from analysis into filesystem modification without making that side effect prominent up front, which can surprise users and cause unintended project changes.
The skill instructs creation of docs/prd.md without a clear user warning that project files will be modified. Unannounced writes are dangerous because users may invoke what appears to be an analysis skill and end up with unsolicited repository changes.
The 'When to use' section is broad and does not clearly constrain activation to explicit user intent. In combination with the skill’s ability to perform research and generate a PRD, this makes accidental invocation more likely and increases the chance of unexpected side effects.
The manifest description emphasizes evaluation through specific analytical frameworks, but the file additionally instructs the agent to perform external web research on failed startups and competitors. That research capability may support the analysis, but it is not declared in the manifest and is broader than the narrowly stated framework-based scoring description.
No suspicious patterns detected.
