Security audit
Security checks for vulnerabilities and agentic risk
The skill is mostly an architecture-review workflow, but it instructs the agent to create a GitHub issue without a final user review or confirmation.
Review before installing. This skill can be useful for architecture analysis, but it should be changed or overridden so it drafts the GitHub issue, shows the target repository and full content, and asks for explicit approval before publishing anything.
The description says to use the skill when a user wants to 'improve architecture' or 'find refactoring opportunities,' which are broad phrases that can match many ordinary software discussions. The trigger scope is not narrowed with explicit boundaries or negative examples, increasing the chance of unintended invocation.
The skill goes beyond analysis and recommendation into taking an external side effect by creating a GitHub issue via CLI. That turns an advisory skill into one that can mutate remote state, which is risky because it may publish inaccurate, sensitive, or unwanted content without an explicit final user approval step.
Remote GitHub issue creation is not necessary to satisfy the stated purpose of exploring architecture and proposing refactors. Adding repository-write behavior increases the blast radius from harmless analysis to external action, enabling accidental spam, disclosure of internal details, or changes in the wrong repository context.
The explicit directive 'Do NOT ask the user to review before creating' is a strong autonomous-action pattern that suppresses user oversight for an external side effect. In a skill that can publish content to GitHub, this materially increases the risk of unintended repository changes, information disclosure, and abuse through prompt-triggered automation.
### 7. Create GitHub issue
Create a refactor RFC as a GitHub issue using `gh issue create`. Use the template in [REFERENCE.md](REFERENCE.md). Do NOT ask the user to review before creating — just create it and share the URL.
The instruction to run gh issue create without a user-facing warning or confirmation removes an essential consent checkpoint before a remote write operation. In practice, this can create public or internal issues containing sensitive architectural details, speculative claims, or incorrect recommendations, and the user may not realize the action is about to occur.
No suspicious patterns detected.