Security audit
Security checks for vulnerabilities and agentic risk
GlobalWire is a disclosed news-formatting skill with limited local-file saving guidance and no evidence of hidden code, credential access, exfiltration, or destructive behavior.
Install this if you want news answers rewritten into a consistent wire-service style. Before using it, be aware that it may auto-apply to broad world-news prompts, it uses an opinionated source hierarchy, and it may create or update Markdown files under ./global-wire during archival workflows; ask for inline-only output or a specific save path if that matters.
The skill description and examples are broad enough that ordinary user requests about world news could unintentionally trigger this skill even when the user did not explicitly ask for wire-style formatting. Over-broad activation can cause prompt-scope confusion, unexpected instruction injection into unrelated conversations, and unnecessary loading of additional skill rules or reference files.
The skill enables implicit invocation while providing only a broad, generic transformation prompt for major world events. This can cause the agent to auto-apply the skill in loosely related contexts, potentially reframing outputs without clear user intent and increasing the risk of unintended influence over sensitive or fast-moving news responses.
Line L067 directs that Chinese output must localize labels and forbids leaving certain labels in English unless the user explicitly requests bilingual formatting. This imposes a language-specific formatting policy in natural language, and the file does not explain a region-specific necessity or present broader language selection options.
The saving rules authorize filesystem changes but do not require a clear warning, confirmation step, or opt-in from the user before modifying the workspace. In a zero-configuration public skill, this is more dangerous because it encourages write behavior in arbitrary environments where users may not expect persistent changes or may lack context about where files will be created.
The file explicitly permits saving even when the user did not ask to save, based on the vague condition that the workflow is 'clearly archival or repeatable.' That creates unauthorized file-modification behavior, which can surprise users, alter workspaces without consent, and be chained with broader agent capabilities to persist data or clutter project directories.
The instructions require that Chinese output must not include English section labels or confidence/watchlist headers unless explicitly requested. This sets a language-format constraint by default rather than offering a neutral user choice, which can be a natural-language locale policy issue under the stated rule.
The file hard-codes a credibility hierarchy that systematically demotes Chinese sources as a class while elevating a mostly Western/English-language outlet set, creating a baked-in editorial bias rather than a transparent, context-sensitive sourcing policy. In a news-briefing skill, this can skew summaries, verification judgments, and timeline construction toward one geopolitical perspective, causing selective omission or underweighting of relevant primary reporting.
