Security audit
Security checks for vulnerabilities and agentic risk
The skill is a transparent multi-agent orchestration helper, but its broad activation rules and trading recommendation workflow need careful review before installation.
Install only if you intentionally want multi-agent delegation. Avoid using it with secrets, private logs, or sensitive business data unless you are comfortable sharing scoped context with multiple worker agents, and treat any trading output as informational rather than financial advice.
The trigger criteria are overly broad and include catch-all language such as use for 'any complex task decomposable into parallel workstreams.' That can cause the skill to activate in contexts where multi-agent delegation is unnecessary or inappropriate, increasing the chance of unintended invocation, excess tool use, and propagation of sensitive user data to multiple sub-agents.
The progress update examples instruct user-facing broadcasts containing Chinese text without any indication that the user's locale or language preference was checked first. While not a direct code-execution issue, this can degrade reliability and trust, confuse users, and produce unauthorized language switching in customer-facing output.
The trigger phrases for the code review pattern are broad enough that ordinary user requests could unintentionally activate the multi-agent workflow. In an orchestration skill, ambiguous activation boundaries can cause unintended delegation, expanded context sharing across workers, and execution of a more powerful workflow than the user explicitly requested.
The market research triggers are generic and could match many normal research requests without clear user intent to invoke multi-agent orchestration. This increases the chance of over-collection, unnecessary parallel processing, and broader propagation of user prompts or sensitive business context to multiple worker roles.
The trading trigger includes common language like asking whether to enter a trade, which can easily overlap with casual financial discussion. In this context, accidental activation is riskier because it routes users into an agent workflow that produces actionable trading analysis, potentially amplifying reliance on unvetted automated recommendations.
The trading workflow explicitly instructs the orchestrator to produce BUY/HOLD/SKIP recommendations with confidence and reasoning, but it provides no warning, limitation, or user-facing disclaimer. That is dangerous because it frames the system as giving actionable financial advice, increasing the likelihood of harmful reliance, regulatory exposure, and inappropriate use in high-stakes decisions.
Document-generation triggers such as "write a report on X" are very broad and likely to overlap with ordinary writing assistance. This can lead to unintended use of a more complex workflow, including unnecessary sharing of user content across researcher, writer, and editor roles, even when a simple single-agent response would suffice.
The debugging triggers are broad and could capture many routine support requests, causing the orchestration layer to activate without clear intent. In debugging contexts this is meaningful because logs, stack traces, and source snippets may contain secrets or proprietary details that would then be distributed to multiple workers unnecessarily.
Natural-language policy review applies to all files. The role patterns present bilingual section titles and trigger phrases, but the file does not state whether users may choose their preferred language or whether the bilingual format is intentional and required for a specific audience.
