Security audit
Security checks for vulnerabilities and agentic risk
This skill only provides reasoning-persona instructions, with no code execution or data access, though users should know it may change the assistant's tone or critique style automatically.
Install this if you want the assistant to apply stronger critique, precedent-checking, or integration-focused reasoning. Be aware that it may activate from broad contexts like brainstorming, code review, or architecture decisions, and it may not always label the active persona unless your agent or workflow requires disclosure.
The skill description and examples encourage broad activation based on vague user phrasing like brainstorming or review-oriented requests. In an agent system, this can cause the persona logic to trigger in contexts the user did not explicitly intend, subtly changing tone, reasoning style, or recommendations without clear consent.
The auto-activation map uses ambiguous labels such as brainstorming, code-review, and exploring new ideas without operational boundaries. In a larger agent framework, those broad categories can match many normal tasks, making the skill activate unexpectedly and influence outputs in ways that are hard to predict or audit.
Allowing the persona to run silently means the system may materially alter reasoning behavior without notifying the user. This reduces transparency and user agency, and makes it harder to detect when advice was shaped by an internally selected adversarial or highly critical framing such as Devil's Advocate or Gonzo mode.
The document describes activation prompts and usage but does not define firm boundaries for when persona activation should and should not occur. Without trigger scope and exclusion rules, benign references, quoted text, analysis of personas, or attacker-crafted prompts can induce unintended mode changes that alter reasoning style and reliability.
The activation wording is conversational and broad enough that ordinary discussion like mentioning a mode or persona could unintentionally trigger behavioral changes. In a reasoning skill, that can cause the agent to adopt a high-agency or overly adversarial stance without explicit user intent, leading to misaligned outputs or bypass of normal conversational safeguards.
No suspicious patterns detected.
