Security audit
Security checks for vulnerabilities and agentic risk
This is a text-only strategic thinking skill; it includes negotiation and influence framing, but it is disclosed, bounded, and does not request code execution or data access.
Install this as a strategy and planning aid, not as authority to manipulate people or bypass professional review. Be especially careful with the psychological game and resource-leverage sections: use them only for transparent, lawful collaboration and avoid high-stakes domains where expert advice or formal process is required.
The '空城计' model gives a stepwise framework for exploiting information asymmetry, shaping perceptions, and causing an opponent to second-guess themselves, which is a reusable deception pattern. In skill context, this is more dangerous because it is operationalized with triggers, prerequisites, and modern business examples, making manipulative conduct easier to justify and apply in real interactions.
This section presents 'borrowing resources' through tactics including lowering the other party's guard,诱骗, and designing scenarios to obtain resources from others, which normalizes manipulative social engineering patterns. Even with later 'win-win' framing, the procedural guidance can be repurposed to pressure, mislead, or extract cooperation from people who would not otherwise consent on fully informed terms.
The activation phrases are very broad and map common user problems to this skill, increasing the chance it will trigger in ambiguous contexts where safer or more domain-specific guidance should apply. Overbroad routing is risky because this skill includes manipulation and influence frameworks that could be surfaced for routine workplace, negotiation, or conflict scenarios without adequate guardrails.
Line L026 states the skill is a '纯文本思维框架' with no executable behavior, which is accurate literally, but elsewhere the document operationalizes tactics such as creating information fog, exploiting asymmetry, and lowering others' guard in concrete step-by-step form. This is not a code-execution contradiction, but it is an intent-level divergence between a harmless framing and the substantive advisory behavior provided.
No suspicious patterns detected.
