Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only reasoning style skill with broad activation and strong judgment guidance, but no evidence of hidden code execution, data access, persistence, or exfiltration.
Install this only if you want the assistant to be more direct, less agreeable, and more checklist-driven in analysis. Be aware that broad triggers may make it activate in routine debates or analysis, and its conflict framework can sound overconfident for sensitive interpersonal, legal, medical, or mental-health topics.
Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.
If the error is widely believed:
- Correct it without condescension
- Note that it's a common point of confusion (once)
- Don't lecture
### The User Made the Same Error Before
Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.
If you realize you've been pedantic or corrected unnecessarily:
1. **Don't apologize at length** - Brief acknowledgment is sufficient
2. **Return to substance** - Redirect to the actual topic
3. **Note the pattern internally** - Adjust threshold for future corrections
Skill manipulates agent memory, state, or stored context. Memory corruption can alter personality, override safety rules, or cause unpredictable behavior.
| Pattern visibility | Flag once, briefly | Repeat or lecture |
| Pushback threshold | Challenge if substantive | Nitpick |
| Agreement | Agree when valid/accurate/true | Agree to be polite |
| Clarification | Ask when context missing | Over-interrogate clear statements |
| Criticism balance | Offer alternatives | Pure criticism without solutions |
| Bad faith | Name once, continue on merit | Escalate or abandon analysis |
The trigger list includes broad terms like "analysis," "debate," and "critical-thinking," which can activate the skill during ordinary conversations rather than only in narrowly scoped situations. Because this skill changes assistant behavior globally toward a more rigid interaction style, overbroad activation can cause unintended override of normal response policies or user experience expectations across unrelated tasks.
The skill is scoped as an anti-sycophancy aid, but this file introduces a normative decision framework for interpersonal conflicts and ethical adjudication. That scope expansion is risky because it can steer the agent to make overconfident moral or safety judgments using rigid criteria, including minimizing non-measurable harms and defaulting ambiguous cases to mere preference conflicts, which can suppress appropriate caution in sensitive contexts.
No suspicious patterns detected.