Security audit
Security checks for vulnerabilities and agentic risk
This is a Markdown-only prompt engineering guide with a minor caution around refusal-troubleshooting advice, but no hidden code, persistence, or data access.
Install this as a prompt-engineering reference skill if you want Claude-focused prompt help. Be careful with the refusal-troubleshooting section: clarifying legitimate intent is reasonable, but do not use it to bypass safety limits or obtain harmful instructions.
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.
- Explain methodology before results
### Don'ts
- Don't make predictions beyond 12 months without caveats
- Don't ignore outliers without investigation
- Don't present correlation as causation
- Don't use jargon without explanation
The troubleshooting advice explicitly suggests reframing a refused harmful request using fictional context and a 'legitimate purpose' explanation. That pattern can help users bypass model safety controls by laundering disallowed intent into ostensibly acceptable framing, especially in a prompt-engineering skill whose purpose is to optimize model behavior.
This markdown file describes the skill in very general terms such as 'Advanced expert in prompt engineering' and 'provides comprehensive guidance' without specifying narrow trigger phrases, scope boundaries, or negative examples. In a skill-dispatch system, such broad wording can overlap with many ordinary prompt-writing or AI-help requests and may cause unintended invocation.
This markdown repeatedly frames troubleshooting instructions around 'Claude' specifically, such as 'Claude invents facts' and 'Ask Claude to cite sources.' That can violate language/locale-style organizational policy constraints on forced platform choice when no opt-in or justification is provided.
No suspicious patterns detected.