Security audit
Security checks for vulnerabilities and agentic risk
This is a prompt-writing skill with some broad and imperfect guidance, but it does not contain code, persistence, credential access, data exfiltration, or hidden behavior.
Before installing, review the broad activation triggers and the English default. The skill appears safe as a prompt-engineering helper, but users may want to narrow triggers and remove the two 'without asking' examples to avoid overconfident prompt generation.
The skill advertises activation for very broad, common intents such as creating prompts, improving prompts, designing assistants, and writing SKILL.md files. This can cause the skill to trigger in many unrelated conversations, increasing the chance that its instructions override more appropriate task-specific behavior and influence outputs in unintended contexts.
Stating 'Primary language: English' imposes a language preference in the skill definition, and the document does not indicate that users may choose another language or opt in to English. Under the policy, forcing a specific language without user choice is a natural-language policy violation unless clearly justified.
The guidance explicitly recommends building prompts 'automatically without asking,' which encourages the model to act on incomplete requirements and make autonomous decisions on the user's behalf. In a prompt- and skill-authoring context, that can embed unsafe assumptions, omit constraints, or generate instructions that overreach the user's actual intent.
**14. Meta Prompting**
- Use AI to design its own prompts. Shifts from asking for answers to asking for "the best way to ask the question." Elevates prompt engineering from personal knowledge to a repeatable, teachable system.
- Best for: Building organizational prompt standards, developing skills or system prompts.
- How to use effectively: Tell AI "you are an expert Prompt Engineer — when the user describes a task, analyze it and build the best prompt automatically without asking."
**15. Continuous (Soft) Prompts**
- Build prompts as structured data formats that computers understand — such as JSON or vectors — instead of natural language. Used to pass information between AI systems or automated pipelines.
The example prompt again instructs the AI to analyze and build a prompt 'without asking questions,' reinforcing autonomous behavior despite earlier sections that require collecting complete information first. This inconsistency can lead the skill to skip clarification, generate overconfident outputs, and normalize acting without sufficient user input.
Few-shot HR Chatbot: Provide example Q&A about leave, documents, and benefits before deploying.
Meta Prompting: "You are an expert Prompt Engineer. When the user describes a task, analyze and build a prompt using: Role / Context / Task / Constraints / Output — without asking questions."
Iterative Refinement: Critique previous response → list 5 weaknesses → suggest improvements → rewrite a better version.
No suspicious patterns detected.