Security audit
Security checks for vulnerabilities and agentic risk
This prompt-refinement skill is coherent and non-executable, with some ambiguity around auto-applying rewritten prompts that users should watch.
This skill is reasonable to install if you want prompts clarified before work begins. Be aware that 'just do it' is ambiguous in the instructions; use explicit wording such as 'show me the refined prompt first' or 'do not refine this' when you want a specific flow.
The skill description is broad enough to trigger on common user requests like improving, optimizing, or clarifying inputs, which can cause the skill to activate outside a narrowly scoped prompt-refinement context. Overbroad routing increases the chance of unintended interception of requests and can alter execution flow or user intent without a clearly bounded handoff.
Using vague phrases like 'just do it' as an auto-apply trigger is dangerous because such language is common in normal conversation and may not reflect informed consent to bypass confirmation. This can cause the skill to execute a refined prompt immediately when the user merely intended to proceed generally, not to waive review of the transformed prompt.
The auto-apply mode permits the skill to refine a prompt and immediately proceed without confirmation, which is a form of autonomous decision-making over user instructions. Because prompt refinement can materially alter scope, constraints, or output expectations, immediate execution reduces the user's opportunity to detect unwanted changes before downstream actions occur.
## Continue Modes
1. **Popup-confirm (default)** — Show refined prompt, popup to choose refined vs original, execute after choice
2. **Auto-apply** — When user says "just do it / skip confirmation", show refined then execute immediately
3. **Optimize-only** — When user only asks to refine without executing, return refined result only
## When NOT to Use
The continue-mode activation relies on ambiguous skip-confirmation phrases without strict boundaries, making it easy for ordinary user wording to be interpreted as consent to bypass review. In a skill that rewrites prompts and influences downstream actions, this ambiguity can silently change task instructions and reduce user oversight.
Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.
## Integration with session-learner
- When user chooses refined or rejects it, this is a **preference signal** for `session-learner`.
- `session-learner` should only summarize rules (e.g., "user prefers seeing refined version before confirming"), never record full prompt text.
- Over time, `session-learner` builds a preference profile that makes `prompt-refiner` increasingly aligned with user habits.
## Priority Rules
The markdown includes user-facing example prompts such as "帮我优化登录性能" and "帮我写个项目介绍" as the sole examples, which implicitly assumes a Chinese-language workflow. Because the file does not offer an opt-in language choice or explain that it is intentionally limited to Chinese users, this creates a language/locale policy concern.
No suspicious patterns detected.