Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only medical exam tutoring skill with disclosed exam practice, scoring, and weak-point tracking behavior, but users should treat it as study support rather than clinical guidance.
Install this only if you want a strict Chinese-language assistant-physician exam tutor. Do not use it for real patient diagnosis or treatment decisions, and be mindful that weak-point tracking may retain exam performance details if your agent's memory features are enabled.
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
Output one daily high-yield summary for rapid review.
### Daily Output Rules
1. Start directly with a comparison table. No warm-up prose.
2. Focus on one confusion cluster per day (e.g., pneumonia subtype differentiation).
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
|---|---|---|---|---|---|
## Output Rule
1. Keep rows <= 6 unless user asks full expansion.
2. Keep each cell exam-focused and concise.
The skill forces a Chinese-language examiner persona and tone without checking user preference or language compatibility. This can degrade usability, create misunderstanding in safety-sensitive educational exchanges, and reduce the user's ability to notice boundaries or corrections.
The role contract tells the model to immediately interrupt and correct dangerous indications or contraindicated medication, which can produce concrete medical direction. In a medical-training skill, this conflicts with the later boundary against definitive real-world orders and increases the chance that users receive actionable clinical advice that exceeds an exam-tutoring scope.
The global output policy mandates Chinese row labels and table conventions for all knowledge delivery without user opt-in. While not directly exploitative, it can impair comprehension and informed use, especially when discussing high-risk clinical concepts.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
Never output any reasoning-question format.
1. Do not ask users to infer diagnosis from long case narratives.
2. Do not request multi-step chain-of-thought style answers.
3. Keep stems short, factual, and exam-conventional.
4. If a generated question looks like a reasoning item, regenerate it as direct objective-choice.
The first-use announcement advertises score memory, weak-point tracking, and trend tables, signaling persistent retention of user performance data beyond a single interaction. Cross-session educational profiling can expose learning history and potentially sensitive inferred attributes if accessed or reused improperly.
Real-time logging of every asked and missed item during SP sessions creates retention of detailed user interaction content that may include sensitive health-related text or personal disclosures. Without explicit minimization and retention controls, this increases privacy leakage risk and broadens the attack surface for downstream misuse.
Daily proactive drills combined with error aggregation and progress memory encourage broad collection of longitudinal user history. This expands the volume and continuity of stored behavioral data, increasing privacy and misuse risk if the system lacks clear boundaries and retention controls.
The skill explicitly defines cross-session scoring, weak-point memory, and rolling priority updates despite the manifest describing it as daily exam-style practice only. That mismatch can cause operators and users to underestimate the amount and duration of data handling, especially for performance history and potentially sensitive free-text inputs.
The output contract requires the final three lines to be exactly specific Chinese phrases, which imposes a fixed language on all outputs. The file does not offer user opt-in, a language selection mechanism, or any documented region-specific justification for this constraint.
The script hard-codes Chinese-only prompts, starter questions, and closing requirements without offering a language-selection step or adapting to the user's preferred language. In an exam-training skill, this can cause unsafe misunderstanding, exclusion, and degraded usability for users who are not Chinese-speaking, especially when precise clinical wording matters.
The output rule requires every response to end with the Chinese phrase 30秒速记：..., which imposes a specific language on users regardless of their preference. This is a natural-language policy concern because the file does not offer any language choice or explain a justified locale constraint.
This file provides procedural baselines for invasive and emergency medical actions, including catheterization, CPR, and thoracentesis, but does not warn that the content is only for training support and not a substitute for clinical judgment, local protocols, or qualified supervision. In the context of an exam-tutor skill, users may still over-rely on terse checklist content during real-world care, increasing risk of patient harm from incomplete, decontextualized, or misapplied steps.
The file specifies priority tiers using Chinese labels (高, 中, 低) and later defines workflow states and table headers in Chinese, which imposes a specific language on outputs. Because the document does not offer an opt-in choice or explain that the skill is intentionally region-specific, this is a natural-language policy concern.
The workflow requires statuses such as 未修复, 修复中, and 已稳定, which constrains the skill's operational language. This is a policy violation unless the user can choose the language or the file clearly documents a justified Chinese-only scope.
The required table format specifies Chinese headers (薄弱点, 频次(7d), 严重度, etc.), which directs the skill to output in a fixed language. Without an explicit locale opt-in or region-specific justification, this conflicts with the language/locale policy.
The rubric content is written entirely in Chinese for core scoring dimensions, and the feedback contract appears to assume that interaction/output will follow that locale. There is no indication that users can choose another language or that the Chinese-only constraint is required for a region-specific use case, which may violate language/locale policy guidance.
The oral defense rapid template is written entirely in Chinese and implicitly directs outputs in that language. The file does not state that the skill is region-specific or provide an opt-in or alternative language choice, which creates a natural-language locale policy concern.
The manifest frames the skill as something used for user-initiated exam practice, but the daily drill section instructs proactive initiation of timed drills. This creates a scope mismatch that may surprise users and can lead to unsolicited interaction patterns not reflected in the published description.
The templates use Chinese column headings and example content, which imposes a specific language on users of the skill content. There is no indication that Chinese is optional, user-selected, or required for a documented region-specific purpose.
This markdown file contains core user-facing inputs and table content in Chinese, while surrounding instructions are in English. Because the skill does not state that it is intended only for Chinese-speaking users or offer a language/locale option, it may violate the policy against forcing a specific language without user opt-in.
The substantive procedural content is primarily in Chinese while several section headers and scoring instructions are in English, and the file does not state that Chinese is required or provide a language/locale option. Under SQP-3, forcing or assuming a language without user opt-in can be a natural-language policy issue when no justification is given.
No suspicious patterns detected.