Security audit
Security checks for vulnerabilities and agentic risk
This skill provides disclosed shopping-comparison advice and does not include hidden code execution, account access, payment actions, or persistence.
Install this if you want a China-market shopping comparison advisor. Treat its output as advice, verify live prices and seller details yourself before buying, and do not provide account credentials or authorize it to place orders or send payment details.
The README is written entirely in Chinese, including the title, description, and all example interactions, with no indication that other languages are supported or that Chinese is optional. Under the policy for natural-language violations, forcing a specific language without user opt-in is a reportable locale/language constraint.
The instruction 'Sound like a decisive Chinese internet shopping advisor' imposes a specific language/locale presentation on all outputs. The file does not offer the user any language or locale choice or indicate that this constraint is optional or region-specific.
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.
- mark any assumptions clearly
Stop before:
- logging into the user's account without consent
- claiming access to private order history
- placing irreversible orders
- sending purchase messages or payment details
The manifest describes a cross-platform buying decision skill for comparing the same product across marketplaces and outputting an optimal purchase path. In Scenario 2, the skill is documented as providing a general Double 11 shopping playbook, including historical price tracking via third-party apps, cart management, presale deposit tactics, and broad deal-hunting strategy, which goes beyond same-product cross-platform comparison.
The default prompt advertises a broad invocation phrase ('Use $buying to compare...') without any trigger constraints, scoping rules, or guardrails about when the skill should activate. In an agent environment, this can cause over-invocation or prompt-surface expansion, increasing the chance the skill is called in unintended contexts and influences buying recommendations without clear user intent.
The manifest description states the skill will '输出最优购买路径' and is written entirely in Chinese, indicating a Chinese-language interaction model without any mention of user language choice. This can violate language/locale policy guidance when a skill appears generally applicable but implicitly enforces one language.
The instruction 'Use this file when the user wants a recommended route rather than a raw winner' does not define clear trigger phrases, scope boundaries, or negative examples. This leaves substantial ambiguity about when the skill should activate versus when another purchasing-comparison behavior should be used.
The file instructs the model to prefer Chinese output phrases regardless of the user's language, which can override user preference and system-level localization expectations. While this is not a direct code-execution or data-theft issue, it is a prompt-control weakness: embedded content is steering output format in a way that may conflict with higher-priority instructions and degrade safe, reliable behavior across users.
The short description is written in Chinese while the default prompt is in English, which implies a language choice has been made by the skill rather than the user. The file does not state that the user can choose their preferred language or that this mixed-language behavior is intentional and justified.
This markdown file consists entirely of Chinese example prompts and does not indicate that language choice is optional or configurable. Under the policy rule, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is clearly documented and justified.
No suspicious patterns detected.