Security audit
Security checks for vulnerabilities and agentic risk
This is a disclosed research-figure workflow that uses image-generation tools and has no persistence or executable install behavior, with practical cautions around language defaults and external image services.
Use this skill when you want a structured paper-diagram workflow. Before rendering, avoid sharing unpublished or proprietary material with image providers unless that is acceptable, watch the default six-image generation behavior for cost or quota impact, and ask for your preferred language if the Chinese-first templates are not suitable.
The skill description and activation scope are very broad, covering common intents like designing or generating figures without strong invocation boundaries. In an agent marketplace or auto-routing environment, this increases the chance the skill is invoked for loosely related requests, causing unintended workflow takeover and image-generation escalation that the user did not clearly request.
Mandating Chinese-only section headers and prompt boilerplate without user language selection can override user intent and reduce transparency about what the system is doing. This is primarily a control and usability risk: users may approve prompts or workflow steps they do not fully understand, and downstream systems may receive language-constrained instructions that were not explicitly requested.
Treating short, ambiguous follow-ups like '继续', '生成', or 'generate' as sufficient authorization to produce multiple candidate images weakens user-consent boundaries. Such minimal phrases are common in conversation and can be misinterpreted as approval for costly or state-changing actions, especially when prior context is incomplete or the user did not intend image generation at that stage.
The skill hard-requires Chinese section headers and footer content in every text-only reply without checking or honoring the user's language preference. This can degrade usability, cause misunderstandings in safety-critical instructions, and create a prompt-quality/accessibility issue, though it is not a direct code-execution or data-exfiltration risk.
The example content is written entirely in Chinese and instructs the next interaction in Chinese without presenting a language-selection mechanism or documenting that Chinese is required for the skill. This can exclude or mislead users who operate in other languages, and in a workflow skill it may cause incorrect use, misunderstanding of state transitions, or unsafe operator decisions due to language barriers.
This markdown file contains natural-language operational guidance entirely in Chinese, including the plan, state, and suggested next-step prompts. Under the language/locale policy rule, forcing a specific language without user opt-in is a policy concern unless the locale limitation is explicitly justified.
The example startup turn is entirely in Chinese and presents the workflow, prompts, and state descriptions as the default interaction language without offering user choice. This can cause users to misunderstand safety-relevant workflow constraints or tool behavior, and may exclude users who cannot read Chinese, but it is primarily a usability and policy-compliance issue rather than a direct security exploit.
The example hardcodes Chinese-language workflow text without user opt-in or a documented locale requirement, which can cause users to misunderstand the current mode, step boundaries, or next actions. In a stateful workflow that governs when image generation occurs, language mismatch increases the chance of uninformed consent, mistaken execution, or bypass of important safety and privacy notices.
This JSON manifest describes the skill in very broad functional terms such as designing, prompting, generating, critiquing, and integrating research-paper figures, but it does not define explicit activation phrases, boundaries, or negative examples. For a manifest file, that can create ambiguity about when the skill should be invoked versus when general figure or paper-help requests should not trigger it.
The navigation policy hard-codes Chinese prompt templates as the recommended and fallback interaction pattern without any user opt-in or language negotiation. This can mislead users, reduce transparency, and cause prompt-routing errors or hidden workflow control when the user operates in another language, especially in a skill that tightly controls multi-step stateful behavior.
All starter messages are written in Chinese and implicitly steer interaction into Chinese without presenting any language choice or opt-in. This can degrade usability, exclude users who do not read Chinese, and increase the chance that users invoke the skill without understanding its constraints or workflow, though it does not directly create code-execution or data-exfiltration risk.
The contract requires fixed Chinese section headers in every text reply, regardless of the user's language or accessibility needs. This can override user preference, create confusion, and cause misleading or unusable outputs in multilingual contexts, especially for users expecting English-only or locale-consistent responses.
The template hard-requires a Chinese-language footer in every TEXT_ONLY reply without any user-language negotiation or documented necessity. This can cause the agent to ignore user locale preferences, reduce transparency, and create usability or accessibility failures that may mislead users about the agent’s state or instructions.
The file presents operational sections and invocation guidance in Chinese while the drafted caption text is in English, but it does not state that the user opted into bilingual output or that the skill is intended for a specific locale. This can violate language/locale policy by imposing a language mode without explicit user choice.
The document prescribes fixed response-format language such as IMAGE_ONLY and TEXT_ONLY and gives instructions only in English, with no indication that users may choose another language or locale. This can be read as a language-policy constraint without explicit opt-in or justification.
This markdown file contains operational instructions and suggested user prompts in Chinese, but does not indicate that the user can choose another language or that the skill is intentionally limited to Chinese. Under the policy rule for language/locale, forcing a specific language without user opt-in is a natural-language policy concern.
The example explicitly routes a later step to external image-generation surfaces and APIs but does not warn the user that content may be sent to external tooling or services. This creates a transparency and consent problem: users may disclose unpublished research details, figures, or sensitive prompts without realizing they are being transmitted beyond the text-only planning stage.
No suspicious patterns detected.
