Security audit
Security checks for vulnerabilities and agentic risk
The skill discloses its cloud image generation and saved API key, and requires confirmation before generating each image.
Expect Chinese-language output. Confirm each image prompt and understand that prompts and selected media go to dLazy, where generation may incur charges. Login saves your API key locally. The contradictory workflow text and stale version references should be clarified if you want a writing-only setup.
The code performs an authentication-configuration operation that the generic description does not disclose. The supplied chunk does not establish credential exfiltration, network access, or how the CLI stores the key.
Authentication configuration is the code chunk's sole purpose, not a supporting implementation detail of any described functionality. The generic description does not disclose this capability. The chunk alone does not establish credential exfiltration or network activity.
The generic description does not declare any authentication purpose, while the supplied shell command explicitly invokes a login operation. The snippet does not establish how authentication works or whether credentials are stored or transmitted.
The generic description does not meaningfully represent the supplied command’s external CLI execution capability. The placeholder prevents determining any specific downstream operation, and the snippet does not establish malicious behavior or particular data access.
The generic description does not disclose that the skill invokes a login command. This is a substantive capability rather than a supporting implementation detail. The supplied code does not establish how authentication works or whether credentials or network resources are accessed.
The generic description does not represent the snippet’s concrete behavior of invoking an external CLI. The command placeholder and absence of the package source prevent determining its downstream actions; no additional capabilities or malicious behavior can be established from this snippet.
The instruction to always conduct adaptation and conversation in Chinese imposes a specific output language regardless of the user's preference. The skill provides neither a language choice nor a documented justification requiring Chinese for its general-purpose adaptation workflow.
The appended mandatory image-generation instructions expand a conversation-only writing task into terminal execution and transmission of prompts to an external service. Although the workflow requires user confirmation and discloses the cloud service elsewhere, conflicting capability and workflow instructions can cause unintended tool use, disclosure of unpublished novel content, or billable API activity beyond the original adaptation task.
The frontmatter describes the skill only as “技能” (“skill”) and leaves triggers empty. Although the body describes webtoon adaptation, the activation metadata provides no conditions for deciding when to invoke the skill.
L089 states that the agent cannot read or write local files and that inputs and outputs remain in the conversation. However, L015 and L025 prescribe CLI authentication commands that explicitly persist an API key to local configuration (L018 and L028), so the documented workflow includes filesystem side effects despite that restriction.
“Always conduct adaptation and conversation in Chinese” imposes a specific language regardless of the user's preference. The skill neither offers a language choice nor documents a justification requiring Chinese-only operation.
No suspicious patterns detected.
