Security audit
Security checks for vulnerabilities and agentic risk
This is a markdown-only anime recommendation skill with source-checking guidance and optional local features, with minor language and privacy caveats.
Installers should expect this skill to steer anime recommendations toward Chinese-first, source-grounded answers. Use the optional local CLI/cache/profile features only if comfortable storing anime history, preference feedback, and possible Bangumi/AniList tokens locally, especially on shared or backed-up machines.
The invocation condition is broad enough to match many ordinary anime recommendation requests, which can cause the skill to activate when the user did not explicitly ask for this specialized workflow. Over-broad triggering can override user preferences, steer responses into unnecessary source-grounded or Chinese-first behavior, and create unintended tool use or browsing exposure.
Mandating a Chinese-first response style without user opt-in can cause the assistant to ignore the user's language preference and reduce transparency or usability. In a multi-skill environment, this is a behavioral integrity issue because the skill can silently alter output language and framing even when the request was made in another language.
A fixed Chinese example output reinforces a default behavior that may cause non-Chinese user requests to be answered in the wrong language. While this is lower severity than direct prompt injection, it still creates a consistent risk of preference misalignment and can bias downstream agent behavior toward an unsolicited language switch.
The document defines the skill as '中文优先' and frames the product around Chinese-first behavior. Under the policy, forcing a specific language or locale without offering user choice is a natural-language policy violation unless clearly optional or justified as a region-specific tool.
The plan explicitly states that the local client will store viewing history, preference feedback, Bangumi/AniList tokens, and personalized profiles, but it does not pair that retention with an explicit privacy notice, consent flow, retention policy, or protection guidance. Even for a local-only tool, these artifacts are privacy-sensitive and could expose identity-linked media preferences or reusable API credentials if the machine is shared, backed up insecurely, or compromised.
No suspicious patterns detected.
