Security audit
Security checks for vulnerabilities and agentic risk
This is a mostly purpose-aligned anonymous chat skill, but its activation scope and disclosures are too broad for a tool that sends chat content and identity tokens to an external service.
Install only if you intentionally want an agent to use ctxly.app for anonymous chat. Do not send secrets, credentials, regulated data, or high-impact instructions through this skill, and require explicit user confirmation before any room token, invite, or message content is sent externally.
The skill uses very broad 'when to use' language, making it likely to activate in loosely related AI, LLM, or automation contexts. In practice, that can cause unintended invocation of an external anonymous chat service, leading to unnecessary transmission of prompts, tokens, or sensitive workflow data to a third party.
The skill describes sending and receiving chat data and treating tokens as identity, but it does not clearly warn users that messages, invites, and bearer-like tokens are transmitted to an external HTTP service. In this context, the omission is significant because the service is anonymous, lacks identity verification, and the free version explicitly lacks encryption and trusted communication guarantees, raising the risk of sensitive data leakage or impersonation.
The skill is described as an anonymous chat tool for room creation, joining, message send/receive, unread checks, and invite retrieval. Line L052 states the capability supports '创建/查询/导出操作' (create/query/export), but the rest of the document explicitly says the free edition does not support history export and focuses on core communication only, creating a direct documentation-level contradiction.
The file describes a lightweight anonymous chat skill backed by specific HTTP endpoints for messaging workflows. L059 says '支持修改/重置/导入操作' (modify/reset/import), which contradicts the stated chat-oriented scope and documented endpoints, and no corresponding capability is otherwise described.
L066 says the output stage supports '导出/保存/转换操作' (export/save/convert). However, L089 explicitly says the free version does not support message persistence or history export, so the documentation contradicts itself about whether exporting is available.
The capability coverage section enumerates an excessively broad set of trigger keywords, including generic terms like Agent, Use, 模型调用, 智能对话, and LLM. This increases the chance of accidental or manipulative activation, which is dangerous because the skill performs external HTTP communication and handles identity-bearing tokens without strong authentication.
