Security audit
Security checks for vulnerabilities and agentic risk
This skill would persistently route OpenClaw model traffic and an API key through a third-party NewCLI proxy, with broad activation wording and promotional links that warrant review.
Install only if you intentionally want OpenClaw to use NewCLI/FoxCode as a model proxy. Independently verify the provider and endpoints, use a dedicated scoped key where possible, avoid sensitive test prompts, back up your OpenClaw config before changing fallback routes, and remove referral-driven or unapproved routes from the setup.
The trigger phrases in the Chinese usage instructions are very broad, including generic requests like '加 Claude', '加 GPT', and '配 newcli'. In an agentic system, such ambiguous activation phrases can cause this skill to fire during ordinary conversation or unrelated admin tasks, leading to unintended provider reconfiguration or steering users toward this third-party service and referral flow.
The English invocation examples use highly generic phrases like 'add Claude', 'add GPT', and 'add Gemini', which are likely to appear in normal operational dialogue. Because this skill configures external model providers, accidental activation could modify system behavior, route traffic to an unintended third-party endpoint, or expose administrators to referral-driven configuration choices without deliberate approval.
The trigger phrases are broad enough to activate on common admin requests such as adding Claude, GPT, or a provider source, which increases the chance the skill runs in contexts the user did not specifically intend. Because the skill instructs transmission of API keys and test prompts to an external proxy service, accidental invocation materially raises exposure risk.
The skill tells the operator to test models by sending prompts and an API key to code.newcli.com, but it does not clearly warn that this is a third-party proxy that will receive credentials and request content. In an admin workflow, omission of this disclosure can lead to uninformed trust decisions and unintended data sharing to a non-primary vendor.
This skill explicitly instructs users to transmit API credentials and test requests to an external service endpoint. External transmission is contextually expected for provider configuration, but it is still security-relevant because the destination is a proxy service and the document does not provide adequate trust, privacy, or secret-handling guidance.
| API Key | 格式为 `sk-ant-oat01-...` 的密钥（三个 provider 共用同一个） |
| 可用模型列表 | 向服务商确认，或查看账户统计页 |
**注意**：NewCLI 的 `/v1/models` 接口有 Cloudflare 防护，无法通过 curl 直接获取模型列表。请以账户后台的实际统计为准，不要依赖文档中的"支持列表"——文档列出的模型不一定全部对你的账户开放。
---
The skill embeds a promotional referral/signup link unrelated to the operational task of configuring providers. While not code-execution or credential-theft by itself, it creates an unnecessary social-engineering element that can bias operators into visiting an unvetted third-party page from within trusted admin guidance.
Distributing a referral URL is not justified by the stated purpose of the skill and introduces a conflict between user interest and author benefit. In security-sensitive operational documentation, such unrelated outbound links increase phishing and trust-abuse risk because admins may assume every included URL is required.
No suspicious patterns detected.