Scientific Internet Access

AI-powered Scientific Internet Access engine for OpenClaw. AI驱动的科学上网术——你的私人科学上网管家。 自动抓取免费节点、测速、筛选，一步步引导小白完成配置。 官网: https://shadowrocket.ai 推荐搭配Claude模型使用，指令遵...

What to consider before installing: - The scripts will actively fetch many public subscription URLs and then attempt TCP connections to dozens of scraped hosts/ports to measure latency. This is required for the feature but can look like port scanning to network monitors and may trigger alerts or be disallowed by your environment. - Scraped node entries often contain raw connection strings (passwords, UUIDs, server addresses). The skill stores these in ~/.openclaw/workspace/nodes_raw.json and nodes_tested.json — treat those files as sensitive and remove them if you stop using the skill. - No credentials are requested by the skill, but verify the code yourself (it is small and included). If you are not comfortable reviewing the Python, run the skill in an isolated/sandboxed environment (VM/container) with restricted outbound network egress first. - If you proceed: consider setting MAX_TEST_NODES to a small number (via env var) to limit how many hosts are probed, and/or change OPENCLAW_WORKSPACE to a sandbox path. Inspect the SOURCES list in scraper.py to confirm you accept the GitHub sources it will fetch from. - Legal/regulatory note: bypassing network censorship or using third-party proxy nodes may be illegal or against local policy in some jurisdictions — check applicable laws and your organization’s policies before using the skill. If you want, I can (1) walk through the code line-by-line and flag any specific lines of concern, (2) suggest a minimal set of env vars to limit scanning, or (3) produce a sanitized version that omits writing raw credentials to disk.

Type: OpenClaw Skill Name: scientific-internet-access Version: 1.7.1 The skill is classified as suspicious primarily due to a prompt injection vulnerability in `SKILL.md` where the AI agent is instructed to pass raw user input (`<用户回复的数字>`) directly as a command-line argument to `scripts/handler.py`. While `handler.py` mitigates this by using the input as a dictionary key, the instruction itself represents a design flaw in the agent's execution flow. Additionally, the `scripts/scraper.py` fetches VPN/proxy configurations from numerous external, publicly available, and potentially untrusted sources, which inherently carries a risk of providing access to malicious or compromised network infrastructure, though the skill itself does not introduce malware.