ProxyBase

Purchase and manage SOCKS5 residential proxies via ProxyBase API with cryptocurrency payments. Supports order creation, payment polling, proxy delivery, band...

This skill appears to do what it says: it contacts api.proxybase.xyz to register an agent (auto-register on first run), create crypto payment invoices, poll confirmations, and write SOCKS5 credentials to {baseDir}/state/.proxy-env or inject them into your OpenClaw gateway. Before installing: 1) Verify you trust https://proxybase.xyz and are OK with the skill creating an agent/account and storing the returned API key under the skill's state directory. 2) Be cautious about using the 'inject-gateway' feature — it will modify your OpenClaw gateway/systemd configuration and restart the gateway, causing all agent network traffic to flow through the purchased proxy (this may expose sensitive requests to the proxy operator). 3) Review the state directory (credentials.env, .proxy-env, orders.json) and backup/remove keys you don't want persisted. 4) If you require stronger control, run commands manually (use the scripts without enabling automated injection/cron) and confirm each payment/address before sending crypto. The test harness references a helper (lib/common.sh) that isn't present in the manifest; this looks like a minor test packaging issue, not malicious behavior.

Type: OpenClaw Skill Name: proxybase-openclaw-skill Version: 1.0.12 The ProxyBase skill facilitates the purchase and management of SOCKS5 proxies via a third-party API (api.proxybase.xyz). It includes a high-risk capability in 'proxybase.sh' ('cmd_inject_gateway') that programmatically modifies the user's systemd service configuration ('openclaw-gateway.service') to inject proxy environment variables. While the script implements extensive security safeguards—such as strict alphanumeric allowlists for API responses, service file verification, and automatic backups—the modification of system configuration files is a sensitive operation. The behavior appears aligned with the stated purpose, and no evidence of malicious intent or data exfiltration was found.