Detect backend origins of Claude Code proxies via response fingerprints.
npx skills add https://github.com/zxc123aa/cc-proxy-detector --skill cc-proxy-detector
Please help me install this Agent Skill. Skill: cc-proxy-detector Source: https://github.com/zxc123aa/cc-proxy-detector/tree/main Command: npx skills add https://github.com/zxc123aa/cc-proxy-detector --skill cc-proxy-detector
Detects the real backend origin of Claude Code proxy/relay services when users need to verify whether traffic is going to official Anthropic, AWS Bedrock (Kiro), or Google Vertex AI (Antigravity), and when proxies attempt to disguise their routing.
Run the detector against your proxy endpoint by setting the base URL and API key in environment variables and then executing the provided Python script to generate a backend verdict report.
High-intent search queries and answers about installing and using this skill.
Claude proxy verification distinguishes official Anthropic, AWS Bedrock, and Google Vertex AI backends by probing the /v1/messages endpoint for tool_id, msg_id, and thinking_signature fingerprints, scoring responses against a three-source matrix to produce structured per-model verdicts.
Yes, you can check for mixed routing by performing a multi-model scan that sends deterministic probes to each model, identifying when a single proxy routes traffic to different backends like AWS Bedrock or Google Vertex AI by comparing extracted response fingerprints and generating per-model verdicts with confidence scores.
Anti-spoofing detection works by analyzing negative evidence from missing Anthropic-only fields and distinguishing dynamic ratelimit behavior from forged headers, catching deep spoofing attempts that disguise routing by validating actual tool execution and thinking signatures against expected backend fingerprints.
Running proxy backend detection requires the requests dependency, a configured proxy base URL, and an API key set as environment variables, enabling the Python script to probe the /v1/messages endpoint and generate structured verdict reports.
Fingerprint analysis limitations include reliance on deterministic probes returning specific fields like thinking_signature and usage, meaning proxies stripping or forging these fields require negative evidence validation, and dynamic ratelimit checks are needed to catch deep spoofing beyond simple header inspection.
LeventsoftAutomate Portainer CE stack deployment and inspection via HTTP API.
salvo-rsForward client requests to configurable backend services via a reverse proxy.
ricardorocheDetect prompt injection attempts in AI and LLM applications.
luis2404123Evade major bot detection systems using residential proxies and TLS fingerprint spoofing.
