Copilot Tunnel Proxy — AI skill for Claude Code


Project: API-Mirror

**A self-contained, protocol-agnostic API translation and routing layer that converts any local tool’s native API calls into the format expected by remote AI model endpoints, without exposing credentials, altering workflow, or requiring cloud-side changes.**

---

Overview

Modern development tools—from terminal-based assistants to in-editor code completions—now demand direct access to proprietary AI model APIs. This creates friction: each tool speaks a different dialect of HTTP, each provider enforces unique authentication schemes, and every endpoint imposes its own rate limits, context windows, and response formats. You end up writing custom middleware for every combination.

**API-Mirror** inverts that problem. Instead of adapting tools one-by-one, it sits as a local transparent proxy that rewrites requests and responses in flight. You configure it once—pointing it at your preferred remote API—and it exposes a universal local endpoint that speaks any protocol your tools require. The result: you replace a dozen brittle integrations with a single, declarative configuration file.

Think of it as a **lingua franca for API communication** between your development environment and external model services. No account sharing, no environment variable soup, no duplicated effort when you switch providers.

---

Why This Exists

The original `copilot-bridge` project demonstrated that GitHub Copilot could be coerced into serving as a generic AI backend. That was clever, but it bound you to Copilot’s specific authentication and rate limits. **API-Mirror** generalizes the concept: you choose the remote provider (OpenAI, Anthropic, or any compatible service), and API-Mirror handles the translation layer so your local tools never know the difference.

Key insight: **the interface your tools expect is not the interface your provider offers.** API-Mirror bridg