Running Claude Code Via Litellm Copilot

# Doc-Verified Notes

This skill is anchored to Xi Xu's article from December 2, 2025:

- [Run Claude Code Cheaply: A Complete Guide to Using LiteLLM with the GitHub Copilot Chat API](https://blog.xi-xu.me/en/2025/12/02/Run-Claude-Code-Cheaply-With-LiteLLM-And-GitHub-Copilot.html)

It is also tightened against current LiteLLM documentation retrieved during package creation:

- LiteLLM GitHub Copilot provider docs, version `v1.81.9-stable`

## Safe Carryovers From The Article

These parts of the article match the current provider model closely enough to keep as the skill's default narrative:

- Claude Code can be pointed at a local LiteLLM endpoint through `ANTHROPIC_BASE_URL`
- Claude Code still expects a non-empty local Anthropic auth token value even though LiteLLM forwards to Copilot instead
- LiteLLM acts as the middle layer between Claude Code and the GitHub Copilot API
- `drop_params: true` remains an important compatibility setting for Anthropic-shaped requests
- the first working request can trigger GitHub device authorization
- verification should be done by watching LiteLLM logs while sending a small Claude Code request

## Doc-Verified Deltas

### Provider naming

The current LiteLLM provider route uses this pattern:

```text
github_copilot/<model>
```

That is the stable rule the skill should teach. The article's example `github_copilot/claude-opus-4.5` still fits this pattern.

### Exact model-name matching

The `ANTHROPIC_MODEL` value in Claude Code must match LiteLLM `model_name` exactly. Treat this as a first-line troubleshooting rule, not a footnote.

### `drop_params: true`

Keep this in the default config examples. It strips Anthropic-specific request fields before forwarding to Copilot and helps avoid avoidable 4xx errors.

### Optional token storage overrides

LiteLLM currently documents these optional environment variables for GitHub Copilot token storage:

- `GITHUB_COPILOT_TOKEN_DIR`
- `GITHUB_COPILOT_ACCESS_TOKEN_FILE`

Only mention them when the user needs custom storage paths or auth troubleshooting. They are not part of the minimum happy path.

### Header overrides are advanced fallback, not default

The article explicitly adds editor-style headers. Current LiteLLM docs also document header override support, but the provider itself is now clearly modeled as `github_copilot/...`.

For this skill, treat explicit `extra_headers` as an advanced fallback when:

- the user already has evidence that header shape matters in their environment
- the basic provider route is reaching Copilot but still needs client emulation tweaks

Do not present custom headers as universally required for the first setup attempt.

## Guidance For Future Revisions

- Preserve the article's user-facing flow because it is clear and practical.
- Prefer current LiteLLM provider mechanics over historical assumptions when they conflict.
- If a user asks for a different Copilot-backed model, keep the `github_copilot/<model>` pattern and avoid claiming that a specific Copilot model is permanently available unless it was re-verified from current docs.


---
name: running-claude-code-via-litellm-copilot
description: Use when routing Claude Code through a local LiteLLM proxy to GitHub Copilot, reducing direct Anthropic spend, configuring ANTHROPIC_BASE_URL or ANTHROPIC_MODEL overrides, or troubleshooting Copilot proxy setup failures such as model-not-found, no localhost traffic, or GitHub 401/403 auth errors.
---

## Overview

Use this skill for the specific workaround where Claude Code keeps its Anthropic-shaped client behavior, but the actual backend traffic is sent to a local LiteLLM proxy and then forwarded to GitHub Copilot.

Treat this as an advanced workaround, not an officially guaranteed GitHub workflow. Help the user succeed technically, but do not promise GitHub support, policy approval, or long-term compatibility.

This skill is guidance-first but execution-aware:

- If the user wants explanation only, provide the smallest corr