Turnpike — Testing skill for Claude Code
A minimal local gateway that proxies Claude/Anthropic and OpenAI-spec clients to any OpenAI- or Anthropic-compatible LLM provider, with model remapping and one-command launch integrations for Claude C.
How to install Turnpike
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open aslamplr/turnpike and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Turnpike does
A minimal local gateway that proxies Claude/Anthropic and OpenAI-spec clients to any OpenAI- or Anthropic-compatible LLM provider, with model remapping and one-command launch integrations for Claude Code and Claude Desktop.
Alternatives in Testing
- Agent Plugins Spec — Agent Plugins Specification v1.0.0 — A minimal standard for packaging agent extensions into distributable plug 1.2k ★
- Next Evals OSS — Evals for Next.js up to 15.5.6 to test AI model competency at Next.js 310 ★
- Robotics Agent Skills — Agent skills that make AI coding assistants write production-grade robotics software 170 ★
README
turnpike
A minimal local LLM gateway in Rust, modeled on Ollama's Claude integration (`internal/proxy/claude_desktop.go` + `middleware/anthropic.go` + `ollama launch`).
It lets Anthropic- and OpenAI-spec clients (Claude Code, Claude Desktop, the Anthropic/OpenAI SDKs) run against **any** upstream provider that speaks the OpenAI or Anthropic wire spec — initially OpenCode Zen — by proxying requests byte-for-byte while rewriting only the `model` field.
Full design documentation — architecture, configuration reference, the gateway, the spec bridge, search middleware, and the launchers — lives in [docs/](docs/).
What it does
- Gateway with reverse proxy + model remap —
turnpike servestarts a loopback HTTP server. Client-facing model ids (e.g.claude-sonnet-5) map ontoprovider/modelpairs; the gateway rewrites themodelJSON field, injects the upstream provider's credentials, and streams the response straight back. Like-for-like specs pass through untouched; when the specs don't match, the bridge translates (below). - One-command client launchers — like
ollama launch claude:turnpike launch claude-codefinds/installs theclaudebinary and spawns it withANTHROPIC_BASE_URLpointed at the gateway and every model tier (opus/sonnet/haiku/subagents) remapped to a route id.turnpike launch claude-desktopwrites a third-party inference gateway profile into Claude Desktop's config library (mirroring Ollama's app), with automatic backup and--restore. Claude Desktop only recognizes UUID-named profile files, so turnpike uses the stable00000000-0000-5000-9000-000000000128.jsonand registers it in_meta.json; olderturnpike.jsonfiles are migrated out automatically, and pre-turnpike snapshots live inturnpike-backups/next to configLibrary.
- Agentic search middleware — when a bridged Claude client declares a
web_searchserver tool and a search provider
Related Skills
Model Provider
Switch the default model provider for sf's spawned build/test sub-agents: anthropic (default), codex, or curso
Gemma 4 31B MTP MLX
Local MLX gateway for running Gemma 4 31B with the Gemma 4 MTP assistant drafter. This is a text-only local ga
Grimoire
Prompt engineering toolkit: version-controlled prompts, AI agent skills manager, sandboxed Claude sessions in
Glm Delegate
Delegate token-heavy execution to swarms of GLM agents in disposable git worktrees — an Anthropic model writes
Vf Check
Run a vibe-force check (format, lint, analyzer, jest, static, local, apex, smoke, verify, all), interpret the
AppReveal
Debug-only in-app MCP server for iOS/macOS/Android/Flutter/ReactNative. Gives LLM agents Playwright-like contr
Related Agents
LLM Orchestrator
Use this agent for LLM integration work — prompt engineering, multi-provider abstraction (OpenRouter, Gemini,
Provider Debugger
Diagnose live-provider compatibility failures for Kimi, GLM, and other OpenAI-compatible endpoints
AI Recon
Delegates to this agent when the user wants to map the AI attack surface of an authorized web application befo