Claude Code for RouterLab

claude-scionos is a RouterLab launcher for the official Claude Code CLI. It keeps the normal Claude Code workflow, while adding guided onboarding, model strategy configuration, secure token storage, and a doctor command for client support.

🇫🇷 Lire en français

Highlights

• Guided launch for first-time users
• --strategy to preselect a routing strategy
• --subagent-model to keep the strategy default or force subagents to Haiku, GPT 5.4 mini, or claude-deepseek-v4-flash on the default RouterLab service
• --service llm for invitation-only RouterLab LLM access
• --no-prompt for automation and CI
• --list-strategies to inspect available routes
• doctor to diagnose local setup quickly
• auth login|status|change|logout|test for secure token management
• Direct Claude Code calls to RouterLab with native model environment variables

Requirements

• Node.js 22 or later
• A RouterLab token from routerlab.ch/keys
• Or an invitation token for --service llm
• On Windows, Git Bash must be installed for Claude Code

Installation

Run directly with npx:

npx claude-scionos

Or install globally:

npm install -g claude-scionos
claude-scionos

Quick Start

Guided mode:

npx claude-scionos

Useful commands:

npx claude-scionos --list-strategies
npx claude-scionos doctor
npx claude-scionos auth login
npx claude-scionos auth login --service llm
npx claude-scionos auth test
npx claude-scionos --strategy aws
npx claude-scionos --service llm --strategy claude
npx claude-scionos --service llm --strategy claude-gpt
npx claude-scionos --service llm --strategy claude-gpt-special
npx claude-scionos --service llm --strategy deepseek-v4
npx claude-scionos --service llm --strategy deepseek-v4-beta
npx claude-scionos --service llm --strategy minimax-m2.7
npx claude-scionos --strategy deepseek-v4 --subagent-model haiku
npx claude-scionos --strategy aws --no-prompt -p "Summarize this repo"

Services

• Default behavior uses https://api.routerlab.ch
• --service llm switches the launcher to https://llm-api.routerlab.ch
• llm is intended for invitation-only access
• Tokens stored with auth login --service llm are kept separate from the default RouterLab token
• llm currently exposes claude, claude-gpt, claude-gpt-special, deepseek-v4, deepseek-v4-beta, minimax-m2.7, and glm-5.1
• routerlab also exposes aws, claude-gpt, deepseek-v4, deepseek-v4-beta, claude-kimi-k2.6, and glm-5.1

Strategies

• default: use Claude Code normally without forcing model environment variables
• aws: sets Claude Code model variables to RouterLab AWS-backed Claude variants (Opus aws-claude-opus-4-8, Sonnet aws-claude-sonnet-4-6, Haiku aws-claude-haiku-4-5-20251001)
• claude: on --service llm, uses Claude Code natively without forcing model environment variables (same as default but via the LLM endpoint)
• claude-gpt: sets model variables to the GPT family Opus uses claude-gpt-5.5, Sonnet uses claude-gpt-5.4, Haiku and subagents use claude-gpt-5.4-mini
• claude-gpt-special: on --service llm, uses claude-gpt-5.5-sp for Opus and Sonnet, and claude-gpt-5.4-mini-sp for Haiku and subagents
• deepseek-v4: uses claude-deepseek-v4-pro for Opus/Sonnet and claude-deepseek-v4-flash for Haiku/subagents
• deepseek-v4-beta: same model mapping as deepseek-v4, available on both routerlab and llm
• minimax-m2.7: on --service llm, shown as minimax-m2.7 and uses claude-minimax-m2.7 for Opus, Sonnet, Haiku, and subagents
• claude-kimi-k2.6: uses claude-kimi-k2.6 for Opus, Sonnet, Haiku, and subagents
• glm-5.1: uses claude-glm-5.1 for Opus, Sonnet, Haiku, and subagents

Use --list-strategies to see the strategies available for the selected service and their live availability when a token is available.

Subagent model

By default, each strategy keeps its own subagent model. On the default RouterLab service, you can override only Claude Code subagents without changing the main Opus, Sonnet, and Haiku model variables:

• --subagent-model default: keep the strategy default
• --subagent-model haiku: force subagents to claude-haiku-4-5-20251001
• --subagent-model gpt-5.4-mini: force subagents to claude-gpt-5.4-mini
• --subagent-model claude-deepseek-v4-flash: force subagents to claude-deepseek-v4-flash

For example, --strategy deepseek-v4 uses claude-deepseek-v4-flash for subagents by default. Adding --subagent-model haiku keeps the DeepSeek main model routing but sets CLAUDE_CODE_SUBAGENT_MODEL=claude-haiku-4-5-20251001.

All --service llm strategies support the same subagent models as the default service.

Token Handling

Token resolution order:

1. ANTHROPIC_AUTH_TOKEN
2. Secure local storage from claude-scionos auth login
3. Service-specific secure local storage from claude-scionos auth login --service llm
4. Manual prompt in guided mode

Secure storage backends:

• Windows: DPAPI-encrypted local file bound to the current user
• macOS: Keychain
• Linux: Secret Service via secret-tool

On Windows, claude-scionos now lets PowerShell handle only DPAPI encryption and decryption, while Node.js writes and reads the secure token file directly. This fixes the case where the secure token file was created but left empty. If an older file is empty or corrupted, claude-scionos treats it as missing instead of trying to use it. Run claude-scionos auth login again to store a fresh token.

Manage the token with:

claude-scionos auth login
claude-scionos auth status
claude-scionos auth change
claude-scionos auth logout
claude-scionos auth test

What --strategy and --no-prompt mean

• --strategy <value> skips the interactive strategy menu and selects the route directly
• --service <value> switches between RouterLab targets. routerlab is the default and llm is invitation-only
• --subagent-model <default|haiku|gpt-5.4-mini|claude-deepseek-v4-flash> overrides the Claude Code subagent model on both the default RouterLab service and --service llm
• --no-prompt disables every interactive question

When --no-prompt is used, the launcher must already have a token from ANTHROPIC_AUTH_TOKEN or secure storage.

Doctor

claude-scionos doctor checks the local setup and prints a support-friendly summary:

• platform and Node.js
• Claude Code installation
• Git Bash on Windows
• secure storage backend
• stored or environment token presence
• RouterLab token validation when a token is available

Compatibility

The wrapper forwards regular Claude Code flags and arguments. RouterLab is called directly by Claude Code; strategies are applied through ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL, ANTHROPIC_DEFAULT_HAIKU_MODEL, and, where needed, CLAUDE_CODE_SUBAGENT_MODEL.

Troubleshooting

claude-scionos doctor should be the first command to run when a client reports an issue.

Common cases:

• Claude Code CLI not found: install @anthropic-ai/claude-code
• Git Bash is required on Windows: install Git for Windows
• ANTHROPIC_AUTH_TOKEN ... is required when using --no-prompt: set the environment variable or store the token first
• Secure token file was created but no encrypted content was written: update to 4.2.0 or later, then re-run claude-scionos auth login
• Stored token is missing on Windows even though you already logged in: re-run claude-scionos auth login because the local DPAPI token file may be empty or corrupted
• secret-tool not found: install a Secret Service client on Linux or rely on the environment variable

Development

npm install
npm test
npm run lint
node index.js

License

MIT. See LICENSE.