v7.3.0

wrapper-scionos 7.3.0

This release fixes long-running proxy streams, makes Codex model selection explicit, and validates macOS Keychain storage and recent OpenCode clients.

Changes

• Claude Code, OpenCode and the shared Desktop/Cowork proxy use separate first-response and resettable inactivity timeouts. Both default to 120 seconds; total streaming duration is unlimited by default and explicitly configurable. Client disconnects still cancel upstream requests.
• Codex no longer silently substitutes another model when the selected family's preferred model is missing. Interactive launches ask for a choice; non-interactive launches require --model. The native catalogue stays within the selected family.
• macOS writes receive secrets through stdin, verify the stored value and use bounded system operations. Existing Keychain entries retain their access controls and remain readable, updateable and removable.
• OpenCode 2 uses its native private --standalone server. Versions 1.18.30 and 2.0.27 pass streaming and a native read-tool cycle. Process-tree cleanup and redirected-output tests now wait for actual startup and preserve the original error.
• English, French and German documentation reflects the tested behavior, secret handling and compatibility limits.

Validation

• GitHub CI: all 37 jobs passed on Windows, macOS and Linux, including real macOS Keychain.
• General matrix: Node 22.13.0, 22, 23.5.0, 24, 25 and 26; real Claude Code 2.1.296 and Codex 0.144.1. OpenCode 1.18.30/2.0.27 passed on Node 22.13.0, 24 and 26 across all three systems.
• Final local unit checks: 229 passed, 1 conditional skip, 0 failures. Coverage: 92.03% lines, 91.69% functions, 83.45% branches; existing thresholds retained.
• Installed/npx package checks passed. npm audit reported zero vulnerabilities. All client inference tests used fake credentials and loopback providers.

Desktop/Cowork profile and proxy tests do not certify the real applications. Production RouterLab inference, every native MCP/session feature and pricing claims remain outside this validation. See the detailed validation record.

The attached archive is the npm package for this version. npm registry publication is handled separately by the package owner. Install the downloaded archive with npm install --global ./wrapper-scionos-7.3.0.tgz, or use npm install --global wrapper-scionos@7.3.0 once it is available on npm.

v7.2.0

wrapper-scionos 7.2.0

This release updates the RouterLab model catalogues and fixes Windows client launches, local proxy handling, and credential storage. RouterLab LLM keeps its existing models and defaults.

Models and menus

• Claude Native: claude-fable-5-1, claude-opus-5-5, claude-sonnet-5-5, and claude-haiku-5-5. Claude Code 2.1.293 or newer is required for this strategy.
• AWS Claude: aws-claude-haiku-4-5, aws-claude-sonnet-5-5, and aws-claude-opus-5-5. The menu displays the 40% discount; this strategy requires Claude Code 2.1.284 or newer.
• OpenAI GPT: gpt-6-astra, gpt-6.1-sol, and gpt-6-luna. RouterLab Codex and OpenCode use gpt-6.1-sol as their service default.
• Trial: deepseek-v4.1-flash-trial, glm-5.3-flash-trial, qwen3.8-flash-trial, gpt-6-luna-trial, and MiniMax-M3-trial. Claude Code's fixed Trial subagent and Codex's preferred Trial model use deepseek-v4.1-flash-trial.
• RouterLab Codex first selects OpenAI GPT-6, Open Source, or Trial; its native /model selector then exposes the selected family's verified models.
• The Desktop menu entry is now Claude Desktop / Cowork.

Existing scripts that explicitly select a removed RouterLab model must use an updated identifier. Discovery still checks availability for the authenticated account before launch.

Fixes and authentication

• Modern npm-generated Windows shims launch their Node entry point directly so quoted Codex runtime settings remain intact.
• Local proxies enforce authenticated loopback access, supported API paths, request timeouts, and model-discovery size limits.
• Desktop profile writes are serialized and recover more reliably on failure; secure-storage failures are reported explicitly.
• Commands supporting an explicit token now accept --token-stdin. Inline --token emits a deprecation warning. Claude Code continues to use masked login, environment variables, or secure storage.
• All six repository Markdown files were reviewed; English, French, and German instructions reflect the current catalogues and client menus.

Validation

• npm run release:check passed locally on Windows.
• GitHub CI passed all six jobs: Windows, macOS, and Linux with Node 22.13.0 and Node 24, including real Codex checks at the supported minimum version 0.144.1.
• Unit and coverage checks: 206 passed, 1 platform-specific test skipped, 0 failed.
• Coverage: 92.34% lines, 91.72% functions, 83.42% branches.
• All four installed/npx entry-mode checks passed.
• Real-client checks passed with Claude Code 2.1.296 (13 scenarios), Codex CLI 0.157.0, and OpenCode 1.18.30, using local test services.
• npm audit: 0 vulnerabilities.
• npm archive inspection: 40 files; no credentials or development caches included.

The attached archive is the prepared npm package. Publication to the npm registry is a separate step.

v7.1.0

Added

• Added grouped RouterLab Codex model catalogues: OpenAI GPT (gpt-6-astra, gpt-6-sol, gpt-6-luna), Open Source (deepseek-v4.1-flash, glm-5.3, glm-5.3-flash, hy4-preview, kimi-k3, minimax-m3, qwen3.8-max), and Trial (deepseek-v4.1-trial, gemini-3.8-flash-trial, glm-5.3-flash-trial, MiniMax-M3-trial, qwen3.8-max-trial).
• Added the RouterLab Trial strategy to Claude Code and OpenCode. Claude Code maps Trial models to Fable, Haiku, Sonnet, and Opus and fixes the subagent to deepseek-v4.1-flash.

Changed

• Updated Claude Code RouterLab Native to claude-fable-5-1 and claude-opus-5-5; aligned its OpenAI GPT and Open Source strategies with the new RouterLab model catalogues.
• Set the RouterLab Codex no-prompt default to gpt-6-sol; OpenCode and RouterLab LLM keep gpt-5.6-sol.

v7.0.0

Full Changelog: v6.0.2...v7.0.0

v6.0.2

Correctif

• Retire deepseek-v4-flash-0731 des choix de sous-agents Claude Code pour le service routerlab.
• Le service llm et les stratégies principales DeepSeek restent inchangés.
• Suite de tests complète : 159 réussis, 1 ignoré.

v6.0.1

Fixed

• Fixed explicit Claude Code subagent model selection for the default routerlab service.
• Preserved the existing selected-subagent behavior for --service llm.
• Updated documentation and regression tests.

v6.0.0

Breaking Changes

• Removed glm-5.2 from the default RouterLab Claude Code, Claude Desktop, and Codex catalogues. Scripts selecting the former glm-5.2 strategy or model must select a replacement.
• Removed glm-5.2, deepseek-v4-pro-0813, deepseek-v4-flash-0731, and grok-4.6 from the RouterLab LLM main-model catalogues for Claude Code, Claude Desktop, and Codex. deepseek-v4-flash-0731 remains available as a Claude Code subagent model.

Added

• Added the RouterLab LLM glm-5.3 strategy as the second Claude Code choice, routing Opus and Sonnet to glm-5.3 and Haiku/subagents to glm-5.3-flash. Both models are also available in Claude Desktop and Codex, and glm-5.3-flash is available in the Claude Code subagent picker.
• Added qwen3.8-max, glm-5.3-flash, grok-4.6, and gemini-3.7-flash to the default RouterLab Claude Code and Codex catalogues. The Claude Desktop RouterLab catalogue mirrors the same additions.
• Added explicit Codex runtime catalog metadata for glm-5.3, glm-5.3-flash, qwen3.8-max, grok-4.6, and gemini-3.7-flash.

Fixed

• Marked kimi-k3 as supporting its documented 1,048,576-token context.
• Preserved Claude Code's 1M mode and normalized compatible decorated requests such as kimi-k3[1m] to the RouterLab identifier kimi-k3 before forwarding. Batch requests are normalized consistently.

v5.3.0

What's Changed

🚀 Added

• RouterLab LLM (--service llm) : Added kimi-k3 support across all coding assistant integration points:
  • Claude Code CLI : available via --service llm --strategy kimi-k3 (with lightweight subagent selection).
  • Claude Desktop : exposed in local mapping proxy under the route claude-kim3.
  • Codex CLI : added kimi-k3 to the allowed model whitelist for discovery and --model kimi-k3 launches.

🧪 Tests & Quality

• Added unit and integration tests covering the kimi-k3 strategy environment, Desktop routing, and Codex model allowlists for --service llm.
• 100% test suite passing.

v5.2.0

Added

• Added grok-4.6 model strategy to RouterLab LLM (--service llm) and exposed it in the Claude Desktop routing catalog as claude-rok4.6 and in Codex CLI allowlist.
• Added combined deepseek (DeepSeek V4) strategy across both services (routerlab and llm), mapping Opus and Sonnet to deepseek-v4-pro-0813, and Haiku/subagents to deepseek-v4-flash-0731.
• Added kimi-k3 and deepseek-v4-pro-0813 to default routerlab service for Claude Code, Claude Desktop (claude-kim3, claude-deev4-pro-0813), and Codex CLI allowlist.
• Added interactive and CLI subagent model selection (--subagent-model ) to the default routerlab service (claude-haiku-4-5, aws-claude-haiku-4-5, deepseek-v4-flash-0731, gpt-5.6-luna), matching the behavior of --service llm.

Changed

• Updated default Claude Native Haiku model identifier from claude-haiku-4-5-20251001 to claude-haiku-4-5 across all strategy environments and Desktop routes.
• Removed deprecated grok-4.5 model.

v5.1.0

Mise à jour des catalogues Codex, Claude Code et Claude Desktop. Ajout de la sélection de sous-agent LLM. Passage à Chalk 6.0.0.