Oneworks Model Services

Use this skill when a user wants to add, update, or troubleshoot OneWorks modelServices.

Writeback Rules

• Put account-level API keys and secrets in global config by default.
• Use project or user config only when the user explicitly asks for an override.
• Do not copy values from merged config back into a source file unless you know the target source.
• Never show API keys, secrets, tokens, or passwords in the final answer.
• If the selected provider is known, prefer built-in defaults for provider, homepageUrl, apiBaseUrl, models, icons, and management capabilities. Only write override fields when the user asks for an override.

Config Shape

Use modelServices. with these common fields:

• provider: built-in provider id.
• title: display name.
• description: short display description.
• apiKey: service API key or secret.
• apiBaseUrl: only needed for custom services or explicit overrides.
• models: only needed for custom services or explicit overrides.
• billing: optional metadata for billing kind, key kind, quota unit, quota windows, and allowed use.
• codingPlan: optional metadata for plan links, protocol base URLs, regions, default models, and restrictions.
• homepageUrl: only needed to override the provider portal.
• management.enabled: enables model/balance/status/secret helper actions when available.
• management.apiKey: optional separate management API key; otherwise reuse apiKey when supported.

Coding Plan and Token Plan Rules

Here "Coding Plan" means a provider billing/quota product, not OneWorks or agent Plan Mode.

• Do not mix plan keys and ordinary pay-as-you-go API keys. A Coding Plan or Token Plan normally needs its dedicated key and dedicated base URL.
• Prefer built-in plan provider ids instead of overriding ordinary API providers:
• kimi-code: Kimi Code, OpenAI https://api.kimi.com/coding/v1, Anthropic https://api.kimi.com/coding/, model kimi-for-coding, Kimi Code API key from the Kimi Code console. Models and quota can be queried through the dedicated /coding/v1 API; do not use Moonshot/Kimi Platform PAYG balance endpoints for it.
• minimax-token-plan: MiniMax Token Plan, OpenAI https://api.minimax.io/v1, Anthropic https://api.minimax.io/anthropic, China Anthropic https://api.minimaxi.com/anthropic, model MiniMax-M3, Subscription Key.
• qwen-coding-plan: Alibaba Cloud Model Studio Coding Plan, OpenAI https://coding.dashscope.aliyuncs.com/v1, Anthropic https://coding.dashscope.aliyuncs.com/apps/anthropic, international OpenAI https://coding-intl.dashscope.aliyuncs.com/v1, international Anthropic https://coding-intl.dashscope.aliyuncs.com/apps/anthropic, plan key commonly sk-sp-*.
• zhipu-coding-plan: Zhipu GLM Coding Plan, OpenAI https://open.bigmodel.cn/api/coding/paas/v4, Anthropic https://open.bigmodel.cn/api/anthropic, model GLM-5.2.
• tencent-tokenhub-coding-plan: Tencent Cloud TokenHub Coding Plan, OpenAI https://api.lkeap.cloud.tencent.com/coding/v3, Anthropic https://api.lkeap.cloud.tencent.com/coding/anthropic.
• volcengine-ark-coding-plan: Volcengine Ark Coding Plan, OpenAI https://ark.cn-beijing.volces.com/api/coding/v3, Anthropic https://ark.cn-beijing.volces.com/api/coding.
• baidu-qianfan-coding-plan: Baidu Qianfan Coding Plan, OpenAI https://qianfan.baidubce.com/v2/coding, Anthropic https://qianfan.baidubce.com/anthropic/coding.
• OpenAI Codex and Anthropic Claude Code subscriptions are product-login entitlements, not model service API keys. ChatGPT/Codex and claude login can use subscription quotas, while OpenAI/Anthropic API keys still bill as ordinary API usage.
• Many plan quotas are request-window based, such as 5-hour, weekly, or monthly request limits. Do not describe them as ordinary token quotas unless the provider calls the product a Token Plan.
• Many providers restrict plans to interactive coding tools and disallow backend batch jobs, automation scripts, or generic API load. Preserve billing.allowedUse: coding_tools_only and show the restriction to the user.
• Plan model lists are catalog defaults and may not be available from /v1/models. Leave models empty to use the built-in catalog, or write models only when the user explicitly wants an allowlist override.
• DeepSeek is currently a pay-as-you-go API provider with coding-agent integrations, not a dedicated Coding Plan provider in this registry.

Built-In Providers

Official providers:

• openai: OpenAI, https://api.openai.com/v1; portal https://platform.openai.com; API keys https://platform.openai.com/api-keys; docs https://platform.openai.com/docs; status https://status.openai.com; model listing via API; balance unsupported.
• anthropic: Anthropic, https://api.anthropic.com/v1; portal https://console.anthropic.com; API keys https://console.anthropic.com/settings/keys; billing https://console.anthropic.com/settings/billing; docs https://docs.anthropic.com; status https://status.claude.com; model listing manual.
• Claude Code should use Anthropic directly with ANTHROPIC_API_KEY; OneWorks normalizes https://api.anthropic.com/v1 to https://api.anthropic.com for the native CLI.
• moonshot-cn: Moonshot/Kimi China, https://api.moonshot.cn/v1; model and balance APIs are supported, so prefer refreshing and saving the provider model list, then use the first saved model as the default candidate. Built-in fallback models are kimi-k2.7-code, kimi-k2.6, kimi-k2.5, kimi-k2-0905-preview, kimi-k2. Portal https://platform.kimi.com; API keys https://platform.kimi.com/console/api-keys; billing https://platform.kimi.com/console/account; docs https://platform.kimi.com/docs; status https://status.moonshot.cn.
• Claude Code should use Kimi's official Anthropic API adapter at https://api.moonshot.cn/anthropic instead of routing through Claude Code Router's OpenAI conversion path.
• moonshot-intl: Moonshot/Kimi International, https://api.moonshot.ai/v1; model and balance APIs are supported, so prefer refreshing and saving the provider model list, then use the first saved model as the default candidate. Built-in fallback models are kimi-k2.7-code, kimi-k2.6, kimi-k2.5, kimi-k2-0905-preview, kimi-k2. Portal https://platform.kimi.ai; API keys https://platform.kimi.ai/console/api-keys; billing https://platform.kimi.ai/console/account; docs https://platform.kimi.ai/docs; status https://status.moonshot.cn.
• Claude Code should use Kimi's official Anthropic API adapter at https://api.moonshot.ai/anthropic instead of routing through Claude Code Router's OpenAI conversion path.
• deepseek: DeepSeek, https://api.deepseek.com; default models deepseek-v4-flash, deepseek-v4-pro, with legacy aliases deepseek-chat, deepseek-reasoner kept for compatibility until their documented 2026-07-24 deprecation; portal https://platform.deepseek.com; API keys https://platform.deepseek.com/api_keys; billing https://platform.deepseek.com/usage; docs https://api-docs.deepseek.com; status https://status.deepseek.com; model and balance APIs are supported.
• Claude Code should use DeepSeek's official Anthropic API adapter at https://api.deepseek.com/anthropic instead of routing through Claude Code Router's OpenAI conversion path.
• minimax: MiniMax, https://api.minimax.io/v1; default model MiniMax-M3; portal https://platform.minimaxi.com; API keys https://platform.minimaxi.com/user-center/basic-information/interface-key; billing https://platform.minimaxi.com/user-center/basic-information/interface-balance; docs https://platform.minimaxi.com/document; status https://status.minimax.io; model API supported, balance manual.
• Claude Code should use MiniMax's official Anthropic API adapter at https://api.minimax.io/anthropic for international accounts or https://api.minimaxi.com/anthropic for China accounts. OneWorks preserves the configured host and switches the path to /anthropic.
• minimax-token-plan: MiniMax Token Plan; use Subscription Key, not a pay-as-you-go API key. Leave models empty for MiniMax-M3; Claude Code should use the plan Anthropic endpoint from the registry.
• qwen: Alibaba Qwen / Bailian, https://dashscope.aliyuncs.com/compatible-mode/v1; default Claude Code compatible models qwen3.7-max, qwen3.7-plus, qwen3.6-flash, qwen3.5-plus, qwen3-coder-next, qwen3-coder-plus; portal https://bailian.console.aliyun.com; API keys https://bailian.console.aliyun.com/?tab=api#/api-key; billing https://usercenter2.aliyun.com/finance; docs https://help.aliyun.com/zh/model-studio; model list is static.
• Claude Code should use Bailian's official Anthropic-compatible endpoints, for example pay-as-you-go https://dashscope.aliyuncs.com/apps/anthropic, Coding Plan https://coding.dashscope.aliyuncs.com/apps/anthropic, Token Plan https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic, or the regional workspace host ending in /apps/anthropic.
• qwen-coding-plan: Alibaba Cloud Model Studio Coding Plan; use the Coding Plan key with the Coding Plan endpoint. Do not use the ordinary DashScope key/base URL for this service.
• zhipu: Zhipu GLM, https://open.bigmodel.cn/api/paas/v4; default Claude Code compatible models glm-5.2[1m], glm-5.2, glm-4.7, glm-4.5-air; portal https://open.bigmodel.cn; API keys https://open.bigmodel.cn/usercenter/proj-mgmt/apikeys; billing https://open.bigmodel.cn/usercenter/resourcepool; docs https://docs.bigmodel.cn; model list is static.
• Claude Code should use Zhipu's official Anthropic-compatible endpoint at https://open.bigmodel.cn/api/anthropic. For glm-5.2[1m], keep CLAUDE_CODE_AUTO_COMPACT_WINDOW=1000000; use glm-4.5-air for the Haiku slot when available.
• zhipu-coding-plan: Zhipu GLM Coding Plan; use the coding endpoint and coding-plan key. Ordinary BigModel endpoints do not count against the plan.
• kimi-code, tencent-tokenhub-coding-plan, volcengine-ark-coding-plan, and baidu-qianfan-coding-plan are dedicated coding subscription providers with static model catalogs and dedicated OpenAI/Anthropic-compatible endpoints.

Cloud providers:

• azure-openai: Azure OpenAI through Azure AI Foundry; portal https://ai.azure.com; billing in Azure portal; docs https://learn.microsoft.com/azure/ai-services/openai; status https://status.azure.com.
• google-gemini: Google Gemini OpenAI-compatible endpoint https://generativelanguage.googleapis.com/v1beta/openai; portal https://aistudio.google.com; API keys https://aistudio.google.com/apikey; billing https://console.cloud.google.com/billing; docs https://ai.google.dev/gemini-api/docs.
• aws-bedrock: AWS Bedrock; portal https://console.aws.amazon.com/bedrock; billing https://console.aws.amazon.com/billing; docs https://docs.aws.amazon.com/bedrock; status https://health.aws.amazon.com/health/status.

Relay and gateway providers:

• openrouter: https://openrouter.ai/api/v1; portal https://openrouter.ai; API keys https://openrouter.ai/settings/keys; billing https://openrouter.ai/settings/credits; docs https://openrouter.ai/docs.
• Claude Code should use OpenRouter's Anthropic Skin directly at https://openrouter.ai/api, with ANTHROPIC_AUTH_TOKEN and empty ANTHROPIC_API_KEY.
• vercel-ai-gateway: https://ai-gateway.vercel.sh/v1; portal https://vercel.com/ai-gateway; billing https://vercel.com/dashboard/usage; docs https://vercel.com/docs/ai-gateway.
• Claude Code should use Vercel AI Gateway's Anthropic-compatible endpoint at https://ai-gateway.vercel.sh, with ANTHROPIC_AUTH_TOKEN and empty ANTHROPIC_API_KEY.
• requesty: https://router.requesty.ai/v1; portal https://requesty.ai; API keys https://app.requesty.ai/api-keys; billing https://app.requesty.ai/billing; docs https://docs.requesty.ai.
• Claude Code should use Requesty's Anthropic-compatible gateway at https://router.requesty.ai, with model IDs such as provider/model-name or policy/name.
• portkey: https://api.portkey.ai/v1; portal https://portkey.ai; API keys https://app.portkey.ai/api-keys; billing https://app.portkey.ai/billing; docs https://portkey.ai/docs.
• Claude Code should use Portkey's unified endpoint at https://api.portkey.ai. If a Portkey provider slug is required, configure extra.claudeCode.portkeyProvider, or set full extra.claudeCode.anthropicCustomHeaders.
• litellm: LiteLLM gateway; portal https://www.litellm.ai; docs https://docs.litellm.ai.
• micu: Micu relay; portal and docs https://micu.hk.
• apiyi: APIYI relay; portal and docs https://apiyi.com.
• yunwu: Yunwu relay; portal and docs https://yunwu.ai.
• custom-openai-compatible: user-defined OpenAI-compatible endpoint; requires explicit apiBaseUrl, usually explicit models, and manual management links.

Workflow

1. Confirm provider, region/site when relevant, API key, whether defaults should be used, and whether management helpers should be enabled.
2. If the provider has a portal, guide the user to open the portal for login, billing/top-up, and secret/API-key creation.
3. Write the minimal config: use provider plus apiKey; omit apiBaseUrl, models, homepageUrl, and icon fields when built-in defaults are enough.
4. For custom OpenAI-compatible services, ask for apiBaseUrl, supported model ids, homepage, and any billing/API-key URLs.
5. After saving, verify with model listing or a short chat request when credentials are available.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: oneworks-ai
• Source: oneworks-ai/app
• License: MIT
• Homepage: https://oneworks.cloud/

Install and usage instructions live in the source repository linked above.