add-models_skill

Overview

This skill adds one or more AI models to the language model constants file by researching official documentation and inserting accurate ModelOption entries. It validates provider mapping, gathers specs (API name, context window, max output tokens, default temperature, and cost tier), updates the constants array, and produces a short summary of changes. The result is ready-to-compile TypeScript additions that follow existing naming and ordering conventions.

How this skill works

You provide a comma-separated list of model names. The skill parses the list, maps each model to a provider, fetches authoritative specs from official docs, and constructs ModelOption objects matching the project's conventions. It inserts the new entries into the proper provider array, optionally creates named constant exports, runs type checks, and returns a concise report of what changed and any ambiguities encountered.

When to use it

• You need to add new or recently released LLMs to the app's available model list.
• Official documentation exists but the project lacks accurate API names or token limits.
• A QA or dev wants standardized ModelOption entries for newly supported providers.
• You want consistent ordering and comments linking to official model docs.
• Preparing the app to offer a new model to end users or to test integration.

Best practices

• Provide exact model names in the comma-separated input (e.g., "gemini 3.1 pro, claude sonnet 4").
• Confirm ambiguous models — if a model could belong to multiple providers, decide primary vs. secondary placement.
• Prefer official docs as sources and include a link comment in the constants file for traceability.
• Place newest or most capable models at the top of the provider array and group by family.
• Run the TypeScript type check (npm run ts) and fix any type issues before finalizing changes.

Example use cases

• Add Google Gemini 2.5 Pro and ensure correct API model name, context window, and cost tier.
• Introduce a new Anthropic Sonnet variant with correct Anthropic and optional Bedrock entries.
• Register OpenRouter-hosted third-party models (e.g., DeepSeek) with provider-prefixed model names.
• Add an Azure-hosted GPT variant and comment out maxOutputTokens per provider convention.
• Prepare MODEL_OPTIONS and TURBO_MODELS updates when a turbo variant is released.