AgentSkillsCN

ai-models

从Anthropic、OpenAI和Google的API中获取最新的AI模型ID，以便在代码中准确引用模型名称。

SKILL.md
--- frontmatter
name: ai-models
description: Fetch latest AI model IDs from Anthropic, OpenAI, and Google APIs to use accurate model names in code
metadata:
  tags: ai, models, api, anthropic, openai, google, claude, gpt, gemini
invocation:
  user_invocable: true

When to use

Use this skill automatically whenever you write code that calls AI APIs and need to specify a model ID.

Triggers:

  • •Writing API calls to Anthropic, OpenAI, or Google AI
  • •Setting model parameters in any AI SDK (Vercel AI SDK, LangChain, etc.)
  • •User asks about latest available models
  • •Any code involving model: or model= for AI providers

How to use

Run the fetch script — it scrapes public sources, no API keys needed:

bash
bash ~/.claude/skills/ai-models/fetch-models.sh

Options:

  • •--all (default): All providers
  • •--anthropic: Anthropic only
  • •--openai: OpenAI only
  • •--google: Google Gemini only

Sources:

  • •Anthropic: official docs page (docs.anthropic.com)
  • •OpenAI: SDK source of truth on GitHub (auto-generated from OpenAPI spec)
  • •Google: official Gemini docs (ai.google.dev)

Rules

  1. •Always run the script before using any model ID in code — never guess or use memorized IDs
  2. •Use the exact model ID from the script output
  3. •Pick the right tier for the use case:
    • •Flagship (Opus / GPT-4o / Gemini Pro): Complex reasoning, high quality
    • •Balanced (Sonnet / GPT-4o-mini / Gemini Flash): Best speed/quality ratio, default choice
    • •Fast (Haiku / GPT-3.5 / Gemini Flash Lite): Speed-first, simple tasks, cost-effective
  4. •Prefer aliases (e.g. claude-sonnet-4-5) over dated versions unless pinning is needed