weryai-chat（维雅聊天）

WeryAI Chat

Use the official WeryAI chat-completions API for general assistant chat and model lookup. This skill is intentionally broad but not specialized: it is for general conversation and prompt-response tasks, not blog writing, social copy, or email drafting.

Example Prompts

• Ask a WeryAI chat model to explain retrieval augmented generation in plain English.
• Send this messages array to WeryAI chat completions and return the assistant response.
• List the currently available WeryAI chat models and their pricing.
• Use GPT_5_4 for this one chat call instead of the default model.

Quick Summary

• Main jobs: general assistant chat, multi-turn chat, chat model lookup, prompt-response
• Default model: GEMINI_3_1_PRO
• Main optional controls: model, messages, maxTokens, temperature, topP
• Main trust signals: dry-run support, model lookup, OpenAI-compatible messages, explicit non-specialized scope

Prerequisites

• WERYAI_API_KEY must be set before calling the API.
• Node.js >=18 is required.
• Real runs use the WeryAI chat completion API and may consume credits.

When to use this skill

Use this skill when the user wants:

• a normal assistant-style answer
• a direct chat-completions call
• a multi-turn conversation via messages
• model lookup or model selection before a chat run

Do not use this skill when the user clearly wants:

• blog writing
• email drafting
• ad copy
• translation or summarization as the main task

Those belong to the existing specialized text/* skills.

OpenAI-compatible message shape

This skill accepts standard chat-completions messages:

[
  { "role": "system", "content": "You are a helpful assistant." },
  { "role": "user", "content": "What is artificial intelligence?" }
]

If you provide messages, they are passed through directly. If you provide only prompt, the runtime builds a simple messages array automatically.

Commands

# List available chat models
node {baseDir}/scripts/models.js

# Simple prompt-response chat
node {baseDir}/scripts/write.js --json '{
  "prompt":"Explain retrieval augmented generation in plain English",
  "temperature":0.7
}'

# Explicit messages array
node {baseDir}/scripts/write.js --json '{
  "model":"GPT_5_4",
  "messages":[
    {"role":"system","content":"You are concise and technical."},
    {"role":"user","content":"Compare RAG and long-context prompting."}
  ]
}'

# Dry-run preview
node {baseDir}/scripts/write.js --json '{
  "prompt":"What is the difference between latency and throughput?"
}' --dry-run

Workflow

1. If the user wants model choice or pricing context first, run models.js.
2. Use write.js for direct prompt-response or explicit messages chat.
3. Prefer --dry-run when validating payload shape without spending credits.
4. Return the assistant response directly when the call succeeds.

Definition of Done

• models.js returns the available chat models and pricing metadata.
• write.js returns at least one assistant completion choice and non-empty text, or a clear API failure.

Re-run Behavior

• Re-running models.js is read-only and safe.
• Re-running write.js --dry-run is safe and does not call the API.
• Re-running write.js creates a fresh chat completion request and may consume additional credits.

References

• Official chat API summary: references/chat-api.md
• Chat completion API: Chat Completion
• Chat models API: Get Chat Models List