WeryAI Chat

Chat, ask, compare, and inspect WeryAI chat models through the official OpenAI-compatible chat completions API. Use when you need general assistant chat, mul...

Security Scan

The skill’s code, runtime instructions, and requested credential (WERYAI_API_KEY) align with its stated purpose of calling the WeryAI chat completions API; nothing indicates hidden or unrelated access.

Purpose & Capability

Name/description, required binary (node), and required env (WERYAI_API_KEY) match the included scripts (models.js, write.js) that call the WeryAI chat and models endpoints. The package implements chat completions and model lookup as advertised; there are no unrelated credentials or binaries requested.

Instruction Scope

SKILL.md instructs the agent to run the included node scripts (models.js, write.js) and to prefer dry-run when validating payloads. The scripts only read environment variables and call the WeryAI endpoints; they do not attempt to read arbitrary user files or unrelated system state. A pre-scan detected a 'system-prompt-override' pattern, but in context this appears to be the normal inclusion of system messages and presets (systemPrompt strings) that are passed to the remote model — not instructions to override the agent host's system prompt or to exfiltrate data.

Install Mechanism

No install spec or remote download is used. This is an instruction/package bundle containing JS scripts that run with node; no external arbitrary code is fetched at install time.

Credentials

The only required credential is WERYAI_API_KEY (declared as primaryEnv), which is appropriate for a client that calls the WeryAI API. The code also reads optional environment variables (WERYAI_BASE_URL, WERYAI_REQUEST_TIMEOUT_MS, WERYAI_TEXT_MODEL) that are not listed in requires.env — this is typical for configuration, but you should verify WERYAI_BASE_URL points to the official API (default is https://api.weryai.com) before setting it.

Persistence & Privilege

The skill does not request always:true, does not modify other skills or system-wide configs, and has normal agent-invocation defaults. It does perform network calls to the WeryAI API (network_required: true), which is expected.

Scan Findings in Context

[system-prompt-override] expected: The pattern likely matches the many 'systemPrompt' strings and the skill-level instructions in SKILL.md. Those are used as model system messages passed to the remote chat API and are expected for a chat skill; they do not appear to be an attempt to overwrite the host agent's system prompt or to exfiltrate secrets.

Assessment

This package is coherent with its stated purpose: it needs only a WERYAI_API_KEY and node to run. Before installing, verify you trust the WeryAI API recipient: (1) ensure WERYAI_BASE_URL (if set) points to the official API (default: https://api.weryai.com), (2) prefer running with --dry-run to inspect request payloads without spending credits, and (3) do not supply the API key to untrusted skill sources. The SKILL.md and presets include system-message templates that will shape model responses—this is expected behavior for a chat client, not a hidden instruction to your local agent. If you need stronger assurance, review the scripts/models.js and scripts/write.js invocations locally or run them in an isolated environment first.

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