@austinchen705/chatbot-intg-skill

An AI skill installer for Claude Code, Codex, and Gemini that guides vendors through complete AI Chatbot API integration — covering auth, chat lifecycle, escalation handling, error resilience, analysis, MCP Server development, and debugging.

Install

Run directly with npx (no install required):

npx @austinchen705/chatbot-intg-skill

Or install globally:

npm install -g @austinchen705/chatbot-intg-skill
chatbot-intg-skill

Usage

Interactive (default)

npx @austinchen705/chatbot-intg-skill

The installer will prompt you to select the target AI coding tool:

  AI Chatbot Integration Skill Installer
  =======================================

  Install for which tools? (comma-separated)
  1) claude  2) codex  3) gemini  4) all
  >

Non-interactive (CI / scripted)

Use the --tool flag to skip the prompt:

# Single tool
npx @austinchen705/chatbot-intg-skill --tool claude

# Multiple tools
npx @austinchen705/chatbot-intg-skill --tool claude,codex

# All tools
npx @austinchen705/chatbot-intg-skill --tool all

What Gets Installed

| Tool | Skill files | Command / Instruction file | |------|-------------|---------------------------| | Claude Code | .claude/skills/chatbot-intg/ | .claude/commands/chatbot-intg.md | | Codex | .codex/skills/chatbot-intg/ | .codex/AGENTS.md (manual step) | | Gemini | .gemini/skills/chatbot-intg/ | .gemini/GEMINI.md (manual step) |

Installed files

SKILL.md                          # Phased integration workflow
resources/
  chatbot_integration_spec.md    # API spec reference (v0.7)
  integration-playbook.md         # Language-specific code templates
  mcp-tool-spec.md                # MCP tool definition rules + multi-language skeletons

Invoking the Skill

Claude Code — the command is auto-detected after install:

/chatbot-intg

Codex — add the following line to .codex/AGENTS.md:

Read .codex/skills/chatbot-intg/SKILL.md for vendor integration workflow

Gemini — add the following line to .gemini/GEMINI.md:

Read .gemini/skills/chatbot-intg/SKILL.md for vendor integration workflow

Integration Workflow

The skill guides vendors through 6 gated phases, each requiring user confirmation before proceeding:

Phase 0: Infrastructure       Language + framework + logging setup
    ↓ (confirm)
Phase 1: Auth & Connectivity  JWT token lifecycle + auto-refresh
    ↓ (confirm)
Phase 2: Core Chat Flow       Session dispatch → send loop → kickout + escalation
    ↓ (confirm)
Phase 3: Error Handling       Unified error framework + retry + circuit breaker
    ↓ (confirm)
Phase 4: Advanced (optional)  Chat Analysis + MCP Server development
    ↓ (confirm)
Phase 5: Acceptance           Integration test checklist + debug handbook

Supported Languages

• Python (FastAPI / Django · httpx / requests · tenacity)
• Java (Spring Boot · OkHttp / RestTemplate · Resilience4j)
• C# (ASP.NET Core · HttpClient · Polly)
• Node.js (Express / NestJS · axios · cockatiel)

Covered APIs

| Method | Route | Description | |--------|-------|-------------| | POST | /api/v1/auth | Obtain JWT Token | | POST | /api/v1/chatbot/dispatch | Create chat session | | GET | /api/v1/chatbot/usage | Query quota | | POST | /api/v1/chatbot/kickout | Close a session | | POST | /api/v1/chatbot/killall | Close all sessions | | POST | /api/v1/chat/send | Send message + receive AI response | | GET | /api/v1/chat/history/{sessionId} | Retrieve chat history | | POST | /api/v1/analysis/create | Submit conversation for analysis | | GET | /api/v1/analysis/{sessionId} | Poll analysis result |

MCP Server Development

Phase 4 of the skill includes guided MCP Server development. The skill helps vendors build custom tools that the AI can call during conversations — based on their own business logic.

How it works

The skill follows a three-step process:

1. Discovery — identify business capabilities to expose as tools

Tool Inventory
──────────────────────────────────────────────────────────
get_order_status   | customer asks about order      | order_id
search_products    | customer looks for a product   | query, category
create_ticket      | customer wants to raise issue  | subject, description

2. Tool Design — apply spec rules before writing code

| Rule | Requirement | |------|-------------| | Naming | snake_case, verb-first (e.g. get_order_status) | | Description | 50–200 chars, includes "Use when..." | | Parameters | Each parameter must have a Description annotation | | Return value | Structured object: { success, data, error } — never plain string | | Stateless | No server-side session dependency per tool call |

3. Implementation — generate skeleton in the vendor's language

Supported languages and their MCP SDK:

| Language | SDK / Library | |----------|--------------| | Python | mcp (FastMCP) | | Java | Spring AI MCP | | C# | ModelContextProtocol.Server | | Node.js | @modelcontextprotocol/sdk |

Transport & Security

All MCP Servers use Streamable HTTP transport with JWT HS256 authentication. The Chatbot system signs and injects the token on every tool request — vendors only need to validate it.

POST /mcp
Authorization: Bearer <chatbot-signed-jwt>

JWT parameters (SecretKey, Issuer, Audience, ExpiryMinutes) are configured per tenant via the Admin API — not hardcoded in the MCP Server.

Example tool skeleton (C#)

[McpServerToolType]
public class OrderTools(IOrderRepository orderRepo)
{
    [McpServerTool(Name = "get_order_status")]
    [Description("Retrieves the current status of a customer order by order ID. "
               + "Use when the customer asks about their order progress or delivery.")]
    public async Task<ToolResult> GetOrderStatusAsync(
        [Description("The order ID provided by the customer (e.g. ORD-12345)")] string orderId)
    {
        var order = await orderRepo.FindByIdAsync(orderId);
        if (order is null)
            return ToolResult.Fail("ORDER_NOT_FOUND", $"Order {orderId} not found");
        return ToolResult.Ok(order);
    }
}

Full skeletons for Python, Java, C#, and Node.js are in resources/mcp-tool-spec.md after installation.

Requirements

• Node.js >= 14
• Run from the root of your project so skill files land in the right directories

License

MIT