### GET /tools/list - List All Registered Tools

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Returns all dynamically created tools with their purposes and callable methods.

```APIDOC
## GET /tools/list - List All Registered Tools

### Description
Returns all dynamically created tools with their purposes and callable methods.

### Method
GET

### Endpoint
/tools/list

### Response
#### Success Response (200)
- **tools** (array) - A list of registered tools.
  - **name** (string) - The name of the tool.
  - **class_name** (string) - The class name of the tool.
  - **code_version_id** (string) - The version ID of the tool's code.
  - **purpose** (string) - The purpose of the tool.
  - **methods** (array) - A list of methods available for the tool.
    - **name** (string) - The name of the method.
    - **description** (string) - A description of the method.
    - **params** (array of strings) - The parameters the method accepts.
  - **created_at** (integer) - Timestamp when the tool was created.

#### Response Example
```json
{
  "tools": [
    {
      "name": "WaterTracker",
      "class_name": "FacetTool",
      "code_version_id": "WaterTracker-1719835200000",
      "purpose": "Track daily water intake in milliliters",
      "methods": [
        {"name": "logIntake", "description": "Record/log new data", "params": ["amount"]},
        {"name": "getTodayTotal", "description": "Get summary/totals", "params": []},
        {"name": "getHistory", "description": "Get historical entries", "params": []}
      ],
      "created_at": 1719835200000
    }
  ]
}
```
```

--------------------------------

### Get Tool Source Code

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Retrieves the raw AI-generated source code for a specific tool by name.

```bash
curl -X GET "https://your-worker.workers.dev/tools/code?name=WaterTracker"
```

--------------------------------

### GET /tools/code - Get Tool Source Code

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Retrieves the AI-generated source code for a specific tool.

```APIDOC
## GET /tools/code - Get Tool Source Code

### Description
Retrieves the AI-generated source code for a specific tool.

### Method
GET

### Endpoint
/tools/code

### Parameters
#### Query Parameters
- **name** (string) - Required - The name of the tool whose source code is to be retrieved.

### Request Example
```bash
curl -X GET "https://your-worker.workers.dev/tools/code?name=WaterTracker"
```

### Response
#### Success Response (200)
- The response body contains the source code of the requested tool as a string.

#### Response Example
```javascript
// Example response for WaterTracker tool code
export class WaterTracker {
  constructor(state, env) {
    this.state = state;
    this.env = env;
  }

  async logIntake(amount) {
    const currentData = await this.state.storage.get('intakeData') || [];
    const newData = [...currentData, { amount, timestamp: Date.now() }];
    await this.state.storage.put('intakeData', newData);
    return { ok: true };
  }

  async getTodayTotal() {
    const intakeData = await this.state.storage.get('intakeData') || [];
    const today = new Date().setHours(0, 0, 0, 0);
    const total = intakeData
      .filter(entry => entry.timestamp >= today)
      .reduce((sum, entry) => sum + entry.amount, 0);
    return { total };
  }

  async getHistory() {
    return await this.state.storage.get('intakeData') || [];
  }
}
```
```

--------------------------------

### GET /api - API Information

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Returns API metadata including available routes and the AI model being used.

```APIDOC
## GET /api - API Information

### Description
Returns API metadata including available routes and the AI model being used.

### Method
GET

### Endpoint
/api

### Response
#### Success Response (200)
- **name** (string) - The name of the project.
- **routes** (array of strings) - A list of available API routes.
- **model** (string) - The AI model currently in use.

#### Response Example
```json
{
  "name": "Project Hephaestus",
  "routes": ["/chat", "/tools/list", "/tools/create", "/tools/call", "/tools/destroy", "/tools/code"],
  "model": "@cf/zai-org/glm-4.7-flash"
}
```
```

--------------------------------

### Wrangler Configuration for Hephaestus Deployment

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Configuration for deploying Hephaestus with required bindings for Durable Objects, Worker Loaders, and Workers AI. Includes static asset configuration and migration setup for SQLite.

```jsonc
{
  "$schema": "node_modules/wrangler/config-schema.json",
  "name": "agents-week-demo",
  "main": "src/index.ts",
  "compatibility_date": "2026-04-01",
  "compatibility_flags": ["nodejs_compat"],

  // Static assets for the chat UI
  "assets": {
    "directory": "./public"
  },

  // Durable Object binding for the supervisor
  "durable_objects": {
    "bindings": [
      {
        "class_name": "SupervisorDO",
        "name": "SUPERVISOR"
      }
    ]
  },

  // Worker Loader for dynamic tool deployment
  "worker_loaders": [
    {
      "binding": "LOADER"
    }
  ],

  // Workers AI binding
  "ai": {
    "binding": "AI"
  },

  // SQLite migrations for the supervisor
  "migrations": [
    {
      "new_sqlite_classes": ["MyDurableObject"],
      "tag": "v1"
    },
    {
      "renamed_classes": [{"from": "MyDurableObject", "to": "SupervisorDO"}],
      "tag": "v2"
    }
  ]
}
```

--------------------------------

### POST /tools/create - Create a New Tool

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Generates and deploys a new tool using AI code generation based on the provided name and purpose.

```APIDOC
## POST /tools/create - Create a New Tool

### Description
Generates and deploys a new tool using AI code generation. The AI writes a Durable Object class based on the provided name and purpose.

### Method
POST

### Endpoint
/tools/create

### Parameters
#### Request Body
- **name** (string) - Required - The desired name for the new tool.
- **purpose** (string) - Required - A description of the tool's functionality.

### Request Example
```json
{
  "name": "TodoList",
  "purpose": "Manage a simple todo list with add, complete, and list operations"
}
```

### Response
#### Success Response (200)
- **name** (string) - The name of the created tool.
- **purpose** (string) - The purpose of the tool.
- **codeVersionId** (string) - The version ID of the tool's code.
- **methods** (array) - A list of methods available for the tool.
  - **name** (string) - The name of the method.
  - **description** (string) - A description of the method.
  - **params** (array of strings) - The parameters the method accepts.
- **alreadyExisted** (boolean) - Indicates if a tool with the same name already existed.

#### Response Example
```json
{
  "name": "TodoList",
  "purpose": "Manage a simple todo list with add, complete, and list operations",
  "codeVersionId": "TodoList-1719835300000",
  "methods": [
    {"name": "addTodo", "description": "Record/log new data", "params": ["text"]},
    {"name": "completeTodo", "description": "Update data", "params": ["id"]},
    {"name": "listTodos", "description": "Get historical entries", "params": []}
  ],
  "alreadyExisted": false
}
```
```

--------------------------------

### Retrieve API Metadata

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Fetches general information about the project, including available endpoints and the AI model in use.

```bash
curl -X GET https://your-worker.workers.dev/api
```

--------------------------------

### Create a New Tool

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Triggers AI code generation to create and deploy a new Durable Object tool based on the provided name and purpose.

```bash
curl -X POST https://your-worker.workers.dev/tools/create \
  -H "Content-Type: application/json" \
  -d '{"name": "TodoList", "purpose": "Manage a simple todo list with add, complete, and list operations"}'
```

--------------------------------

### List Registered Tools

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Retrieves a list of all dynamically created tools, including their purposes and available methods.

```bash
curl -X GET https://your-worker.workers.dev/tools/list
```

--------------------------------

### Call Tool Method

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Invokes a specific method on a registered tool. Requires the tool name, method name, and arguments.

```bash
# Add a todo item
curl -X POST https://your-worker.workers.dev/tools/call \
  -H "Content-Type: application/json" \
  -d '{"name": "TodoList", "method": "addTodo", "args": ["Buy groceries"]}'

# List all todos
curl -X POST https://your-worker.workers.dev/tools/call \
  -H "Content-Type: application/json" \
  -d '{"name": "TodoList", "method": "listTodos", "args": []}'
```

--------------------------------

### Create Workers AI Supervisor Model

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Creates an AI SDK compatible model for streaming text using the `workers-ai-provider`. It initializes the provider with the `env.AI` binding and specifies the AI model to use.

```typescript
// Create AI SDK compatible model for streamText
import { createWorkersAI } from "workers-ai-provider";

export function createSupervisorModel(env: Env) {
  const workersai = createWorkersAI({ binding: env.AI });
  return workersai(AI_MODEL); // "@cf/zai-org/glm-4.7-flash"
}
```

--------------------------------

### POST /tools/call - Call a Tool Method

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Invokes a specific method on a registered tool with the provided arguments.

```APIDOC
## POST /tools/call - Call a Tool Method

### Description
Invokes a specific method on a registered tool with the provided arguments.

### Method
POST

### Endpoint
/tools/call

### Parameters
#### Request Body
- **name** (string) - Required - The name of the tool to call.
- **method** (string) - Required - The name of the method to invoke on the tool.
- **args** (array) - Required - An array of arguments to pass to the method.

### Request Example
```json
# Add a todo item
{
  "name": "TodoList",
  "method": "addTodo",
  "args": ["Buy groceries"]
}

# List all todos
{
  "name": "TodoList",
  "method": "listTodos",
  "args": []
}
```

### Response
#### Success Response (200)
- **tool** (string) - The name of the tool that was called.
- **method** (string) - The name of the method that was invoked.
- **result** (any) - The result returned by the tool method. The structure of this field depends on the specific tool and method called.

#### Response Example
```json
# Example for addTodo method
{
  "tool": "TodoList",
  "method": "addTodo",
  "result": {"ok": true, "id": 1, "text": "Buy groceries", "createdAt": 1719835400000}
}

# Example for listTodos method
{
  "tool": "TodoList",
  "method": "listTodos",
  "result": [
    {"id": 1, "text": "Buy groceries", "completed": 0, "created_at": 1719835400000}
  ]
}
```
```

--------------------------------

### SupervisorDO Class - Core Durable Object Methods

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Documentation for the core methods of the SupervisorDO class, which orchestrates tool management and interactions.

```APIDOC
## SupervisorDO Class

### Core Durable Object Methods

The SupervisorDO class is the central orchestrator that handles all requests and manages the tool lifecycle.

### createTool

#### Description
Creates a new tool using AI code generation. It normalizes the name, generates code via AI, stores it in SQLite, and creates a facet.

#### Parameters
- **name** (string) - Required - The name of the tool to create.
- **purpose** (string) - Required - The purpose of the tool.

#### Returns
- **name** (string) - The name of the created tool.
- **purpose** (string) - The purpose of the created tool.
- **codeVersionId** (string) - The version ID of the generated code.
- **methods** (array) - An array of method descriptors for the tool.
- **alreadyExisted** (boolean) - Indicates if a tool with the same name already existed.

### callTool

#### Description
Calls a method on any registered tool. It retrieves the tool, gets the facet stub, and invokes the specified method with the provided arguments.

#### Parameters
- **name** (string) - Required - The name of the tool.
- **method** (string) - Required - The name of the method to call on the tool.
- **args** (array) - Required - An array of arguments to pass to the method.

#### Returns
- (any) - The result of the method call on the tool.

### destroyTool

#### Description
Deletes a tool and cleans up associated resources, including its Durable Object facet and stored code.

#### Parameters
- **name** (string) - Required - The name of the tool to delete.

#### Returns
- void

### handleChatStream

#### Description
Handles streaming chat interactions with an AI model using the Vercel AI SDK. It sets up the model, system prompt, user prompt, and available tools, and returns a streaming response.

#### Parameters
- **message** (string) - Required - The user's chat message.

#### Returns
- Response - A streaming response object.

```

--------------------------------

### Implement SupervisorDO Orchestrator

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Central class for managing tool lifecycles, routing requests, and handling AI streaming interactions.

```typescript
import { DurableObject } from "cloudflare:workers";
import { Hono } from "hono";
import { streamText, tool } from "ai";
import { z } from "zod";

export class SupervisorDO extends DurableObject<Env> {
  private readonly router: Hono;

  constructor(ctx: DurableObjectState, env: Env) {
    super(ctx, env);
    // Initialize SQLite schema for tool registry
    this.ctx.blockConcurrencyWhile(async () => this.initializeSchema());
    this.router = createSupervisorRouter(this);
  }

  // All HTTP requests are handled by the Hono router
  async fetch(request: Request): Promise<Response> {
    return this.router.fetch(request, this.env);
  }

  // Create a new tool using AI code generation
  async createTool(name: string, purpose: string): Promise<{
    name: string;
    purpose: string;
    codeVersionId: string;
    methods: MethodDescriptor[];
    alreadyExisted: boolean;
  }> {
    // Normalizes name, generates code via AI, stores in SQLite, creates facet
  }

  // Call a method on any registered tool
  async callTool(name: string, method: string, args: unknown[]): Promise<unknown> {
    const tool = this.getTool(name);
    const facet = this.getFacetStub(tool.name, tool.code_version_id, tool.class_name);
    const rpc = facet as Record<string, (...a: unknown[]) => Promise<unknown>>;
    return rpc[method](...args);
  }

  // Delete a tool and clean up resources
  async destroyTool(name: string): Promise<void> {
    this.ctx.facets.delete(normalizedName);
    this.ctx.storage.sql.exec("DELETE FROM tool_registry WHERE name = ?", normalizedName);
  }

  // Handle streaming chat with AI using Vercel AI SDK
  handleChatStream(message: string): Response {
    const result = streamText({
      model: createSupervisorModel(this.env),
      system: buildAgentSystemPrompt(),
      prompt: message,
      tools: {
        list_tools: tool({ /* ... */ }),
        create_tool: tool({ /* ... */ }),
        call_tool: tool({ /* ... */ }),
        // Plus dynamically registered tool methods
      },
      stopWhen: stepCountIs(20),
    });
    return result.toUIMessageStreamResponse();
  }
}
```

--------------------------------

### Destroy a Tool via API

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Sends a POST request to remove a tool from the registry and clean up associated resources.

```bash
curl -X POST https://your-worker.workers.dev/tools/destroy \
  -H "Content-Type: application/json" \
  -d '{"name": "TodoList"}'
```

--------------------------------

### Send AI Chat Request

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Sends a message to the supervisor. The response is a Server-Sent Events (SSE) stream containing tool inputs, outputs, and text deltas.

```bash
curl -X POST https://your-worker.workers.dev/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Track my water intake. I just drank 500ml."}'
```

--------------------------------

### WaterTracker API Endpoints

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

This section details the available methods for interacting with the WaterTracker Durable Object.

```APIDOC
## POST /logIntake

### Description
Records a new water intake entry.

### Method
POST

### Endpoint
/logIntake

### Parameters
#### Request Body
- **amount** (number) - Required - The amount of water intake in milliliters.

### Request Example
```json
{
  "amount": 500
}
```

### Response
#### Success Response (200)
- **ok** (boolean) - Indicates if the operation was successful.
- **amount** (number) - The logged amount of water intake.
- **loggedAt** (number) - The timestamp when the intake was logged.

#### Response Example
```json
{
  "ok": true,
  "amount": 500,
  "loggedAt": 1719835200000
}
```

## GET /getTodayTotal

### Description
Retrieves the total water intake for the current day.

### Method
GET

### Endpoint
/getTodayTotal

### Response
#### Success Response (200)
- **total** (number) - The total amount of water intake for the current day in milliliters.

#### Response Example
```json
{
  "total": 1500
}
```

## GET /getHistory

### Description
Retrieves a list of historical water intake entries, limited to the 50 most recent entries.

### Method
GET

### Endpoint
/getHistory

### Response
#### Success Response (200)
- **id** (number) - The unique identifier for the intake entry.
- **amount** (number) - The amount of water intake in milliliters.
- **logged_at** (number) - The timestamp when the intake was logged.

#### Response Example
```json
[
  {
    "id": 1,
    "amount": 500,
    "logged_at": 1719835200000
  },
  {
    "id": 2,
    "amount": 750,
    "logged_at": 1719748800000
  }
]
```
```

--------------------------------

### POST /chat - AI Chat with Streaming Response

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Sends a message to the AI supervisor and receives a streaming response. The AI can use built-in tools and dynamically registered tools.

```APIDOC
## POST /chat - AI Chat with Streaming Response

### Description
Sends a message to the AI supervisor and receives a streaming response. The AI can use built-in tools (list_tools, create_tool, call_tool) and any dynamically registered tools to fulfill requests.

### Method
POST

### Endpoint
/chat

### Parameters
#### Request Body
- **message** (string) - Required - The user's message to the AI.

### Request Example
```json
{
  "message": "Track my water intake. I just drank 500ml."
}
```

### Response
#### Success Response (200)
- The response is a Server-Sent Events (SSE) stream containing various event types such as `tool-input-available`, `tool-output-available`, `text-delta`, and `[DONE]`.

#### Response Example
```
data: {"type":"tool-input-available","toolName":"create_tool","toolCallId":"abc123","input":{"name":"WaterTracker","purpose":"Track daily water intake in milliliters"}}
data: {"type":"tool-output-available","toolCallId":"abc123"}
data: {"type":"tool-input-available","toolName":"call_tool","toolCallId":"def456","input":{"name":"WaterTracker","method":"logIntake","args":[500]}}
data: {"type":"tool-output-available","toolCallId":"def456"}
data: {"type":"text-delta","delta":"I've created a water tracker and logged your 500ml intake!"}
data: [DONE]
```
```

--------------------------------

### POST /tools/destroy - Delete a Tool

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Removes a tool from the registry and cleans up its Durable Object facet and stored code.

```APIDOC
## POST /tools/destroy

### Description
Removes a tool from the registry and cleans up its Durable Object facet and stored code.

### Method
POST

### Endpoint
/tools/destroy

### Parameters
#### Request Body
- **name** (string) - Required - The name of the tool to delete.

### Request Example
```bash
curl -X POST https://your-worker.workers.dev/tools/destroy \
  -H "Content-Type: application/json" \
  -d '{"name": "TodoList"}'
```

### Response
#### Success Response (200)
- **ok** (boolean) - Indicates if the operation was successful.
- **deleted** (string) - The name of the deleted tool.

#### Response Example
```json
{
  "ok": true,
  "deleted": "TodoList"
}
```
```

--------------------------------

### Define WaterTracker Durable Object

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Implements a Durable Object for tracking water intake with SQLite persistence. Requires Cloudflare Workers environment.

```typescript
import { DurableObject } from "cloudflare:workers";
export class FacetTool extends DurableObject {
  constructor(ctx, env) {
    super(ctx, env);
    this.sql = ctx.storage.sql;
    this.sql.exec("CREATE TABLE IF NOT EXISTS intake (id INTEGER PRIMARY KEY AUTOINCREMENT, amount INTEGER NOT NULL, logged_at INTEGER NOT NULL)");
  }
  async logIntake(amount) {
    const now = Date.now();
    this.sql.exec("INSERT INTO intake (amount, logged_at) VALUES (?, ?)", amount, now);
    return { ok: true, amount, loggedAt: now };
  }
  async getTodayTotal() {
    const startOfDay = new Date().setHours(0,0,0,0);
    const rows = this.sql.exec("SELECT SUM(amount) as total FROM intake WHERE logged_at >= ?", startOfDay).toArray();
    return { total: rows[0]?.total || 0 };
  }
  async getHistory() {
    return this.sql.exec("SELECT id, amount, logged_at FROM intake ORDER BY logged_at DESC LIMIT 50").toArray();
  }
}
```

--------------------------------

### Extract Method Signatures from Generated Code

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Extracts method signatures from generated JavaScript code. It parses `async methodName(params)` patterns, skipping common methods like `constructor`, `fetch`, and `alarm`. The `MethodDescriptor` type is used to store extracted information.

```typescript
// Extract method signatures from generated code
export function extractMethods(code: string): MethodDescriptor[] {
  const methods: MethodDescriptor[] = [];
  const pattern = /async\s+(\w+)\s*\(([^)]*)\)/g;
  const skip = new Set(["constructor", "fetch", "alarm"]);

  let match;
  while ((match = pattern.exec(code)) !== null) {
    const methodName = match[1];
    if (!skip.has(methodName)) {
      const params = match[2].split(",").map((p) => p.trim()).filter(Boolean);
      methods.push({ name: methodName, description: inferMethodDescription(methodName), params });
    }
  }
  return methods;
}
```

--------------------------------

### Core TypeScript Type Definitions

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Defines essential TypeScript types used throughout the application, including descriptors for tool methods, records for tool registry and code versions, request bodies for chat and tool APIs, AI call options, and chat message formats.

```typescript
// Method descriptor for tool RPC methods
export type MethodDescriptor = {
  name: string;
  description: string;
  params: string[];
};

// Tool registry record stored in SQLite
export type ToolRecord = {
  name: string;
  class_name: string;
  code_version_id: string;
  purpose: string;
  methods: string; // JSON stringified MethodDescriptor[]
  created_at: number;
};

// Tool code version stored in SQLite
export type ToolCodeRecord = {
  code_version_id: string;
  module_code: string;
  created_at: number;
};

// Chat API request body
export type ChatBody = {
  message: string;
};

// Tool API request body
export type ToolBody = {
  name: string;
  purpose?: string;
  method?: string;
  args?: unknown[];
};

// AI call options
export type RunAiOptions = {
  maxCompletionTokens?: number;
  timeoutMs?: number;
};

// Chat message format
export type ChatMessage = {
  role: "system" | "user" | "assistant";
  content: string;
};
```

--------------------------------

### Call Workers AI with Timeout Handling

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Low-level function to call Workers AI with timeout handling and response extraction. It uses `withTimeout` to enforce a specified timeout and `extractTextFromAiResponse` to process the AI's reply. Ensure `env.AI` is bound.

```typescript
import { AI_MODEL } from "../constants";
import type { ChatMessage, RunAiOptions } from "../types";

const DEFAULT_TIMEOUT_MS = 20_000;

export async function runAi(
  env: Env,
  messages: ChatMessage[],
  options: RunAiOptions = {}
): Promise<string> {
  const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
  const payload = {
    messages,
    max_completion_tokens: options.maxCompletionTokens ?? 500,
    chat_template_kwargs: { enable_thinking: false },
  };

  const raw = await withTimeout(env.AI.run(AI_MODEL, payload), timeoutMs);
  return extractTextFromAiResponse(raw);
}
```

--------------------------------

### Generate JavaScript Durable Object Code

Source: https://context7.com/apeacock1991/agents-week-demo-2026/llms.txt

Generates JavaScript Durable Object code using Workers AI based on the tool name and purpose. Requires a `runAiFn` that handles AI calls and returns raw completion. Ensure the generated code includes `export class ${FACET_CLASS_NAME}` and `async `.

```typescript
import { FACET_CLASS_NAME } from "../constants";
import { buildCodePrompt } from "../prompts";
import type { ChatMessage, RunAiOptions } from "../types";

export async function generateToolCode(
  name: string,
  purpose: string,
  runAiFn: (messages: ChatMessage[], options?: RunAiOptions) => Promise<string>,
): Promise<string> {
  const codePrompt = buildCodePrompt(name, purpose);
  const raw = await runAiFn(
    [
      {
        role: "system",
        content: "You write compact, valid JavaScript Durable Object classes..."
      },
      { role: "user", content: codePrompt }
    ],
    { maxCompletionTokens: 900, timeoutMs: 25_000 }
  );

  let code = stripCodeFences(raw);
  code = fixCommonSqlMistakes(code);

  if (code.includes(`export class ${FACET_CLASS_NAME}`) && code.includes("async ")) {
    return code;
  }
  throw new Error(`AI returned invalid tool code for "${name}" (${purpose})`);
}
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.