@operor/cli 0.1.0

package/src/commands/test.ts

@@ -0,0 +1,177 @@
+import { readConfig } from '../config.js';
+import { writeFileSync } from 'fs';
+import { formatTimestamp } from '../utils.js';
+
+export async function runTest(options: {
+  csv?: string;
+  tag?: string;
+  report?: string;
+  real?: boolean;
+  allowWrites?: boolean;
+  dryRun?: boolean;
+} = {}): Promise<void> {
+  const config = readConfig();
+
+  const { Operor } = await import('@operor/core');
+  const { MockProvider } = await import('@operor/provider-mock');
+  const { MockShopifySkill } = await import('@operor/testing');
+
+  // If --csv is provided, run CSV test suite mode
+  if (options.csv) {
+    const { CSVLoader, TestSuiteRunner, SkillTestHarness } = await import('@operor/testing');
+
+    const os = new Operor({ debug: false, batchWindowMs: 0 });
+    const provider = new MockProvider();
+    const shopify = new MockShopifySkill();
+
+    // Wrap skill with safety harness if --real or --dry-run
+    let skill: any = shopify;
+    if (options.real || options.dryRun) {
+      skill = new SkillTestHarness(shopify, {
+        allowWrites: options.allowWrites ?? false,
+        dryRun: options.dryRun ?? false,
+      });
+      await skill.authenticate();
+    }
+
+    await os.addProvider(provider);
+    await os.addSkill(skill);
+
+    os.createAgent({
+      name: 'Test Agent',
+      purpose: 'Handle customer support conversations',
+      personality: 'empathetic and solution-focused',
+      triggers: ['order_tracking', 'general'],
+      tools: [shopify.tools.get_order, shopify.tools.create_discount, shopify.tools.search_products],
+      rules: [{
+        name: 'Auto-compensation',
+        condition: async (_ctx: any, toolResults: any[]) => {
+          const order = toolResults.find((t) => t.name === 'get_order');
+          return order?.success && order.result?.isDelayed && order.result?.delayDays >= 2;
+        },
+        action: async () => {
+          const discount = await shopify.tools.create_discount.execute({ percent: 10, validDays: 30 });
+          return { type: 'discount_created', code: discount.code, percent: 10, validDays: 30 };
+        },
+      }],
+    });
+
+    await os.start();
+
+    let testCases = await CSVLoader.fromFile(options.csv);
+
+    // Filter by tag if specified
+    if (options.tag) {
+      testCases = testCases.filter((tc) =>
+        tc.tags?.some((t) => t.toLowerCase() === options.tag!.toLowerCase())
+      );
+      if (testCases.length === 0) {
+        console.error(`No test cases found with tag "${options.tag}"`);
+        process.exit(1);
+      }
+    }
+
+    console.log(`\n  Running ${testCases.length} test case(s) from ${options.csv}...\n`);
+
+    // Set up LLM if available for evaluation
+    let llm: any;
+    if (config.LLM_PROVIDER && config.LLM_API_KEY) {
+      const { AIProvider } = await import('@operor/llm');
+      llm = new AIProvider({
+        provider: config.LLM_PROVIDER as any,
+        apiKey: config.LLM_API_KEY,
+        model: config.LLM_MODEL,
+      });
+    }
+
+    const suiteRunner = new TestSuiteRunner({
+      agentOS: os,
+      llm,
+    });
+
+    const result = await suiteRunner.runSuite(testCases);
+    await os.stop();
+
+    // Display results
+    for (const r of result.results) {
+      const icon = r.evaluation.passed ? '✓' : '✗';
+      const status = r.evaluation.passed ? 'PASS' : 'FAIL';
+      console.log(`  ${icon} [${formatTimestamp()}] [${status}] ${r.testCase.id}: ${r.testCase.question}`);
+      if (!r.evaluation.passed) {
+        console.log(`    Score: ${r.evaluation.score.toFixed(2)} — ${r.evaluation.reasoning}`);
+      }
+    }
+
+    console.log(`\n  Results: ${result.passed}/${result.total} passed (avg score: ${result.averageScore.toFixed(2)})`);
+    console.log(`  Duration: ${(result.totalDuration / 1000).toFixed(1)}s`);
+    console.log(`  Status: ${result.failed === 0 ? '✓ ALL PASSED' : '✗ SOME FAILED'}\n`);
+
+    if (options.report) {
+      writeFileSync(options.report, JSON.stringify(result, null, 2), 'utf-8');
+      console.log(`  Report saved to ${options.report}\n`);
+    }
+
+    process.exit(result.failed === 0 ? 0 : 1);
+    return;
+  }
+
+  // Default: run built-in demo scenarios
+  console.log('\n  Running Operor test scenarios...\n');
+
+  const os = new Operor({ debug: true, batchWindowMs: 0 });
+  const provider = new MockProvider();
+  const shopify = new MockShopifySkill();
+
+  await os.addProvider(provider);
+  await os.addSkill(shopify);
+
+  os.createAgent({
+    name: 'Test Agent',
+    purpose: 'Validate agent processing pipeline',
+    personality: 'helpful',
+    triggers: ['order_tracking', 'general'],
+    tools: [shopify.tools.get_order, shopify.tools.create_discount],
+    rules: [{
+      name: 'Auto-compensation',
+      condition: async (_ctx: any, toolResults: any[]) => {
+        const order = toolResults.find(t => t.name === 'get_order');
+        return order?.success && order.result?.isDelayed && order.result?.delayDays >= 2;
+      },
+      action: async () => {
+        const discount = await shopify.tools.create_discount.execute({ percent: 10, validDays: 30 });
+        return { type: 'discount_created', code: discount.code, percent: 10, validDays: 30 };
+      },
+    }],
+  });
+
+  let processed = 0;
+  const total = 4;
+  const done = new Promise<void>(resolve => {
+    os.on('message:processed', () => { if (++processed >= total) resolve(); });
+  });
+
+  await os.start();
+
+  const scenarios = [
+    { phone: '+1', msg: 'Where is my order #12345?', label: 'Delayed order (auto-compensation)' },
+    { phone: '+2', msg: 'Check order #67890', label: 'On-time order' },
+    { phone: '+3', msg: 'Where is order #99999?', label: 'Order not found' },
+    { phone: '+4', msg: 'Hello!', label: 'General greeting' },
+  ];
+
+  for (const s of scenarios) {
+    console.log(`  [${formatTimestamp()}] Test: ${s.label}`);
+    provider.simulateIncomingMessage(s.phone, s.msg);
+    await new Promise(r => setTimeout(r, 100));
+  }
+
+  await done;
+  await os.stop();
+
+  const discounts = shopify.getDiscounts();
+  console.log(`\n  Results: ${processed}/${total} messages processed`);
+  console.log(`  Discounts created: ${discounts.length}`);
+  console.log(`  Status: ${processed === total ? '✓ PASS' : '✗ FAIL'}\n`);
+
+  process.exit(processed === total ? 0 : 1);
+}

package/src/commands/vibe.ts

@@ -0,0 +1,478 @@
+import * as readline from 'node:readline';
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import { readConfig } from '../config.js';
+
+interface VibeOptions {
+  action?: string;
+  agent?: string;
+}
+
+interface LLMMessage {
+  role: 'system' | 'user' | 'assistant';
+  content: string;
+}
+
+const WELCOME = `
+  Welcome to Operor Vibe -- your agent copilot!
+
+  I can help you with:
+
+    1. Create      Create a new agent from a description
+    2. Customize   Update an agent's instructions and behavior
+    3. Connect     Add skills, knowledge sources, and tools
+    4. Improve     Refine responses based on feedback
+
+  What would you like to do?
+`;
+
+const AGENTS_DIR = path.resolve(process.cwd(), 'agents');
+
+const META_PROMPT = `You are Operor Vibe, an AI copilot that helps users create and customize AI agents.
+
+You generate file-based agent definitions for the Operor framework. Each agent lives in an agents/<name>/ directory with these files:
+
+## INSTRUCTIONS.md Format
+
+YAML frontmatter followed by markdown body:
+
+\`\`\`
+---
+name: <agent-slug>
+purpose: <one-line description>
+triggers:
+  - <intent keywords that route to this agent>
+channels:              # optional — omit to handle all channels
+  - whatsapp
+  - telegram
+  - wati
+skills:              # optional — configured via mcp.json
+  - shopify
+  - stripe
+  - github
+knowledgeBase: false   # set true to enable KB retrieval
+guardrails:            # optional
+  maxResponseLength: 1500
+  blockedTopics: []
+  escalationTriggers: []
+  systemRules: []
+---
+
+## Role
+<What this agent does and who it serves>
+
+## Scope
+<Bullet list of what the agent CAN do>
+
+## Out of Scope
+<Bullet list of what the agent should NOT do>
+
+## Style
+<Communication style guidelines>
+
+## Escalation
+<When and how to hand off to humans>
+
+## Examples
+<2-3 example user/agent exchanges>
+\`\`\`
+
+## IDENTITY.md Format
+Short personality card:
+- Name, Emoji, Nature, Vibe
+
+## SOUL.md Format
+Deeper behavioral guide:
+- Core Principles, Boundaries, Tone, When Things Go Wrong
+
+## Available skills
+configured via mcp.json (e.g. shopify, stripe, github)
+
+## Available channels
+whatsapp, telegram, wati
+
+## Your behavior
+- Be conversational and friendly
+- When the user describes what they want, ask 2-3 clarifying questions naturally before generating
+- Questions should cover: target audience, tone/personality, key capabilities, skills needed
+- When you have enough info, generate all three files
+- Output each file clearly marked with these exact delimiters:
+  --- FILE: INSTRUCTIONS.md ---
+  (file content)
+  --- FILE: IDENTITY.md ---
+  (file content)
+  --- FILE: SOUL.md ---
+  (file content)
+- After generating, summarize what you created`;
+
+const CUSTOMIZE_PROMPT = `You are Operor Vibe, helping the user customize an existing agent.
+
+The user will provide feedback about their agent. You have access to the current INSTRUCTIONS.md content.
+
+- Ask clarifying questions if the feedback is vague
+- When ready to apply changes, output the updated file with:
+  --- FILE: INSTRUCTIONS.md ---
+  (full updated content)
+- Explain what you changed`;
+
+const CONNECT_PROMPT = `You are Operor Vibe, helping the user connect skills and knowledge sources to an agent.
+
+Available skills: configured via mcp.json (e.g. shopify, stripe, github)
+Available channels: whatsapp, telegram, wati
+
+The user will tell you what to add. You have access to the current INSTRUCTIONS.md content.
+
+- Update the YAML frontmatter to add the requested skills/channels
+- Output the updated file with:
+  --- FILE: INSTRUCTIONS.md ---
+  (full updated content)
+- Explain what you connected`;
+
+const IMPROVE_PROMPT = `You are Operor Vibe, helping the user improve an agent's responses.
+
+The user will provide examples of bad responses and what they'd prefer instead.
+You have access to the current INSTRUCTIONS.md content.
+
+- Analyze the gap between actual and desired behavior
+- Update the Examples section and/or Style section to address the issue
+- Output the updated file with:
+  --- FILE: INSTRUCTIONS.md ---
+  (full updated content)
+- Explain what you improved`;
+
+function createLLM(config: ReturnType<typeof readConfig>) {
+  // Lazy import to avoid loading AI SDK at CLI startup
+  return import('@operor/llm').then(({ AIProvider }) => {
+    const maxTokens = config.VIBE_MAX_TOKENS ? Number(config.VIBE_MAX_TOKENS) : 4000;
+    return new AIProvider({
+      provider: config.LLM_PROVIDER as 'openai' | 'anthropic' | 'google' | 'groq' | 'ollama',
+      apiKey: config.LLM_API_KEY,
+      model: config.LLM_MODEL,
+      maxTokens,
+    });
+  });
+}
+
+function listAgents(): string[] {
+  if (!fs.existsSync(AGENTS_DIR)) return [];
+  return fs.readdirSync(AGENTS_DIR, { withFileTypes: true })
+    .filter(e => e.isDirectory() && !e.name.startsWith('_'))
+    .map(e => e.name);
+}
+
+function readAgentFile(agentName: string, fileName: string): string | null {
+  const filePath = path.join(AGENTS_DIR, agentName, fileName);
+  if (!fs.existsSync(filePath)) return null;
+  return fs.readFileSync(filePath, 'utf-8');
+}
+
+function writeAgentFile(agentName: string, fileName: string, content: string): void {
+  const dir = path.join(AGENTS_DIR, agentName);
+  fs.mkdirSync(dir, { recursive: true });
+  fs.writeFileSync(path.join(dir, fileName), content, 'utf-8');
+}
+
+export function parseGeneratedFiles(text: string): Record<string, string> {
+  const files: Record<string, string> = {};
+  const pattern = /---\s*FILE:\s*(\S+)\s*---\n([\s\S]*?)(?=---\s*FILE:|$)/g;
+  let match: RegExpExecArray | null;
+  while ((match = pattern.exec(text)) !== null) {
+    const name = match[1].trim();
+    const content = match[2].trim();
+    if (name && content) files[name] = content;
+  }
+  return files;
+}
+
+export function extractAgentName(files: Record<string, string>): string {
+  const instructions = files['INSTRUCTIONS.md'] || '';
+  const nameMatch = instructions.match(/^name:\s*(.+)$/m);
+  if (nameMatch) return nameMatch[1].trim();
+  return 'new-agent';
+}
+
+function prompt(rl: readline.Interface, question: string): Promise<string> {
+  return new Promise((resolve) => {
+    rl.question(question, (answer) => resolve(answer.trim()));
+  });
+}
+
+async function conversationLoop(
+  rl: readline.Interface,
+  llm: { complete: (msgs: LLMMessage[], opts?: any) => Promise<{ text: string }> },
+  systemPrompt: string,
+  initialUserMessage?: string,
+  maxTokens = 4000,
+): Promise<{ history: LLMMessage[]; lastResponse: string }> {
+  const history: LLMMessage[] = [{ role: 'system', content: systemPrompt }];
+
+  if (initialUserMessage) {
+    history.push({ role: 'user', content: initialUserMessage });
+    process.stdout.write('\n  Thinking...');
+    const response = await llm.complete(history, { maxTokens });
+    process.stdout.write('\r' + ' '.repeat(20) + '\r');
+    history.push({ role: 'assistant', content: response.text });
+    console.log(`\n  Vibe: ${response.text}\n`);
+
+    // Check if files were generated
+    const files = parseGeneratedFiles(response.text);
+    if (Object.keys(files).length > 0) {
+      return { history, lastResponse: response.text };
+    }
+  }
+
+  // Continue conversation
+  while (true) {
+    const input = await prompt(rl, '  You: ');
+    if (!input || input === 'exit' || input === 'quit' || input === 'done') {
+      return { history, lastResponse: history[history.length - 1]?.content || '' };
+    }
+
+    history.push({ role: 'user', content: input });
+    process.stdout.write('  Thinking...');
+    const response = await llm.complete(history, { maxTokens });
+    process.stdout.write('\r' + ' '.repeat(20) + '\r');
+    history.push({ role: 'assistant', content: response.text });
+    console.log(`\n  Vibe: ${response.text}\n`);
+
+    // Check if files were generated in this turn
+    const files = parseGeneratedFiles(response.text);
+    if (Object.keys(files).length > 0) {
+      return { history, lastResponse: response.text };
+    }
+  }
+}
+
+async function handleCreate(
+  rl: readline.Interface,
+  llm: { complete: (msgs: LLMMessage[], opts?: any) => Promise<{ text: string }> },
+  maxTokens: number,
+): Promise<void> {
+  console.log('\n  Describe the agent you want to create:\n');
+  const description = await prompt(rl, '  You: ');
+  if (!description) return;
+
+  const { lastResponse } = await conversationLoop(rl, llm, META_PROMPT, description, maxTokens);
+  const files = parseGeneratedFiles(lastResponse);
+
+  if (Object.keys(files).length === 0) {
+    console.log('\n  No agent files were generated. Try again with more detail.\n');
+    return;
+  }
+
+  const agentName = extractAgentName(files);
+  console.log(`\n  Writing agent files to agents/${agentName}/\n`);
+
+  for (const [fileName, content] of Object.entries(files)) {
+    writeAgentFile(agentName, fileName, content);
+    console.log(`    + agents/${agentName}/${fileName}`);
+  }
+
+  console.log(`\n  Agent "${agentName}" created! Run "operor start" to use it.\n`);
+}
+
+async function handleCustomize(
+  rl: readline.Interface,
+  llm: { complete: (msgs: LLMMessage[], opts?: any) => Promise<{ text: string }> },
+  maxTokens: number,
+  targetAgent?: string,
+): Promise<void> {
+  const agents = listAgents();
+  if (agents.length === 0) {
+    console.log('\n  No agents found. Create one first with "operor vibe create".\n');
+    return;
+  }
+
+  let agentName = targetAgent;
+  if (!agentName) {
+    console.log('\n  Available agents:\n');
+    agents.forEach((a, i) => console.log(`    ${i + 1}. ${a}`));
+    const choice = await prompt(rl, '\n  Select an agent (name or number): ');
+    const idx = parseInt(choice, 10);
+    agentName = idx > 0 && idx <= agents.length ? agents[idx - 1] : choice;
+  }
+
+  const instructions = readAgentFile(agentName, 'INSTRUCTIONS.md');
+  if (!instructions) {
+    console.log(`\n  Agent "${agentName}" not found or has no INSTRUCTIONS.md.\n`);
+    return;
+  }
+
+  console.log(`\n  Customizing agent "${agentName}". What would you like to change?\n`);
+  const feedback = await prompt(rl, '  You: ');
+  if (!feedback) return;
+
+  const contextMessage = `Current INSTRUCTIONS.md for agent "${agentName}":\n\n${instructions}\n\nUser feedback: ${feedback}`;
+  const { lastResponse } = await conversationLoop(rl, llm, CUSTOMIZE_PROMPT, contextMessage, maxTokens);
+  const files = parseGeneratedFiles(lastResponse);
+
+  if (files['INSTRUCTIONS.md']) {
+    writeAgentFile(agentName, 'INSTRUCTIONS.md', files['INSTRUCTIONS.md']);
+    console.log(`\n  Updated agents/${agentName}/INSTRUCTIONS.md\n`);
+  } else {
+    console.log('\n  No changes generated.\n');
+  }
+}
+
+async function handleConnect(
+  rl: readline.Interface,
+  llm: { complete: (msgs: LLMMessage[], opts?: any) => Promise<{ text: string }> },
+  maxTokens: number,
+  targetAgent?: string,
+): Promise<void> {
+  const agents = listAgents();
+  if (agents.length === 0) {
+    console.log('\n  No agents found. Create one first with "operor vibe create".\n');
+    return;
+  }
+
+  let agentName = targetAgent;
+  if (!agentName) {
+    console.log('\n  Available agents:\n');
+    agents.forEach((a, i) => console.log(`    ${i + 1}. ${a}`));
+    const choice = await prompt(rl, '\n  Select an agent (name or number): ');
+    const idx = parseInt(choice, 10);
+    agentName = idx > 0 && idx <= agents.length ? agents[idx - 1] : choice;
+  }
+
+  const instructions = readAgentFile(agentName, 'INSTRUCTIONS.md');
+  if (!instructions) {
+    console.log(`\n  Agent "${agentName}" not found or has no INSTRUCTIONS.md.\n`);
+    return;
+  }
+
+  console.log(`\n  What would you like to connect to "${agentName}"?`);
+  console.log('  (skills: configured via mcp.json, e.g. shopify, stripe, github)');
+  console.log('  (channels: whatsapp, telegram, wati)\n');
+  const request = await prompt(rl, '  You: ');
+  if (!request) return;
+
+  const contextMessage = `Current INSTRUCTIONS.md for agent "${agentName}":\n\n${instructions}\n\nUser request: ${request}`;
+  const { lastResponse } = await conversationLoop(rl, llm, CONNECT_PROMPT, contextMessage, maxTokens);
+  const files = parseGeneratedFiles(lastResponse);
+
+  if (files['INSTRUCTIONS.md']) {
+    writeAgentFile(agentName, 'INSTRUCTIONS.md', files['INSTRUCTIONS.md']);
+    console.log(`\n  Updated agents/${agentName}/INSTRUCTIONS.md\n`);
+  } else {
+    console.log('\n  No changes generated.\n');
+  }
+}
+
+async function handleImprove(
+  rl: readline.Interface,
+  llm: { complete: (msgs: LLMMessage[], opts?: any) => Promise<{ text: string }> },
+  maxTokens: number,
+  targetAgent?: string,
+): Promise<void> {
+  const agents = listAgents();
+  if (agents.length === 0) {
+    console.log('\n  No agents found. Create one first with "operor vibe create".\n');
+    return;
+  }
+
+  let agentName = targetAgent;
+  if (!agentName) {
+    console.log('\n  Available agents:\n');
+    agents.forEach((a, i) => console.log(`    ${i + 1}. ${a}`));
+    const choice = await prompt(rl, '\n  Select an agent (name or number): ');
+    const idx = parseInt(choice, 10);
+    agentName = idx > 0 && idx <= agents.length ? agents[idx - 1] : choice;
+  }
+
+  const instructions = readAgentFile(agentName, 'INSTRUCTIONS.md');
+  if (!instructions) {
+    console.log(`\n  Agent "${agentName}" not found or has no INSTRUCTIONS.md.\n`);
+    return;
+  }
+
+  console.log(`\n  Improving agent "${agentName}".`);
+  console.log('  Share an example of a bad response and what you\'d prefer:\n');
+  const feedback = await prompt(rl, '  You: ');
+  if (!feedback) return;
+
+  const contextMessage = `Current INSTRUCTIONS.md for agent "${agentName}":\n\n${instructions}\n\nUser feedback on response quality: ${feedback}`;
+  const { lastResponse } = await conversationLoop(rl, llm, IMPROVE_PROMPT, contextMessage, maxTokens);
+  const files = parseGeneratedFiles(lastResponse);
+
+  if (files['INSTRUCTIONS.md']) {
+    writeAgentFile(agentName, 'INSTRUCTIONS.md', files['INSTRUCTIONS.md']);
+    console.log(`\n  Updated agents/${agentName}/INSTRUCTIONS.md\n`);
+  } else {
+    console.log('\n  No changes generated.\n');
+  }
+}
+
+function resolveAction(input: string): string | null {
+  const normalized = input.toLowerCase().trim();
+  if (['1', 'create'].includes(normalized)) return 'create';
+  if (['2', 'customize'].includes(normalized)) return 'customize';
+  if (['3', 'connect'].includes(normalized)) return 'connect';
+  if (['4', 'improve'].includes(normalized)) return 'improve';
+  return null;
+}
+
+export async function runVibe(opts: VibeOptions): Promise<void> {
+  const config = readConfig();
+
+  if (!config.LLM_PROVIDER || !config.LLM_API_KEY) {
+    console.error('\n  LLM not configured. Run "operor setup" first to configure your LLM provider.\n');
+    process.exit(1);
+  }
+
+  let llm: Awaited<ReturnType<typeof createLLM>>;
+  try {
+    llm = await createLLM(config);
+  } catch (err: any) {
+    console.error(`\n  Failed to initialize LLM: ${err.message}\n`);
+    process.exit(1);
+  }
+
+  const maxTokens = config.VIBE_MAX_TOKENS ? Number(config.VIBE_MAX_TOKENS) : 4000;
+
+  const rl = readline.createInterface({
+    input: process.stdin,
+    output: process.stdout,
+  });
+
+  // Graceful Ctrl+C
+  rl.on('close', () => {
+    console.log('\n  Goodbye!\n');
+    process.exit(0);
+  });
+
+  let action = opts.action ? resolveAction(opts.action) : null;
+
+  if (!action) {
+    console.log(WELCOME);
+    const choice = await prompt(rl, '  > ');
+    action = resolveAction(choice);
+    if (!action) {
+      console.log('\n  Invalid choice. Use: create, customize, connect, or improve.\n');
+      rl.close();
+      return;
+    }
+  }
+
+  try {
+    switch (action) {
+      case 'create':
+        await handleCreate(rl, llm, maxTokens);
+        break;
+      case 'customize':
+        await handleCustomize(rl, llm, maxTokens, opts.agent);
+        break;
+      case 'connect':
+        await handleConnect(rl, llm, maxTokens, opts.agent);
+        break;
+      case 'improve':
+        await handleImprove(rl, llm, maxTokens, opts.agent);
+        break;
+    }
+  } catch (err: any) {
+    if (err.code === 'ERR_USE_AFTER_CLOSE') return; // readline closed via Ctrl+C
+    console.error(`\n  Error: ${err.message}\n`);
+  }
+
+  rl.close();
+}