@butlerw/vellum 0.1.0

package/src/index.tsx

@@ -0,0 +1,614 @@
+#!/usr/bin/env node
+import {
+  AgentLoop,
+  APPROVAL_POLICIES,
+  type ApprovalPolicy,
+  BUILTIN_CODING_MODES,
+  CODING_MODES,
+  type CodingMode,
+  OnboardingWizard as CoreOnboardingWizard,
+  createAgentFactory,
+  LLM,
+  normalizeMode,
+  SANDBOX_POLICIES,
+  type SandboxPolicy,
+  UnifiedToolContainer,
+} from "@vellum/core";
+import { ProviderRegistry } from "@vellum/provider";
+import { createId } from "@vellum/shared";
+import { Command } from "commander";
+import { render } from "ink";
+import { registerDelegateCommand } from "./agents/commands/index.js";
+import { App } from "./app.js";
+import { handleAgentsGenerate } from "./commands/agents/generate.js";
+import { handleAgentsShow } from "./commands/agents/show.js";
+import { handleAgentsValidate } from "./commands/agents/validate.js";
+import {
+  renderCredentialsAdd,
+  renderCredentialsList,
+  renderCredentialsRemove,
+} from "./commands/credentials.js";
+import { executeInit } from "./commands/init.js";
+import { createLspCommand } from "./commands/lsp.js";
+import {
+  handleSkillCreate,
+  handleSkillList,
+  handleSkillShow,
+  handleSkillValidate,
+} from "./commands/skill.js";
+import { getEffectiveThinkingConfig } from "./commands/think.js";
+import type { CommandResult } from "./commands/types.js";
+import { getOrCreateOrchestrator } from "./orchestrator-singleton.js";
+import { executeShutdownCleanup, getShutdownCleanup, setShutdownCleanup } from "./shutdown.js";
+import { BufferedStdout, createCompatStdout, setActiveStdout } from "./tui/buffered-stdout.js";
+import { initI18n } from "./tui/i18n/index.js";
+
+import {
+  getAlternateBufferSetting,
+  getDefaultAlternateBufferEnabled,
+} from "./tui/i18n/settings-integration.js";
+import { isConptyTerminal } from "./tui/utils/detectTerminal.js";
+import { version } from "./version.js";
+
+// ============================================
+// Helper: Get message from CommandResult
+// ============================================
+
+/**
+ * Extract display message from CommandResult
+ */
+function getResultMessage(result: CommandResult): string {
+  switch (result.kind) {
+    case "success":
+      return result.message ?? "";
+    case "error":
+      return result.message;
+    case "interactive":
+      return result.prompt.message;
+    case "pending":
+      return result.operation.message;
+  }
+}
+// ============================================
+// T-VIRTUAL-SCROLL: Working Stdio Proxy for Ink
+// ============================================
+
+/**
+ * Creates working stdio proxies for Ink rendering.
+ *
+ * This fixes an issue where Ink's stdout/stderr writes can get intercepted
+ * or cause issues in VS Code terminal. By proxying through the original
+ * write functions bound to process.stdout/stderr, we ensure atomic writes.
+ *
+ * Pattern adapted from Gemini CLI's terminal handling.
+ */
+function createWorkingStdio(): {
+  stdout: typeof process.stdout;
+  stderr: typeof process.stderr;
+} {
+  const originalWrite = process.stdout.write.bind(process.stdout);
+  const originalErrWrite = process.stderr.write.bind(process.stderr);
+
+  const inkStdout = new Proxy(process.stdout, {
+    get(target, prop, receiver) {
+      if (prop === "write") return originalWrite;
+      const value = Reflect.get(target, prop, receiver);
+      return typeof value === "function" ? value.bind(target) : value;
+    },
+  });
+
+  const inkStderr = new Proxy(process.stderr, {
+    get(target, prop, receiver) {
+      if (prop === "write") return originalErrWrite;
+      const value = Reflect.get(target, prop, receiver);
+      return typeof value === "function" ? value.bind(target) : value;
+    },
+  });
+
+  return {
+    stdout: inkStdout as typeof process.stdout,
+    stderr: inkStderr as typeof process.stderr,
+  };
+}
+
+// ============================================
+// Graceful Shutdown Setup (T030)
+// ============================================
+
+/**
+ * Handle process signals for graceful shutdown.
+ */
+function setupGlobalShutdownHandlers(): void {
+  const signals: NodeJS.Signals[] = ["SIGINT", "SIGTERM", "SIGQUIT"];
+
+  for (const signal of signals) {
+    process.on(signal, () => {
+      console.log(`\n[CLI] Received ${signal}, shutting down gracefully...`);
+      executeShutdownCleanup();
+      // Give time for cleanup, then exit
+      setTimeout(() => process.exit(0), 100);
+    });
+  }
+}
+
+// Setup handlers early
+setupGlobalShutdownHandlers();
+
+// =============================================================================
+// T037-T040: Mode CLI Flag Interfaces
+// =============================================================================
+
+/**
+ * Chat command options including mode flags.
+ */
+export interface ChatOptions {
+  /** Model to use for AI responses */
+  model: string;
+  /** Provider to use (anthropic, openai, etc.) */
+  provider: string;
+  /** Language/locale for UI */
+  language?: string;
+  /** Coding mode (vibe, plan, spec) */
+  mode: CodingMode;
+  /** Approval policy override */
+  approval?: ApprovalPolicy;
+  /** Sandbox policy override */
+  sandbox?: SandboxPolicy;
+  /** Full-auto shortcut flag */
+  fullAuto?: boolean;
+  /** UI theme (dark, parchment, dracula, etc.) */
+  theme?: string;
+  /** Force banner display on startup */
+  banner?: boolean;
+}
+
+/**
+ * Parse mode flag with validation.
+ * Supports both new modes (vibe, plan, spec) and legacy modes (code, draft, debug, ask).
+ * @param value - User input value
+ * @returns Validated CodingMode
+ */
+function parseMode(value: string): CodingMode {
+  const result = normalizeMode(value);
+  return result.mode;
+}
+
+/**
+ * Parse approval policy flag with validation.
+ * @param value - User input value
+ * @returns Validated ApprovalPolicy
+ */
+function parseApproval(value: string): ApprovalPolicy {
+  const valid = APPROVAL_POLICIES as readonly string[];
+  if (!valid.includes(value)) {
+    throw new Error(`Invalid approval: ${value}. Valid options: ${valid.join(", ")}`);
+  }
+  return value as ApprovalPolicy;
+}
+
+/**
+ * Parse sandbox policy flag with validation.
+ * @param value - User input value
+ * @returns Validated SandboxPolicy
+ */
+function parseSandbox(value: string): SandboxPolicy {
+  const valid = SANDBOX_POLICIES as readonly string[];
+  if (!valid.includes(value)) {
+    throw new Error(`Invalid sandbox: ${value}. Valid options: ${valid.join(", ")}`);
+  }
+  return value as SandboxPolicy;
+}
+
+const program = new Command();
+
+program.name("vellum").description("Next-generation AI coding agent").version(version);
+
+// =============================================================================
+// T037-T040: Chat Command with Mode Flags
+// =============================================================================
+
+program
+  .command("chat", { isDefault: true })
+  .description("Start interactive chat session")
+  .option("-m, --model <model>", "Model to use", "claude-sonnet-4-20250514")
+  .option("-p, --provider <provider>", "Provider to use", "anthropic")
+  .option("-l, --language <locale>", "Language/locale to use (e.g., en, zh)")
+  // T037: --mode flag
+  .option("--mode <mode>", `Set coding mode (${CODING_MODES.join("|")})`, parseMode, "vibe")
+  // T038: --approval flag
+  .option(
+    "--approval <policy>",
+    `Set approval policy (${APPROVAL_POLICIES.join("|")})`,
+    parseApproval
+  )
+  // T039: --sandbox flag
+  .option("--sandbox <policy>", `Set sandbox policy (${SANDBOX_POLICIES.join("|")})`, parseSandbox)
+  // T040: --full-auto shortcut
+  .option("--full-auto", "Shortcut for --mode=vibe --approval=full-auto", false)
+  // Theme selection
+  .option("--theme <theme>", "UI theme (dark|parchment|dracula|etc.)", "parchment")
+  .option("--banner", "Show banner on startup", false)
+  .action(async (options: ChatOptions) => {
+    // T040: Apply --full-auto shortcut
+    let effectiveMode = options.mode;
+    let effectiveApproval = options.approval;
+
+    if (options.fullAuto) {
+      effectiveMode = "vibe";
+      effectiveApproval = "full-auto";
+    }
+
+    // Initialize i18n before rendering (T019)
+    initI18n({ cliLanguage: options.language });
+
+    // Load user's saved configuration from onboarding
+    // CLI flags override user config if explicitly provided
+    let effectiveProvider = options.provider;
+    let effectiveModel = options.model;
+
+    try {
+      const wizard = new CoreOnboardingWizard();
+      const loadResult = await wizard.loadState();
+      if (loadResult.ok) {
+        const userConfig = wizard.generateConfig();
+        if (userConfig.provider) {
+          effectiveProvider = userConfig.provider;
+        }
+        if (userConfig.model) {
+          effectiveModel = userConfig.model;
+        }
+      }
+    } catch {
+      // Use CLI defaults if config loading fails
+    }
+
+    // Create AgentLoop with real LLM provider
+    let agentLoop: AgentLoop | undefined;
+    let initError: Error | undefined;
+    try {
+      // Initialize credential manager for secure credential resolution
+      const { createCredentialManager } = await import("./commands/auth.js");
+      const credentialManager = await createCredentialManager();
+
+      // Initialize provider registry with credential manager
+      const providerRegistry = new ProviderRegistry({
+        credentialManager: credentialManager,
+      });
+      LLM.initialize(providerRegistry);
+
+      // Get mode config for the selected coding mode
+      const modeConfig = BUILTIN_CODING_MODES[effectiveMode];
+
+      // Initialize orchestrator singleton for task delegation
+      const orchestrator = getOrCreateOrchestrator();
+
+      // Create PromptBuilder with MD prompts (REQ-001: Use MD prompts)
+      const { promptBuilder, cleanup } = await createAgentFactory({
+        cwd: process.cwd(),
+        projectRoot: process.cwd(),
+        role: "base", // Load base.md for identity
+        mode: effectiveMode, // Load vibe.md/plan.md/spec.md
+      });
+
+      // Register cleanup for graceful shutdown
+      setShutdownCleanup(cleanup);
+
+      // Create unified tool container (T045: Single source of truth for tools)
+      const toolContainer = new UnifiedToolContainer({
+        cwd: process.cwd(),
+      });
+      toolContainer.registerBuiltins();
+
+      // Create AgentLoop with PromptBuilder and unified tool container
+      agentLoop = new AgentLoop({
+        sessionId: createId(),
+        mode: modeConfig,
+        providerType: effectiveProvider,
+        model: effectiveModel,
+        cwd: process.cwd(),
+        projectRoot: process.cwd(),
+        interactive: true,
+        orchestrator,
+        promptBuilder, // Use MD-loaded prompts
+        // Dynamic thinking config getter for runtime /think toggling
+        getThinkingConfig: getEffectiveThinkingConfig,
+        // T045: Wire unified tool container
+        tools: toolContainer.getProviderToolDefinitions(),
+        toolExecutor: toolContainer.getExecutor(),
+        // Enable AGENTS.md and Skills integration for project context
+        enableAgentsIntegration: true,
+        enableSkillsIntegration: true,
+      });
+    } catch (error) {
+      initError = error instanceof Error ? error : new Error(String(error));
+      console.error("[CLI] Failed to initialize agent:", initError.message);
+      // Continue without agentLoop - App will show error banner and fall back to echo mode
+    }
+
+    const isVSCodeTerminal =
+      process.env.TERM_PROGRAM === "vscode" ||
+      process.env.VSCODE_INJECTION === "1" ||
+      Boolean(process.env.VSCODE_GIT_IPC_HANDLE);
+    const isScreenReaderActive = Boolean(
+      process.env.ACCESSIBILITY === "true" || process.env.SCREEN_READER === "true"
+    );
+    // T-VIRTUAL-SCROLL: Enable debug mode for static output testing
+    // Set VELLUM_STATIC_OUTPUT=1 to enable non-replacing output (each update renders separately)
+    const isStaticOutputMode = Boolean(process.env.VELLUM_STATIC_OUTPUT === "1");
+
+    // T-VIRTUAL-SCROLL: Get working stdio proxies for Ink
+    // Always use proxied stdio to ensure proper write handling in VS Code terminal
+    const { stdout: inkStdoutProxy, stderr: inkStderr } = createWorkingStdio();
+
+    // Synchronized Output (DEC 2026): Only enable on Windows + VS Code terminal
+    // by reusing the existing detection in createCompatStdout().
+    const compatStdout = createCompatStdout();
+    const useBufferedStdout = compatStdout instanceof BufferedStdout;
+
+    // Set active stdout for modules that need to write outside Ink
+    if (useBufferedStdout) {
+      setActiveStdout(compatStdout);
+    }
+
+    // Preserve existing behavior everywhere else (proxy stdout).
+    const inkStdout: NodeJS.WriteStream = useBufferedStdout ? compatStdout : inkStdoutProxy;
+
+    if (useBufferedStdout) {
+      // Compose with existing shutdown cleanup so we don't lose agent cleanup.
+      const previousCleanup = getShutdownCleanup();
+      setShutdownCleanup(() => {
+        previousCleanup?.();
+        (compatStdout as BufferedStdout).dispose();
+      });
+    }
+
+    // Build Ink render options
+    // Key changes from Gemini CLI analysis:
+    // - Always use stdio proxy (prevents write interception issues)
+    // - patchConsole: false (prevent Ink from hijacking console)
+    // - exitOnCtrlC: false (let app handle Ctrl+C for proper cleanup)
+    // - incrementalRendering linked to alternateBuffer state
+    // - Removed maxFps: 20 (let Ink manage frame rate naturally)
+    const userAltBufferSetting = getAlternateBufferSetting();
+    const defaultAltBuffer = getDefaultAlternateBufferEnabled();
+    const resolvedAltBuffer = userAltBufferSetting ?? defaultAltBuffer;
+    const isConpty = isConptyTerminal();
+    const allowAlternateBuffer = !isConpty || userAltBufferSetting === true;
+    const useAlternateBuffer =
+      !isScreenReaderActive && !isStaticOutputMode && resolvedAltBuffer && allowAlternateBuffer;
+    const incrementalRendering = !isStaticOutputMode;
+    const inkRenderOptions = isVSCodeTerminal
+      ? {
+          stdout: inkStdout,
+          stderr: inkStderr,
+          stdin: process.stdin,
+          patchConsole: useAlternateBuffer,
+          exitOnCtrlC: false,
+          alternateBuffer: useAlternateBuffer,
+          incrementalRendering,
+          debug: isStaticOutputMode,
+        }
+      : {
+          stdout: inkStdout,
+          stderr: inkStderr,
+          stdin: process.stdin,
+          patchConsole: useAlternateBuffer,
+          exitOnCtrlC: false,
+          alternateBuffer: useAlternateBuffer,
+          incrementalRendering,
+          debug: isStaticOutputMode,
+        };
+
+    render(
+      <App
+        model={effectiveModel}
+        provider={effectiveProvider}
+        mode={effectiveMode}
+        approval={effectiveApproval}
+        sandbox={options.sandbox}
+        theme={options.theme as import("./tui/theme/index.js").ThemeName}
+        banner={options.banner}
+        agentLoop={agentLoop}
+        initError={initError}
+      />,
+      inkRenderOptions
+    );
+  });
+
+program
+  .command("run <prompt>")
+  .description("Run a single prompt")
+  .option("-m, --model <model>", "Model to use", "claude-sonnet-4-20250514")
+  .action(async (prompt, options) => {
+    console.log(`Running: ${prompt} with model ${options.model}`);
+    // TODO: Implement single run
+  });
+
+program
+  .command("config")
+  .description("Manage configuration")
+  .action(() => {
+    console.log("Config management coming soon");
+  });
+
+// =============================================================================
+// Credentials Command (T022)
+// =============================================================================
+
+const credentialsCmd = program
+  .command("credentials")
+  .description("Manage API credentials for LLM providers");
+
+credentialsCmd
+  .command("list", { isDefault: true })
+  .description("List all stored credentials (masked values)")
+  .action(() => {
+    renderCredentialsList();
+  });
+
+credentialsCmd
+  .command("add <provider>")
+  .description("Add or update credential for a provider")
+  .action((provider: string) => {
+    renderCredentialsAdd(provider);
+  });
+
+credentialsCmd
+  .command("remove <provider>")
+  .alias("rm")
+  .description("Remove credential for a provider")
+  .action((provider: string) => {
+    renderCredentialsRemove(provider);
+  });
+
+// =============================================================================
+// Init Command (T039-T041)
+// =============================================================================
+
+program
+  .command("init")
+  .description("Initialize AGENTS.md for your project")
+  .option("-f, --force", "Overwrite existing AGENTS.md without prompting")
+  .option("-m, --minimal", "Skip wizard prompts, use defaults")
+  .action(async (options) => {
+    const result = await executeInit({
+      force: options.force,
+      minimal: options.minimal,
+      nonInteractive: false,
+    });
+    process.exit(result.exitCode);
+  });
+
+// =============================================================================
+// Agents Command Group (T042-T046)
+// =============================================================================
+
+const agentsCmd = program.command("agents").description("Manage AGENTS.md configuration");
+
+agentsCmd
+  .command("show", { isDefault: true })
+  .description("Display merged AGENTS.md configuration")
+  .option("-j, --json", "Output as JSON")
+  .option("-v, --verbose", "Show all details including sources")
+  .option("-s, --scope <path>", "Show config for specific file/directory")
+  .action(async (options) => {
+    const result = await handleAgentsShow({
+      json: options.json,
+      verbose: options.verbose,
+      scope: options.scope,
+    });
+    console.log(getResultMessage(result));
+    process.exit(result.kind === "success" ? 0 : 1);
+  });
+
+agentsCmd
+  .command("validate [file]")
+  .description("Validate AGENTS.md syntax and structure")
+  .option("-v, --verbose", "Show verbose output")
+  .option("-j, --json", "Output as JSON")
+  .action(async (file, options) => {
+    const result = await handleAgentsValidate({
+      file,
+      verbose: options.verbose,
+      json: options.json,
+    });
+    console.log(getResultMessage(result));
+    process.exit(result.kind === "success" ? 0 : 1);
+  });
+
+agentsCmd
+  .command("generate")
+  .description("Generate AGENTS.md based on detected project stack")
+  .option("-o, --output <path>", "Output file path (default: ./AGENTS.md)")
+  .option("-m, --merge", "Merge with existing file")
+  .option("--dry-run", "Preview generated content without writing")
+  .action(async (options) => {
+    const result = await handleAgentsGenerate({
+      output: options.output,
+      merge: options.merge,
+      dryRun: options.dryRun,
+    });
+    console.log(getResultMessage(result));
+    process.exit(result.kind === "success" ? 0 : 1);
+  });
+
+// =============================================================================
+// Skill Command Group (T033-T037)
+// =============================================================================
+
+const skillCmd = program.command("skill").description("Manage skills for AI context");
+
+skillCmd
+  .command("list", { isDefault: true })
+  .description("List all available skills")
+  .option("-s, --source <source>", "Filter by source (workspace, user, global, builtin)")
+  .option("-j, --json", "Output as JSON")
+  .option("-v, --verbose", "Show full descriptions and triggers")
+  .action(async (options) => {
+    const result = await handleSkillList({
+      source: options.source,
+      json: options.json,
+      verbose: options.verbose,
+    });
+    console.log(getResultMessage(result));
+    process.exit(result.kind === "success" ? 0 : 1);
+  });
+
+skillCmd
+  .command("show <name>")
+  .description("Show details of a specific skill")
+  .option("-c, --content", "Show full SKILL.md content")
+  .option("-j, --json", "Output as JSON")
+  .action(async (name, options) => {
+    const result = await handleSkillShow(name, {
+      content: options.content,
+      json: options.json,
+    });
+    console.log(getResultMessage(result));
+    process.exit(result.kind === "success" ? 0 : 1);
+  });
+
+skillCmd
+  .command("create <name>")
+  .description("Create a new skill from template")
+  .option("-l, --location <location>", "Location: workspace, user, or global")
+  .option("-f, --force", "Overwrite if skill already exists")
+  .option("-n, --non-interactive", "Non-interactive mode (use defaults)")
+  .action(async (name, options) => {
+    const result = await handleSkillCreate(name, {
+      location: options.location,
+      force: options.force,
+      nonInteractive: options.nonInteractive,
+    });
+    process.exit(result.exitCode);
+  });
+
+skillCmd
+  .command("validate")
+  .description("Validate skill(s)")
+  .option("-s, --skill <name>", "Validate single skill by name")
+  .option("--strict", "Treat warnings as errors")
+  .option("-j, --json", "Output as JSON")
+  .action(async (options) => {
+    const result = await handleSkillValidate({
+      skill: options.skill,
+      strict: options.strict,
+      json: options.json,
+    });
+    console.log(getResultMessage(result));
+    process.exit(result.kind === "success" ? 0 : 1);
+  });
+
+// =============================================================================
+// Agent Commands (T044 - Multi-Agent Orchestration)
+// =============================================================================
+
+registerDelegateCommand(program);
+
+// =============================================================================
+// LSP Command (Phase 30)
+// =============================================================================
+
+program.addCommand(createLspCommand());
+
+program.parse();