whale-code 6.5.11 → 6.6.0

package/dist/cli/services/__tests__/agent-loop-utils-helpers.test.js

@@ -1,466 +0,0 @@
-/**
- * Tests for agent-loop.ts — utility functions: truncateResult, classifyToolError, canUseAgent,
- * getSessionTokens, getServerToolCount
- */
-
-import { describe, it, expect, vi, beforeEach } from "vitest";
-
-// ============================================================================
-// Mocks — must be declared before imports
-// ============================================================================
-
-vi.mock("../local-tools.js", () => ({
-  LOCAL_TOOL_DEFINITIONS: [{
-    name: "bash",
-    description: "Run shell commands",
-    input_schema: {
-      type: "object",
-      properties: {
-        command: {
-          type: "string"
-        }
-      }
-    }
-  }, {
-    name: "read_file",
-    description: "Read a file",
-    input_schema: {
-      type: "object",
-      properties: {
-        path: {
-          type: "string"
-        }
-      }
-    }
-  }],
-  executeLocalTool: vi.fn(async name => ({
-    success: true,
-    output: `local-tool-result:${name}`
-  })),
-  isLocalTool: vi.fn(name => ["bash", "read_file"].includes(name))
-}));
-vi.mock("../interactive-tools.js", () => ({
-  INTERACTIVE_TOOL_DEFINITIONS: [{
-    name: "ask_user_question",
-    description: "Ask user",
-    input_schema: {
-      type: "object",
-      properties: {
-        question: {
-          type: "string"
-        }
-      }
-    }
-  }],
-  executeInteractiveTool: vi.fn(async () => ({
-    success: true,
-    output: "user answered"
-  }))
-}));
-vi.mock("../config-store.js", () => ({
-  loadConfig: vi.fn(() => ({
-    access_token: "test-token",
-    refresh_token: "test-refresh",
-    proxy_url: "https://proxy.test"
-  })),
-  resolveConfig: vi.fn(() => ({
-    storeId: "store-123",
-    proxyUrl: "https://proxy.test"
-  })),
-  getProxyUrl: vi.fn(() => "https://proxy.test")
-}));
-vi.mock("../auth-service.js", () => ({
-  getValidToken: vi.fn(async () => "test-jwt-token"),
-  refreshSession: vi.fn(async () => ({
-    success: true,
-    config: {
-      access_token: "refreshed-token"
-    }
-  }))
-}));
-vi.mock("../server-tools.js", () => ({
-  isServerTool: vi.fn(() => false),
-  loadServerToolDefinitions: vi.fn(async () => []),
-  executeServerTool: vi.fn(async () => ({
-    success: true,
-    output: "server-result"
-  })),
-  getServerStatus: vi.fn(() => ({
-    connected: false
-  })),
-  setServerToolContext: vi.fn()
-}));
-vi.mock("../telemetry.js", () => ({
-  nextTurn: vi.fn(() => 1),
-  createTurnContext: vi.fn(() => ({
-    traceId: "trace-1",
-    spanId: "span-1",
-    conversationId: "conv-1",
-    userId: "user-1",
-    userEmail: "test@test.com"
-  })),
-  logSpan: vi.fn(),
-  generateSpanId: vi.fn(() => "span-gen-1"),
-  flushCliSpans: vi.fn()
-}));
-vi.mock("../error-logger.js", () => ({
-  captureError: vi.fn(),
-  addBreadcrumb: vi.fn()
-}));
-vi.mock("../agent-events.js", () => ({
-  AgentEventEmitter: vi.fn(),
-  setGlobalEmitter: vi.fn(),
-  clearGlobalEmitter: vi.fn()
-}));
-vi.mock("../mcp-client.js", () => ({
-  mcpClientManager: {
-    getTools: vi.fn(() => []),
-    isMcpTool: vi.fn(() => false),
-    callTool: vi.fn(async () => ({
-      success: true,
-      output: "mcp-result"
-    }))
-  }
-}));
-vi.mock("../hooks.js", () => ({
-  loadHooks: vi.fn(() => []),
-  runBeforeToolHook: vi.fn(async () => ({
-    allow: true
-  })),
-  runAfterToolHook: vi.fn(async () => ({})),
-  runSessionHook: vi.fn(async () => {})
-}));
-vi.mock("../../../shared/agent-core.js", () => {
-  class MockLoopDetector {
-    resetTurn = vi.fn();
-    getSessionStats = vi.fn(() => ({
-      totalErrors: 0,
-      failedStrategies: 0,
-      consecutiveFailedTurns: 0
-    }));
-  }
-  return {
-    LoopDetector: MockLoopDetector,
-    COMPACTION_TRIGGER_TOKENS: 167_000,
-    COMPACTION_TOTAL_BUDGET: 2_000_000,
-    DEFAULT_SESSION_COST_BUDGET_USD: 25,
-    AGENT_DEFAULTS: {
-      maxTurns: 200,
-      temperature: 1.0,
-      maxConcurrentTools: 7,
-      enableDelegation: true,
-      maxDurationMs: 900_000,
-      maxDurationMsWorkflow: 120_000,
-      enableModelRouting: true,
-      subagentMaxTokens: 32000,
-      subagentMaxTurns: 25,
-      subagentTemperature: 1.0,
-      subagentMaxRetries: 3,
-      maxContinuations: 10,
-      subagentMaxContinuations: 5,
-      maxMessageChars: 100_000,
-      maxToolResultChars: 2048,
-      compactionTriggerTokens: 167_000,
-      compactionTotalBudget: 2_000_000,
-      sessionCostBudgetUsd: 25.00,
-      teammateMaxTotalTurns: 50,
-      teamTimeoutMs: 600_000,
-      workerStallMs: 180_000,
-      teamReviewMaxTokens: 1024,
-      apiTimeoutMs: 90_000,
-      toolTimeoutMs: 60_000,
-      toolResultTruncateChars: 500_000
-    },
-    emitCostWarningIfNeeded: vi.fn(),
-    getCompactionConfig: vi.fn(() => ({
-      isNative: true,
-      triggerTokens: 167_000,
-      totalBudget: 2_000_000
-    })),
-    getContextManagement: vi.fn(() => ({
-      betas: [],
-      config: {
-        edits: []
-      }
-    })),
-    resolveAgentLoopConfig: vi.fn(() => ({
-      maxTurns: 200,
-      temperature: 1.0,
-      maxTokens: 32000,
-      maxConcurrentTools: 7,
-      enableDelegation: true,
-      maxDurationMs: 900_000,
-      enableModelRouting: true,
-      maxMessageChars: 100_000,
-      contextOverrides: {},
-      subagentMaxTokens: 32000,
-      subagentMaxTurns: 25,
-      subagentTemperature: 1.0,
-      maxContinuations: 10,
-      subagentMaxContinuations: 5,
-      subagentMaxRetries: 3,
-      maxToolResultChars: 2048,
-      defaultsUsed: []
-    }))
-  };
-});
-vi.mock("../agent-config.js", () => ({
-  loadCLIAgentConfig: vi.fn(async () => null),
-  resetAgentConfigCache: vi.fn()
-}));
-vi.mock("../../../shared/sse-parser.js", () => ({
-  parseSSEStream: vi.fn(() => async function* () {}()),
-  processStreamWithCallbacks: vi.fn(async () => ({
-    text: "Hello from Claude",
-    toolUseBlocks: [],
-    thinkingBlocks: [],
-    thinkingTokens: 0,
-    citations: [],
-    usage: {
-      inputTokens: 100,
-      outputTokens: 50,
-      cacheCreationTokens: 0,
-      cacheReadTokens: 0
-    },
-    compactionContent: null,
-    contextManagementApplied: false,
-    stopReason: "end_turn"
-  })),
-  collectStreamResult: vi.fn(async () => ({
-    text: "summary",
-    usage: {
-      inputTokens: 10,
-      outputTokens: 5,
-      cacheCreationTokens: 0,
-      cacheReadTokens: 0
-    }
-  }))
-}));
-vi.mock("../../../shared/api-client.js", () => ({
-  callServerProxy: vi.fn(async () => new ReadableStream()),
-  callTranscribe: vi.fn(async () => "transcribed text"),
-  buildAPIRequest: vi.fn(() => ({
-    maxTokens: 16384,
-    betas: ["context-management-2025-06-27"],
-    contextManagement: {
-      edits: []
-    },
-    thinking: undefined
-  })),
-  buildSystemBlocks: vi.fn((prompt, cost) => [{
-    type: "text",
-    text: prompt,
-    cache_control: {
-      type: "ephemeral"
-    }
-  }, {
-    type: "text",
-    text: cost
-  }]),
-  prepareWithCaching: vi.fn((tools, messages) => ({
-    tools,
-    messages
-  })),
-  trimGeminiContext: vi.fn(msgs => msgs),
-  trimOpenAIContext: vi.fn(msgs => msgs),
-  requestProviderCompaction: vi.fn(async () => null)
-}));
-vi.mock("../../../shared/constants.js", () => ({
-  MODELS: {
-    HAIKU: "claude-haiku-4-5-20251001",
-    SONNET: "claude-sonnet-4-6",
-    OPUS: "claude-opus-4-6"
-  },
-  getProvider: vi.fn(() => "anthropic")
-}));
-vi.mock("../../../shared/tool-dispatch.js", () => ({
-  dispatchTools: vi.fn(async () => ({
-    results: [],
-    bailOut: false,
-    bailMessage: undefined
-  })),
-  buildAssistantContent: vi.fn(({
-    text,
-    toolUseBlocks
-  }) => {
-    const blocks = [];
-    if (text) blocks.push({
-      type: "text",
-      text
-    });
-    for (const t of toolUseBlocks) blocks.push({
-      type: "tool_use",
-      id: t.id,
-      name: t.name,
-      input: t.input
-    });
-    return blocks;
-  })
-}));
-vi.mock("../memory-manager.js", () => ({
-  loadMemory: vi.fn(async () => "memory content"),
-  addMemory: vi.fn(async () => {}),
-  removeMemory: vi.fn(async () => {}),
-  listMemories: vi.fn(async () => [])
-}));
-vi.mock("../git-context.js", () => ({
-  refreshGitContext: vi.fn(async () => "git context"),
-  resetGitContext: vi.fn()
-}));
-vi.mock("../claude-md-loader.js", () => ({
-  loadClaudeMd: vi.fn(async () => "claude md"),
-  reloadClaudeMd: vi.fn(async () => "claude md"),
-  resetClaudeMdCache: vi.fn()
-}));
-vi.mock("../permission-modes.js", () => ({
-  setPermissionMode: vi.fn(),
-  getPermissionMode: vi.fn(() => "normal"),
-  isToolAllowedByPermission: vi.fn(() => true)
-}));
-vi.mock("../model-manager.js", () => ({
-  setModel: vi.fn(),
-  setModelById: vi.fn(),
-  getModel: vi.fn(() => "claude-sonnet-4-6"),
-  getModelShortName: vi.fn(() => "sonnet"),
-  estimateCostUsd: vi.fn(() => 0.005)
-}));
-vi.mock("../session-persistence.js", () => ({
-  saveSession: vi.fn(async () => {}),
-  loadSession: vi.fn(async () => null),
-  listSessions: vi.fn(async () => []),
-  findLatestSessionForCwd: vi.fn(async () => null)
-}));
-vi.mock("../system-prompt.js", () => ({
-  buildSystemPrompt: vi.fn(async () => "You are a helpful AI assistant.")
-}));
-vi.mock("../background-processes.js", () => ({
-  listProcesses: vi.fn(() => []),
-  listBackgroundAgents: vi.fn(() => [])
-}));
-
-// ============================================================================
-// Imports
-// ============================================================================
-
-import { resetSessionState, truncateResult, classifyToolError, canUseAgent, getSessionTokens, getServerToolCount } from "../agent-loop.js";
-import { loadServerToolDefinitions } from "../server-tools.js";
-import { loadConfig } from "../config-store.js";
-
-// ============================================================================
-// Tests
-// ============================================================================
-
-describe("agent-loop — utility functions", () => {
-  beforeEach(() => {
-    vi.clearAllMocks();
-    resetSessionState();
-    vi.mocked(loadConfig).mockReturnValue({
-      access_token: "test-token",
-      refresh_token: "test-refresh",
-      proxy_url: "https://proxy.test"
-    });
-  });
-  describe("truncateResult", () => {
-    it("returns short strings unchanged", () => {
-      expect(truncateResult("hello", 100)).toBe("hello");
-    });
-    it("returns exact length strings unchanged", () => {
-      const s = "a".repeat(50);
-      expect(truncateResult(s, 50)).toBe(s);
-    });
-    it("truncates long strings with char count", () => {
-      const s = "a".repeat(200);
-      const result = truncateResult(s, 50);
-      expect(result.length).toBeLessThan(200);
-      expect(result).toContain("... (200 chars total)");
-    });
-    it("keeps first maxLen characters", () => {
-      const result = truncateResult("abcdefghij", 5);
-      expect(result.startsWith("abcde")).toBe(true);
-    });
-  });
-  describe("classifyToolError", () => {
-    it("detects timeout errors", () => {
-      expect(classifyToolError("Command timed out after 30s")).toBe("timeout");
-      expect(classifyToolError("Request timeout")).toBe("timeout");
-    });
-    it("detects permission errors", () => {
-      expect(classifyToolError("Permission denied")).toBe("permission");
-      expect(classifyToolError("EACCES: /etc/secret")).toBe("permission");
-    });
-    it("detects not found errors", () => {
-      expect(classifyToolError("File not found: /tmp/missing")).toBe("not_found");
-      expect(classifyToolError("No such file or directory")).toBe("not_found");
-    });
-    it("detects command not found errors", () => {
-      expect(classifyToolError("exit code 127")).toBe("command_not_found");
-      expect(classifyToolError("command not found: foo")).toBe("not_found");
-    });
-    it("detects import errors", () => {
-      expect(classifyToolError("Import error: cannot find module")).toBe("import_error");
-    });
-    it("detects syntax errors", () => {
-      expect(classifyToolError("SyntaxError: unexpected token")).toBe("syntax_error");
-      expect(classifyToolError("Parse error at line 5")).toBe("syntax_error");
-    });
-    it("detects env managed errors", () => {
-      expect(classifyToolError("externally-managed environment")).toBe("env_managed");
-    });
-    it("returns unknown for unrecognized errors", () => {
-      expect(classifyToolError("Something went wrong")).toBe("unknown");
-    });
-    it("is case insensitive", () => {
-      expect(classifyToolError("TIMED OUT")).toBe("timeout");
-      expect(classifyToolError("PERMISSION DENIED")).toBe("permission");
-    });
-  });
-  describe("canUseAgent", () => {
-    it("returns ready when config has tokens", () => {
-      expect(canUseAgent().ready).toBe(true);
-    });
-    it("returns ready when env has API key", () => {
-      vi.mocked(loadConfig).mockReturnValueOnce({});
-      process.env.ANTHROPIC_API_KEY = "sk-test";
-      expect(canUseAgent().ready).toBe(true);
-      delete process.env.ANTHROPIC_API_KEY;
-    });
-    it("returns not ready with reason when no auth", () => {
-      vi.mocked(loadConfig).mockReturnValueOnce({});
-      delete process.env.ANTHROPIC_API_KEY;
-      const result = canUseAgent();
-      expect(result.ready).toBe(false);
-      expect(result.reason).toContain("whale login");
-    });
-  });
-  describe("getSessionTokens", () => {
-    it("returns zero tokens initially", () => {
-      const t = getSessionTokens();
-      expect(t.input).toBe(0);
-      expect(t.output).toBe(0);
-    });
-    it("resets to zero after resetSessionState", () => {
-      resetSessionState();
-      const t = getSessionTokens();
-      expect(t.input).toBe(0);
-      expect(t.output).toBe(0);
-    });
-  });
-  describe("getServerToolCount", () => {
-    it("returns 0 when server tools unavailable", async () => {
-      vi.mocked(loadServerToolDefinitions).mockRejectedValueOnce(new Error("offline"));
-      expect(await getServerToolCount()).toBe(0);
-    });
-    it("returns count of loaded tools", async () => {
-      vi.mocked(loadServerToolDefinitions).mockResolvedValueOnce([{
-        name: "tool1",
-        description: "d",
-        input_schema: {}
-      }, {
-        name: "tool2",
-        description: "d",
-        input_schema: {}
-      }]);
-      expect(await getServerToolCount()).toBe(2);
-    });
-  });
-});