whale-code 6.5.6 → 6.5.8

package/dist/cli/services/__tests__/telemetry.part2.test.js

@@ -0,0 +1,195 @@
+import { describe, it, expect, beforeEach, vi } from "vitest";
+
+// ============================================================================
+// Mocks
+// ============================================================================
+
+const mockCreateAuthenticatedClient = vi.fn().mockReturnValue({
+  from: vi.fn()
+});
+const mockGetValidToken = vi.fn().mockResolvedValue(null);
+const mockResolveConfig = vi.fn().mockReturnValue({
+  supabaseUrl: "https://test.supabase.co",
+  supabaseKey: "test-key",
+  serverUrl: "https://whale-agent.fly.dev",
+  storeId: "store-default"
+});
+const mockLoadConfig = vi.fn().mockReturnValue({
+  user_id: "uid-1",
+  email: "u@test.com"
+});
+const mockCaptureError = vi.fn();
+
+// Mock fetch globally
+const mockFetch = vi.fn().mockResolvedValue({
+  ok: true,
+  json: async () => ({
+    success: true,
+    queued: 1
+  })
+});
+vi.stubGlobal("fetch", mockFetch);
+vi.mock("@supabase/supabase-js", () => ({
+  createClient: vi.fn()
+}));
+vi.mock("../config-store.js", () => ({
+  resolveConfig: () => mockResolveConfig(),
+  loadConfig: () => mockLoadConfig()
+}));
+vi.mock("../auth-service.js", () => ({
+  getValidToken: () => mockGetValidToken(),
+  createAuthenticatedClient: (...args) => mockCreateAuthenticatedClient(...args)
+}));
+vi.mock("../error-logger.js", () => ({
+  captureError: (...args) => mockCaptureError(...args)
+}));
+vi.mock("module", () => ({
+  createRequire: () => () => ({
+    version: "2.0.0-test"
+  })
+}));
+import { createTurnContext, logSpan, flushCliSpans } from "../telemetry.js";
+
+// Helper: log a span and flush immediately, returning the parsed spans from the fetch body
+async function logAndFlush(opts) {
+  logSpan(opts);
+  flushCliSpans();
+  // Wait for the async _sendSpans to complete
+  await vi.waitFor(() => {
+    expect(mockFetch).toHaveBeenCalled();
+  });
+  const lastCall = mockFetch.mock.calls[mockFetch.mock.calls.length - 1];
+  const body = JSON.parse(lastCall[1].body);
+  // Return the last span in the batch
+  return body.spans[body.spans.length - 1];
+}
+
+// ============================================================================
+// logSpan — spans are buffered and flushed to Fly.io server
+// ============================================================================
+
+describe("logSpan", () => {
+  beforeEach(() => {
+    vi.clearAllMocks();
+    mockResolveConfig.mockReturnValue({
+      supabaseUrl: "https://test.supabase.co",
+      supabaseKey: "test-key",
+      serverUrl: "https://whale-agent.fly.dev",
+      storeId: "store-default"
+    });
+  });
+  it("does not throw (fire-and-forget)", () => {
+    const ctx = createTurnContext();
+    expect(() => logSpan({
+      action: "test.action",
+      durationMs: 100,
+      context: ctx
+    })).not.toThrow();
+  });
+  it("sends spans to server via HTTP POST on flush", async () => {
+    const ctx = createTurnContext();
+    logSpan({
+      action: "test.insert",
+      durationMs: 50,
+      context: ctx
+    });
+    flushCliSpans();
+    await vi.waitFor(() => {
+      expect(mockFetch).toHaveBeenCalled();
+    });
+    const [url, options] = mockFetch.mock.calls[mockFetch.mock.calls.length - 1];
+    expect(url).toBe("https://whale-agent.fly.dev");
+    expect(options.method).toBe("POST");
+    expect(options.headers["Authorization"]).toBe("Bearer test-key");
+    const body = JSON.parse(options.body);
+    expect(body.mode).toBe("telemetry_ingest");
+    expect(body.spans.length).toBeGreaterThanOrEqual(1);
+    expect(body.spans.find(s => s.action === "test.insert")).toBeTruthy();
+  });
+  it("sets severity to error and status_code to ERROR when error present", async () => {
+    const ctx = createTurnContext();
+    const span = await logAndFlush({
+      action: "test.error",
+      durationMs: 10,
+      context: ctx,
+      error: "something failed"
+    });
+    expect(span.severity).toBe("error");
+    expect(span.status_code).toBe("ERROR");
+    expect(span.error_message).toBe("something failed");
+  });
+  it("bridges errors to captureError", async () => {
+    const ctx = createTurnContext();
+    logSpan({
+      action: "test.bridge",
+      durationMs: 10,
+      context: ctx,
+      error: "bridge test"
+    });
+    await vi.waitFor(() => {
+      expect(mockCaptureError).toHaveBeenCalledWith(expect.objectContaining({
+        errorType: "test.bridge",
+        errorMessage: "bridge test",
+        severity: "error"
+      }));
+    });
+  });
+  it("does not call captureError when no error", async () => {
+    const ctx = createTurnContext();
+    logSpan({
+      action: "test.ok",
+      durationMs: 10,
+      context: ctx
+    });
+    flushCliSpans();
+    await vi.waitFor(() => {
+      expect(mockFetch).toHaveBeenCalled();
+    });
+    expect(mockCaptureError).not.toHaveBeenCalled();
+  });
+  it("includes storeId from opts", async () => {
+    const ctx = createTurnContext();
+    const span = await logAndFlush({
+      action: "test.store",
+      durationMs: 10,
+      context: ctx,
+      storeId: "store-42"
+    });
+    expect(span.store_id).toBe("store-42");
+  });
+  it("passes details into the details field", async () => {
+    const ctx = createTurnContext();
+    const span = await logAndFlush({
+      action: "test.details",
+      durationMs: 5,
+      context: ctx,
+      details: {
+        foo: "bar"
+      }
+    });
+    expect(span.details).toMatchObject({
+      foo: "bar"
+    });
+  });
+  it("uses explicit severity override", async () => {
+    const ctx = createTurnContext();
+    const span = await logAndFlush({
+      action: "test.warn",
+      durationMs: 1,
+      context: ctx,
+      severity: "warn"
+    });
+    expect(span.severity).toBe("warn");
+  });
+  it("does not throw on fetch error", async () => {
+    mockFetch.mockRejectedValueOnce(new Error("network fail"));
+    const ctx = createTurnContext();
+    expect(() => logSpan({
+      action: "test.fetcherr",
+      durationMs: 1,
+      context: ctx
+    })).not.toThrow();
+    flushCliSpans();
+    // Should not throw even though fetch failed
+  });
+});

package/dist/cli/services/__tests__/telemetry.test.js

@@ -0,0 +1,176 @@
+import { describe, it, expect, vi } from "vitest";
+
+// ============================================================================
+// Mocks
+// ============================================================================
+
+const mockCreateAuthenticatedClient = vi.fn().mockReturnValue({
+  from: vi.fn()
+});
+const mockGetValidToken = vi.fn().mockResolvedValue(null);
+const mockResolveConfig = vi.fn().mockReturnValue({
+  supabaseUrl: "https://test.supabase.co",
+  supabaseKey: "test-key",
+  serverUrl: "https://whale-agent.fly.dev",
+  storeId: "store-default"
+});
+const mockLoadConfig = vi.fn().mockReturnValue({
+  user_id: "uid-1",
+  email: "u@test.com"
+});
+const mockCaptureError = vi.fn();
+
+// Mock fetch globally
+const mockFetch = vi.fn().mockResolvedValue({
+  ok: true,
+  json: async () => ({
+    success: true,
+    queued: 1
+  })
+});
+vi.stubGlobal("fetch", mockFetch);
+vi.mock("@supabase/supabase-js", () => ({
+  createClient: vi.fn()
+}));
+vi.mock("../config-store.js", () => ({
+  resolveConfig: () => mockResolveConfig(),
+  loadConfig: () => mockLoadConfig()
+}));
+vi.mock("../auth-service.js", () => ({
+  getValidToken: () => mockGetValidToken(),
+  createAuthenticatedClient: (...args) => mockCreateAuthenticatedClient(...args)
+}));
+vi.mock("../error-logger.js", () => ({
+  captureError: (...args) => mockCaptureError(...args)
+}));
+vi.mock("module", () => ({
+  createRequire: () => () => ({
+    version: "2.0.0-test"
+  })
+}));
+import { setConversationId, getConversationId, initializeTelemetryClient, generateTraceId, generateSpanId, nextTurn, createTurnContext, getTurnNumber, logSpan, flushCliSpans } from "../telemetry.js";
+
+// Helper: log a span and flush immediately, returning the parsed spans from the fetch body
+async function logAndFlush(opts) {
+  logSpan(opts);
+  flushCliSpans();
+  // Wait for the async _sendSpans to complete
+  await vi.waitFor(() => {
+    expect(mockFetch).toHaveBeenCalled();
+  });
+  const lastCall = mockFetch.mock.calls[mockFetch.mock.calls.length - 1];
+  const body = JSON.parse(lastCall[1].body);
+  // Return the last span in the batch
+  return body.spans[body.spans.length - 1];
+}
+
+// ============================================================================
+// Session state
+// ============================================================================
+
+describe("session state", () => {
+  it("getConversationId returns a UUID initially", () => {
+    const id = getConversationId();
+    expect(id).toMatch(/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/);
+  });
+  it("setConversationId changes the conversation id", () => {
+    const custom = "custom-conv-123";
+    setConversationId(custom);
+    expect(getConversationId()).toBe(custom);
+  });
+});
+
+// ============================================================================
+// Turn management
+// ============================================================================
+
+describe("turn management", () => {
+  it("nextTurn increments the turn number", () => {
+    const before = getTurnNumber();
+    const n = nextTurn();
+    expect(n).toBe(before + 1);
+    expect(getTurnNumber()).toBe(n);
+  });
+  it("nextTurn increments on each call", () => {
+    const a = nextTurn();
+    const b = nextTurn();
+    const c = nextTurn();
+    expect(b).toBe(a + 1);
+    expect(c).toBe(b + 1);
+  });
+});
+
+// ============================================================================
+// Trace/Span ID generators
+// ============================================================================
+
+describe("generateTraceId", () => {
+  it("returns a UUID", () => {
+    const id = generateTraceId();
+    expect(id).toMatch(/^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/);
+  });
+  it("generates unique IDs", () => {
+    const a = generateTraceId();
+    const b = generateTraceId();
+    expect(a).not.toBe(b);
+  });
+});
+describe("generateSpanId", () => {
+  it("returns a 16-character hex string", () => {
+    const id = generateSpanId();
+    expect(id).toMatch(/^[0-9a-f]{16}$/);
+  });
+  it("generates unique IDs", () => {
+    const a = generateSpanId();
+    const b = generateSpanId();
+    expect(a).not.toBe(b);
+  });
+});
+
+// ============================================================================
+// createTurnContext
+// ============================================================================
+
+describe("createTurnContext", () => {
+  it("returns an ExecutionContext with defaults", () => {
+    const ctx = createTurnContext();
+    expect(ctx.source).toBe("whale_cli");
+    expect(ctx.serviceName).toBe("whale-code");
+    expect(ctx.traceId).toBeTruthy();
+    expect(ctx.spanId).toBeTruthy();
+    expect(ctx.traceFlags).toBe(1);
+    expect(ctx.conversationId).toBeTruthy();
+    expect(ctx.turnNumber).toBeDefined();
+  });
+  it("loads user info from config", () => {
+    const ctx = createTurnContext();
+    expect(ctx.userId).toBe("uid-1");
+    expect(ctx.userEmail).toBe("u@test.com");
+  });
+  it("applies overrides", () => {
+    const ctx = createTurnContext({
+      model: "sonnet",
+      inputTokens: 100,
+      outputTokens: 200
+    });
+    expect(ctx.model).toBe("sonnet");
+    expect(ctx.inputTokens).toBe(100);
+    expect(ctx.outputTokens).toBe(200);
+  });
+  it("overrides can replace source", () => {
+    const ctx = createTurnContext({
+      source: "custom_source"
+    });
+    expect(ctx.source).toBe("custom_source");
+  });
+});
+
+// ============================================================================
+// initializeTelemetryClient
+// ============================================================================
+
+describe("initializeTelemetryClient", () => {
+  it("does not throw", () => {
+    expect(() => initializeTelemetryClient("some-token")).not.toThrow();
+  });
+});

package/dist/cli/services/agent-event-types.d.ts

@@ -0,0 +1,148 @@
+/**
+ * Agent Event type definitions.
+ * Extracted from agent-events.ts so the emitter class stays under 300 lines.
+ */
+import type Anthropic from "@anthropic-ai/sdk";
+export interface AgentTextEvent {
+    type: "text";
+    text: string;
+    accumulated: string;
+}
+export interface AgentToolStartEvent {
+    type: "tool_start";
+    id: string;
+    name: string;
+}
+export interface AgentToolEndEvent {
+    type: "tool_end";
+    id: string;
+    name: string;
+    success: boolean;
+    result?: string;
+    input?: Record<string, unknown>;
+    durationMs: number;
+}
+export interface AgentUsageEvent {
+    type: "usage";
+    inputTokens: number;
+    outputTokens: number;
+    model?: string;
+    costUsd?: number;
+    cacheReadTokens?: number;
+    cacheCreationTokens?: number;
+}
+export interface AgentDoneEvent {
+    type: "done";
+    text: string;
+    messages: Anthropic.MessageParam[];
+}
+export interface AgentErrorEvent {
+    type: "error";
+    error: string;
+}
+export interface AgentCompactEvent {
+    type: "compact";
+    before: number;
+    after: number;
+    tokensSaved: number;
+}
+export interface ToolOutputEvent {
+    type: "tool_output";
+    toolName: string;
+    line: string;
+}
+export interface ToolProgressEvent {
+    type: "tool_progress";
+    toolName: string;
+    progress: {
+        phase?: string;
+        elapsed_s: number;
+        stdout_lines: number;
+        stderr_lines: number;
+        stdout_bytes: number;
+        last_line?: string;
+    };
+}
+export interface SubagentStartEvent {
+    type: "subagent_start";
+    id: string;
+    agentType: string;
+    model: string;
+    description: string;
+}
+export interface SubagentProgressEvent {
+    type: "subagent_progress";
+    id: string;
+    agentType: string;
+    message: string;
+    turn?: number;
+    toolName?: string;
+}
+export interface SubagentDoneEvent {
+    type: "subagent_done";
+    id: string;
+    agentType: string;
+    success: boolean;
+    output: string;
+    tokens: {
+        input: number;
+        output: number;
+    };
+    tools: string[];
+    durationMs: number;
+}
+export interface SubagentToolStartEvent {
+    type: "subagent_tool_start";
+    agentId: string;
+    toolName: string;
+    toolInput?: Record<string, unknown>;
+}
+export interface SubagentToolEndEvent {
+    type: "subagent_tool_end";
+    agentId: string;
+    toolName: string;
+    success: boolean;
+    durationMs: number;
+}
+export interface TeamStartEvent {
+    type: "team_start";
+    teamId: string;
+    name: string;
+    teammateCount: number;
+    taskCount: number;
+}
+export interface TeamProgressEvent {
+    type: "team_progress";
+    teamId: string;
+    teammateId: string;
+    teammateName: string;
+    message: string;
+    taskId?: string;
+}
+export interface TeamTaskEvent {
+    type: "team_task";
+    teamId: string;
+    teammateId: string;
+    taskId: string;
+    taskDescription: string;
+    status: "started" | "completed" | "failed";
+    result?: string;
+}
+export interface TeamDoneEvent {
+    type: "team_done";
+    teamId: string;
+    success: boolean;
+    summary: string;
+    tasksCompleted: number;
+    tasksTotal: number;
+    tokens: {
+        input: number;
+        output: number;
+    };
+    durationMs: number;
+}
+export interface AgentThinkingEvent {
+    type: "thinking";
+    chunkCount: number;
+}
+export type AgentEvent = AgentTextEvent | AgentThinkingEvent | AgentToolStartEvent | AgentToolEndEvent | AgentUsageEvent | AgentDoneEvent | AgentErrorEvent | AgentCompactEvent | ToolOutputEvent | ToolProgressEvent | SubagentStartEvent | SubagentProgressEvent | SubagentDoneEvent | SubagentToolStartEvent | SubagentToolEndEvent | TeamStartEvent | TeamProgressEvent | TeamTaskEvent | TeamDoneEvent;

package/dist/cli/services/agent-event-types.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=agent-event-types.js.map

package/dist/cli/services/agent-event-types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent-event-types.js","names":[],"sources":["../../../src/cli/services/agent-event-types.ts"],"sourcesContent":["/**\n * Agent Event type definitions.\n * Extracted from agent-events.ts so the emitter class stays under 300 lines.\n */\n\nimport type Anthropic from \"@anthropic-ai/sdk\";\n\n// ============================================================================\n// CORE AGENT EVENTS\n// ============================================================================\n\nexport interface AgentTextEvent {\n  type: \"text\";\n  text: string;\n  accumulated: string;\n}\n\nexport interface AgentToolStartEvent {\n  type: \"tool_start\";\n  id: string;\n  name: string;\n}\n\nexport interface AgentToolEndEvent {\n  type: \"tool_end\";\n  id: string;\n  name: string;\n  success: boolean;\n  result?: string;\n  input?: Record<string, unknown>;\n  durationMs: number;\n}\n\nexport interface AgentUsageEvent {\n  type: \"usage\";\n  inputTokens: number;\n  outputTokens: number;\n  model?: string;\n  costUsd?: number;\n  cacheReadTokens?: number;\n  cacheCreationTokens?: number;\n}\n\nexport interface AgentDoneEvent {\n  type: \"done\";\n  text: string;\n  messages: Anthropic.MessageParam[];\n}\n\nexport interface AgentErrorEvent {\n  type: \"error\";\n  error: string;\n}\n\nexport interface AgentCompactEvent {\n  type: \"compact\";\n  before: number;\n  after: number;\n  tokensSaved: number;\n}\n\nexport interface ToolOutputEvent {\n  type: \"tool_output\";\n  toolName: string;\n  line: string;\n}\n\nexport interface ToolProgressEvent {\n  type: \"tool_progress\";\n  toolName: string;\n  progress: {\n    phase?: string;\n    elapsed_s: number;\n    stdout_lines: number;\n    stderr_lines: number;\n    stdout_bytes: number;\n    last_line?: string;\n  };\n}\n\n// ============================================================================\n// SUBAGENT EVENTS\n// ============================================================================\n\nexport interface SubagentStartEvent {\n  type: \"subagent_start\";\n  id: string;\n  agentType: string;\n  model: string;\n  description: string;\n}\n\nexport interface SubagentProgressEvent {\n  type: \"subagent_progress\";\n  id: string;\n  agentType: string;\n  message: string;\n  turn?: number;\n  toolName?: string;\n}\n\nexport interface SubagentDoneEvent {\n  type: \"subagent_done\";\n  id: string;\n  agentType: string;\n  success: boolean;\n  output: string;\n  tokens: { input: number; output: number };\n  tools: string[];\n  durationMs: number;\n}\n\nexport interface SubagentToolStartEvent {\n  type: \"subagent_tool_start\";\n  agentId: string;\n  toolName: string;\n  toolInput?: Record<string, unknown>;\n}\n\nexport interface SubagentToolEndEvent {\n  type: \"subagent_tool_end\";\n  agentId: string;\n  toolName: string;\n  success: boolean;\n  durationMs: number;\n}\n\n// ============================================================================\n// TEAM EVENTS\n// ============================================================================\n\nexport interface TeamStartEvent {\n  type: \"team_start\";\n  teamId: string;\n  name: string;\n  teammateCount: number;\n  taskCount: number;\n}\n\nexport interface TeamProgressEvent {\n  type: \"team_progress\";\n  teamId: string;\n  teammateId: string;\n  teammateName: string;\n  message: string;\n  taskId?: string;\n}\n\nexport interface TeamTaskEvent {\n  type: \"team_task\";\n  teamId: string;\n  teammateId: string;\n  taskId: string;\n  taskDescription: string;\n  status: \"started\" | \"completed\" | \"failed\";\n  result?: string;\n}\n\nexport interface TeamDoneEvent {\n  type: \"team_done\";\n  teamId: string;\n  success: boolean;\n  summary: string;\n  tasksCompleted: number;\n  tasksTotal: number;\n  tokens: { input: number; output: number };\n  durationMs: number;\n}\n\n// ============================================================================\n// THINKING EVENT\n// ============================================================================\n\nexport interface AgentThinkingEvent {\n  type: \"thinking\";\n  chunkCount: number;\n}\n\n// ============================================================================\n// UNION TYPE\n// ============================================================================\n\nexport type AgentEvent =\n  | AgentTextEvent\n  | AgentThinkingEvent\n  | AgentToolStartEvent\n  | AgentToolEndEvent\n  | AgentUsageEvent\n  | AgentDoneEvent\n  | AgentErrorEvent\n  | AgentCompactEvent\n  | ToolOutputEvent\n  | ToolProgressEvent\n  | SubagentStartEvent\n  | SubagentProgressEvent\n  | SubagentDoneEvent\n  | SubagentToolStartEvent\n  | SubagentToolEndEvent\n  | TeamStartEvent\n  | TeamProgressEvent\n  | TeamTaskEvent\n  | TeamDoneEvent;\n"],"mappings":"","ignoreList":[]}

package/dist/cli/services/agent-loop-iteration.d.ts

@@ -0,0 +1,13 @@
+/**
+ * Agent Loop Iteration — context trimming, compaction handling, provider compaction
+ * Extracted from agent-loop.ts for file size limits.
+ */
+import type Anthropic from "@anthropic-ai/sdk";
+import type { AgentLoopCallbacks } from "./agent-loop-types.js";
+import type { AgentEventEmitter } from "./agent-loop-types.js";
+export declare function trimContextForProvider(currentModel: string, cachedMessages: Anthropic.MessageParam[], prevIterationInputTokens: number, callbacks: AgentLoopCallbacks, emitter?: AgentEventEmitter): Anthropic.MessageParam[];
+export declare function handleProviderCompaction(currentModel: string, messages: Anthropic.MessageParam[], systemPrompt: string, inputTokens: number, compactionCount: number, token: string, callbacks: AgentLoopCallbacks, emitter?: AgentEventEmitter, turnCtx?: any, storeId?: string): Promise<{
+    compacted: boolean;
+    compactionCount: number;
+    decrementIteration: boolean;
+}>;

package/dist/cli/services/agent-loop-setup.d.ts

@@ -0,0 +1,32 @@
+/**
+ * Agent Loop Setup — initialization and configuration for runAgentLoop.
+ * Extracted from agent-loop.ts.
+ */
+import type Anthropic from "@anthropic-ai/sdk";
+import { createTurnContext } from "./telemetry.js";
+import { LoopDetector, resolveAgentLoopConfig } from "../../shared/agent-core.js";
+import type { AgentLoopOptions } from "./agent-loop-types.js";
+import { buildToolExecutor } from "./agent-loop-tools.js";
+export interface AgentLoopSetupResult {
+    tools: any[];
+    systemPrompt: string;
+    messages: Anthropic.MessageParam[];
+    effectiveMaxTurns: number;
+    shellSummarization: boolean;
+    sessionStart: number;
+    storeId: string;
+    turnNum: number;
+    turnCtx: ReturnType<typeof createTurnContext>;
+    activeModel: string;
+    effectiveBudget: number;
+    ctxOverrides: any;
+    ctxMgmt: any;
+    effectiveCompactionTrigger: number;
+    effectiveCompactionBudget: number;
+    maxContinuations: number;
+    loopDetector: LoopDetector;
+    toolExecutor: ReturnType<typeof buildToolExecutor>;
+    hooks: any[];
+    resolved: ReturnType<typeof resolveAgentLoopConfig> | null;
+}
+export declare function setupAgentLoop(opts: AgentLoopOptions): Promise<AgentLoopSetupResult>;

package/dist/cli/services/agent-loop-shell-summarize.d.ts

@@ -0,0 +1,11 @@
+/**
+ * Agent Loop — shell output summarization.
+ *
+ * Summarizes long bash output using Haiku via server proxy to save context window.
+ */
+import type { ToolResultMessage } from "../../shared/types.js";
+/**
+ * Post-process tool results to summarize long bash output.
+ * Only affects bash tool results that exceed size/line thresholds.
+ */
+export declare function summarizeLongToolResults(toolResults: ToolResultMessage[], toolNames: Map<string, string>, proxyUrl: string, token: string, shellSummarization: boolean, storeId?: string): Promise<ToolResultMessage[]>;

package/dist/cli/services/agent-loop-shell-summarize.js

@@ -0,0 +1,94 @@
+/**
+ * Agent Loop — shell output summarization.
+ *
+ * Summarizes long bash output using Haiku via server proxy to save context window.
+ */
+
+import { parseSSEStream, collectStreamResult } from "../../shared/sse-parser.js";
+import { callServerProxy, buildAPIRequest } from "../../shared/api-client.js";
+import { MODELS } from "../../shared/constants.js";
+// ============================================================================
+// CONSTANTS
+// ============================================================================
+
+const SHELL_SUMMARIZE_LINE_THRESHOLD = 800;
+const SHELL_SUMMARIZE_SIZE_THRESHOLD = 200_000; // 200KB
+const SHELL_SUMMARIZE_MAX_INPUT = 300_000; // 300KB max to summarizer
+const SHELL_SUMMARIZE_ORIGINAL_PREVIEW_LINES = 20;
+
+// ============================================================================
+// HELPERS
+// ============================================================================
+
+/**
+ * Check if shell output should be summarized based on line count or size.
+ */
+function shouldSummarizeShellOutput(output) {
+  if (output.length > SHELL_SUMMARIZE_SIZE_THRESHOLD) return true;
+  const lineCount = output.split("\n").length;
+  return lineCount > SHELL_SUMMARIZE_LINE_THRESHOLD;
+}
+
+/**
+ * Summarize long shell output using Haiku via server proxy.
+ * Returns summarized output or original if summarization fails.
+ */
+async function summarizeShellOutput(output, proxyUrl, token, storeId) {
+  const lineCount = output.split("\n").length;
+  const truncatedForSummary = output.length > SHELL_SUMMARIZE_MAX_INPUT ? output.slice(0, SHELL_SUMMARIZE_MAX_INPUT) + "\n... (truncated for summarization)" : output;
+  try {
+    const summaryConfig = buildAPIRequest({
+      model: MODELS.HAIKU,
+      contextProfile: "subagent"
+    });
+    const stream = await callServerProxy({
+      proxyUrl,
+      token,
+      model: MODELS.HAIKU,
+      system: [{
+        type: "text",
+        text: "You are a concise technical summarizer. Summarize shell/command output preserving key information, errors, warnings, file paths, and actionable items. Be brief but thorough."
+      }],
+      messages: [{
+        role: "user",
+        content: `Summarize this shell output concisely, preserving key information, errors, and actionable items:\n\n${truncatedForSummary}`
+      }],
+      tools: [],
+      apiConfig: summaryConfig,
+      storeId,
+      timeoutMs: 15_000
+    });
+    const result = await collectStreamResult(parseSSEStream(stream));
+    const summary = result.text.trim();
+    if (!summary) return output;
+    const originalLines = output.split("\n");
+    const preview = originalLines.slice(0, SHELL_SUMMARIZE_ORIGINAL_PREVIEW_LINES).join("\n");
+    return `[Summarized from ${lineCount} lines]\n\n${summary}\n\n[First ${SHELL_SUMMARIZE_ORIGINAL_PREVIEW_LINES} lines of original output]\n${preview}`;
+  } catch {
+    return output;
+  }
+}
+
+// ============================================================================
+// PUBLIC API
+// ============================================================================
+
+/**
+ * Post-process tool results to summarize long bash output.
+ * Only affects bash tool results that exceed size/line thresholds.
+ */
+export async function summarizeLongToolResults(toolResults, toolNames, proxyUrl, token, shellSummarization, storeId) {
+  if (!shellSummarization) return toolResults;
+  const tasks = toolResults.map(async result => {
+    const toolName = toolNames.get(result.tool_use_id);
+    if (toolName !== "bash" || typeof result.content !== "string") return result;
+    if (!shouldSummarizeShellOutput(result.content)) return result;
+    const summarized = await summarizeShellOutput(result.content, proxyUrl, token, storeId);
+    return {
+      ...result,
+      content: summarized
+    };
+  });
+  return Promise.all(tasks);
+}
+//# sourceMappingURL=agent-loop-shell-summarize.js.map