whale-code 6.5.11 → 6.6.0

package/dist/shared/api-client-retry.test.js

@@ -1,307 +0,0 @@
-import { describe, it, expect, vi, beforeEach, afterEach } from "vitest";
-import { callServerProxy } from "./api-client.js";
-// ============================================================================
-// callServerProxy — Retry, Fallback, Abort, Timeout behavior
-// ============================================================================
-
-describe("callServerProxy (retry/fallback/abort)", () => {
-  let originalFetch;
-  beforeEach(() => {
-    originalFetch = globalThis.fetch;
-  });
-  afterEach(() => {
-    globalThis.fetch = originalFetch;
-  });
-  function mockFetch(impl) {
-    globalThis.fetch = vi.fn(impl);
-  }
-  function makeMockStream() {
-    return new ReadableStream({
-      start(controller) {
-        controller.enqueue(new TextEncoder().encode("data: hello\n\n"));
-        controller.close();
-      }
-    });
-  }
-  const baseConfig = {
-    proxyUrl: "https://whale-agent.fly.dev/proxy",
-    token: "test-jwt-token",
-    model: "claude-opus-4-6",
-    system: [{
-      type: "text",
-      text: "You are helpful."
-    }],
-    messages: [{
-      role: "user",
-      content: "Hello"
-    }],
-    tools: [{
-      name: "bash",
-      description: "Run commands"
-    }],
-    apiConfig: {
-      betas: ["context-management-2025-06-27"],
-      contextManagement: {
-        edits: [{
-          type: "clear_tool_uses_20250919",
-          trigger: {
-            type: "input_tokens",
-            value: 80_000
-          },
-          keep: {
-            type: "tool_uses",
-            value: 3
-          }
-        }]
-      },
-      maxTokens: 16384
-    }
-  };
-
-  // ---------- Retry behavior ----------
-
-  it("retries on 429 status with exponential backoff", async () => {
-    let callCount = 0;
-    mockFetch(async () => {
-      callCount++;
-      if (callCount <= 2) {
-        return new Response("Rate limited", {
-          status: 429
-        });
-      }
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    const result = await callServerProxy({
-      ...baseConfig
-    });
-    expect(result).toBeInstanceOf(ReadableStream);
-    expect(callCount).toBe(3);
-  });
-  it("retries on 500 status", async () => {
-    let callCount = 0;
-    mockFetch(async () => {
-      callCount++;
-      if (callCount === 1) {
-        return new Response("Internal Server Error", {
-          status: 500
-        });
-      }
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    const result = await callServerProxy({
-      ...baseConfig
-    });
-    expect(result).toBeInstanceOf(ReadableStream);
-    expect(callCount).toBe(2);
-  });
-  it("retries on 529 status (overloaded)", async () => {
-    let callCount = 0;
-    mockFetch(async () => {
-      callCount++;
-      if (callCount === 1) {
-        return new Response("Overloaded", {
-          status: 529
-        });
-      }
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    const result = await callServerProxy({
-      ...baseConfig
-    });
-    expect(result).toBeInstanceOf(ReadableStream);
-    expect(callCount).toBe(2);
-  });
-  it("throws after MAX_RETRIES (3) retries exhausted", async () => {
-    mockFetch(async () => new Response("Rate limited", {
-      status: 429
-    }));
-    await expect(callServerProxy({
-      ...baseConfig
-    })).rejects.toThrow("Proxy error (429)");
-    // 1 initial + 3 retries = 4 total attempts
-    expect(globalThis.fetch).toHaveBeenCalledTimes(4);
-  }, 15_000);
-  it("calls onRetry callback on each retry", async () => {
-    let callCount = 0;
-    mockFetch(async () => {
-      callCount++;
-      if (callCount <= 2) {
-        return new Response("Rate limited", {
-          status: 429
-        });
-      }
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    const onRetry = vi.fn();
-    await callServerProxy({
-      ...baseConfig,
-      onRetry
-    });
-    expect(onRetry).toHaveBeenCalledTimes(2);
-    expect(onRetry).toHaveBeenNthCalledWith(1, 1, 3, expect.stringContaining("429"));
-    expect(onRetry).toHaveBeenNthCalledWith(2, 2, 3, expect.stringContaining("429"));
-  });
-
-  // ---------- Fallback model ----------
-
-  it("switches to fallback model on last retry attempt", async () => {
-    let callCount = 0;
-    let lastModel;
-    mockFetch(async (_url, opts) => {
-      callCount++;
-      lastModel = JSON.parse(opts.body).model;
-      if (callCount <= 3) {
-        return new Response("Rate limited", {
-          status: 429
-        });
-      }
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    const onFallback = vi.fn();
-    await callServerProxy({
-      ...baseConfig,
-      fallbackModel: "claude-sonnet-4-6",
-      onFallback
-    });
-    expect(onFallback).toHaveBeenCalledWith("claude-opus-4-6", "claude-sonnet-4-6");
-    expect(lastModel).toBe("claude-sonnet-4-6");
-    expect(callCount).toBe(4);
-  }, 15_000);
-  it("does not call onFallback when no fallbackModel is set", async () => {
-    let callCount = 0;
-    mockFetch(async () => {
-      callCount++;
-      if (callCount <= 2) {
-        return new Response("Rate limited", {
-          status: 429
-        });
-      }
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    const onFallback = vi.fn();
-    await callServerProxy({
-      ...baseConfig,
-      onFallback
-    });
-    expect(onFallback).not.toHaveBeenCalled();
-  });
-
-  // ---------- Abort signal ----------
-
-  it("throws immediately when signal is already aborted", async () => {
-    const controller = new AbortController();
-    controller.abort();
-    mockFetch(async () => {
-      throw new DOMException("Aborted", "AbortError");
-    });
-    await expect(callServerProxy({
-      ...baseConfig,
-      signal: controller.signal
-    })).rejects.toThrow();
-    // Should not retry on abort
-    expect(globalThis.fetch).toHaveBeenCalledTimes(1);
-  });
-  it("does not retry when signal is aborted", async () => {
-    const controller = new AbortController();
-    let callCount = 0;
-    mockFetch(async () => {
-      callCount++;
-      controller.abort();
-      throw new DOMException("Aborted", "AbortError");
-    });
-    await expect(callServerProxy({
-      ...baseConfig,
-      signal: controller.signal
-    })).rejects.toThrow();
-    expect(callCount).toBe(1);
-  });
-
-  // ---------- Timeout ----------
-
-  it("applies timeout when timeoutMs is set and no signal provided", async () => {
-    let receivedSignal;
-    mockFetch(async (_url, opts) => {
-      receivedSignal = opts.signal;
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    await callServerProxy({
-      ...baseConfig,
-      timeoutMs: 5000
-    });
-
-    // The fetch should have received an AbortSignal from the timeout controller
-    expect(receivedSignal).toBeDefined();
-    expect(receivedSignal).toBeInstanceOf(AbortSignal);
-  });
-  it("does not create timeout controller when signal is already provided", async () => {
-    const controller = new AbortController();
-    let receivedSignal;
-    mockFetch(async (_url, opts) => {
-      receivedSignal = opts.signal;
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    await callServerProxy({
-      ...baseConfig,
-      signal: controller.signal,
-      timeoutMs: 5000
-    });
-
-    // Should use the user-provided signal, not a timeout-created one
-    expect(receivedSignal).toBe(controller.signal);
-  });
-
-  // ---------- Network errors (retryable by message) ----------
-
-  it("retries on network error with 'timeout' in message", async () => {
-    let callCount = 0;
-    mockFetch(async () => {
-      callCount++;
-      if (callCount === 1) {
-        throw Object.assign(new Error("Request timeout"), {
-          status: undefined
-        });
-      }
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    const result = await callServerProxy({
-      ...baseConfig
-    });
-    expect(result).toBeInstanceOf(ReadableStream);
-    expect(callCount).toBe(2);
-  });
-  it("retries on error with 'overloaded' in message", async () => {
-    let callCount = 0;
-    mockFetch(async () => {
-      callCount++;
-      if (callCount === 1) {
-        throw new Error("API is overloaded");
-      }
-      return new Response(makeMockStream(), {
-        status: 200
-      });
-    });
-    const result = await callServerProxy({
-      ...baseConfig
-    });
-    expect(result).toBeInstanceOf(ReadableStream);
-    expect(callCount).toBe(2);
-  });
-});

package/dist/shared/api-client-system-trimming.test.js

@@ -1,261 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { buildSystemBlocks, prepareWithCaching, trimOpenAIContext, trimGeminiContext } from "./api-client.js";
-
-// ============================================================================
-// buildSystemBlocks
-// ============================================================================
-
-describe("buildSystemBlocks", () => {
-  it("returns system prompt with cache_control when caching enabled", () => {
-    const blocks = buildSystemBlocks("You are helpful.", undefined, true);
-    expect(blocks).toHaveLength(1);
-    expect(blocks[0]).toEqual({
-      type: "text",
-      text: "You are helpful.",
-      cache_control: {
-        type: "ephemeral"
-      }
-    });
-  });
-  it("returns system prompt without cache_control when caching disabled", () => {
-    const blocks = buildSystemBlocks("You are helpful.", undefined, false);
-    expect(blocks).toHaveLength(1);
-    expect(blocks[0]).toEqual({
-      type: "text",
-      text: "You are helpful."
-    });
-  });
-  it("defaults to caching enabled", () => {
-    const blocks = buildSystemBlocks("Hello.");
-    expect(blocks[0]).toHaveProperty("cache_control", {
-      type: "ephemeral"
-    });
-  });
-  it("appends cost context as a second block", () => {
-    const blocks = buildSystemBlocks("System prompt.", "Session cost: $0.05");
-    expect(blocks).toHaveLength(2);
-    expect(blocks[0].text).toBe("System prompt.");
-    expect(blocks[1]).toEqual({
-      type: "text",
-      text: "Session cost: $0.05"
-    });
-  });
-  it("cost context block does not have cache_control (dynamic content)", () => {
-    const blocks = buildSystemBlocks("System.", "Cost: $1.00", true);
-    expect(blocks[1]).not.toHaveProperty("cache_control");
-  });
-  it("omits cost context block when undefined", () => {
-    const blocks = buildSystemBlocks("System.", undefined, true);
-    expect(blocks).toHaveLength(1);
-  });
-  it("omits cost context block when empty string (falsy)", () => {
-    const blocks = buildSystemBlocks("System.", "", true);
-    expect(blocks).toHaveLength(1);
-  });
-});
-
-// ============================================================================
-// prepareWithCaching
-// ============================================================================
-
-describe("prepareWithCaching", () => {
-  it("delegates to addPromptCaching and returns cached tools and messages", () => {
-    const tools = [{
-      name: "tool1"
-    }, {
-      name: "tool2"
-    }];
-    const messages = [{
-      role: "user",
-      content: "first"
-    }, {
-      role: "assistant",
-      content: "second"
-    }, {
-      role: "user",
-      content: "third"
-    }];
-    const result = prepareWithCaching(tools, messages);
-
-    // Last tool should have cache_control
-    expect(result.tools[result.tools.length - 1]).toHaveProperty("cache_control", {
-      type: "ephemeral"
-    });
-    // First tool should not
-    expect(result.tools[0]).not.toHaveProperty("cache_control");
-  });
-  it("does not mutate original arrays", () => {
-    const tools = [{
-      name: "t1"
-    }];
-    const messages = [{
-      role: "user",
-      content: "a"
-    }, {
-      role: "assistant",
-      content: "b"
-    }];
-    prepareWithCaching(tools, messages);
-    expect(tools[0]).not.toHaveProperty("cache_control");
-    expect(messages[0]).toEqual({
-      role: "user",
-      content: "a"
-    });
-  });
-  it("handles empty tools and messages", () => {
-    const result = prepareWithCaching([], []);
-    expect(result.tools).toEqual([]);
-    expect(result.messages).toEqual([]);
-  });
-});
-
-// ============================================================================
-// trimOpenAIContext
-// ============================================================================
-
-describe("trimOpenAIContext", () => {
-  function makeMessages(toolResultCount) {
-    const msgs = [];
-    for (let i = 0; i < toolResultCount; i++) {
-      msgs.push({
-        role: "assistant",
-        content: [{
-          type: "tool_use",
-          id: `t${i}`,
-          name: "bash",
-          input: {}
-        }]
-      });
-      msgs.push({
-        role: "user",
-        content: [{
-          type: "tool_result",
-          tool_use_id: `t${i}`,
-          content: `Result ${i}`
-        }]
-      });
-    }
-    return msgs;
-  }
-  it("returns same reference when below threshold", () => {
-    const msgs = makeMessages(10);
-    const result = trimOpenAIContext(msgs, 100_000); // 100K < 150K default threshold
-    expect(result).toBe(msgs); // exact same reference
-  });
-  it("returns same reference when above threshold but few tool results", () => {
-    const msgs = makeMessages(3); // 3 tool results <= keepRecent=5
-    const result = trimOpenAIContext(msgs, 200_000);
-    expect(result).toBe(msgs);
-  });
-  it("trims oldest tool results when above threshold", () => {
-    const msgs = makeMessages(10);
-    const result = trimOpenAIContext(msgs, 200_000, 150_000, 5);
-    expect(result).not.toBe(msgs); // new array
-    // Should have 5 trimmed (oldest) and 5 kept
-    let trimmedCount = 0;
-    let keptCount = 0;
-    for (const msg of result) {
-      if (Array.isArray(msg.content)) {
-        for (const block of msg.content) {
-          if (block.type === "tool_result") {
-            if (block.content === "[trimmed]") trimmedCount++;else keptCount++;
-          }
-        }
-      }
-    }
-    expect(trimmedCount).toBe(5);
-    expect(keptCount).toBe(5);
-  });
-  it("preserves most recent tool results (not oldest)", () => {
-    const msgs = makeMessages(8);
-    const result = trimOpenAIContext(msgs, 200_000, 150_000, 5);
-    // Last 5 tool results should be preserved
-    const toolResults = [];
-    for (const msg of result) {
-      if (Array.isArray(msg.content)) {
-        for (const block of msg.content) {
-          if (block.type === "tool_result") {
-            toolResults.push(block);
-          }
-        }
-      }
-    }
-    // First 3 should be trimmed, last 5 should be kept
-    expect(toolResults[0].content).toBe("[trimmed]");
-    expect(toolResults[1].content).toBe("[trimmed]");
-    expect(toolResults[2].content).toBe("[trimmed]");
-    expect(toolResults[3].content).toBe("Result 3");
-    expect(toolResults[7].content).toBe("Result 7");
-  });
-  it("respects custom threshold and keepRecent", () => {
-    const msgs = makeMessages(6);
-    const result = trimOpenAIContext(msgs, 100_000, 50_000, 2); // lower threshold, keep 2
-    let trimmedCount = 0;
-    for (const msg of result) {
-      if (Array.isArray(msg.content)) {
-        for (const block of msg.content) {
-          if (block.type === "tool_result" && block.content === "[trimmed]") trimmedCount++;
-        }
-      }
-    }
-    expect(trimmedCount).toBe(4); // 6 - 2 = 4 trimmed
-  });
-  it("handles messages with non-array content", () => {
-    const msgs = [{
-      role: "user",
-      content: "Hello"
-    }, ...makeMessages(6)];
-    // Should not crash on string content
-    const result = trimOpenAIContext(msgs, 200_000, 150_000, 3);
-    expect(result[0].content).toBe("Hello");
-  });
-});
-
-// ============================================================================
-// trimGeminiContext
-// ============================================================================
-
-describe("trimGeminiContext", () => {
-  function makeMessages(toolResultCount) {
-    const msgs = [];
-    for (let i = 0; i < toolResultCount; i++) {
-      msgs.push({
-        role: "assistant",
-        content: [{
-          type: "tool_use",
-          id: `t${i}`,
-          name: "bash",
-          input: {}
-        }]
-      });
-      msgs.push({
-        role: "user",
-        content: [{
-          type: "tool_result",
-          tool_use_id: `t${i}`,
-          content: `Result ${i}`
-        }]
-      });
-    }
-    return msgs;
-  }
-  it("returns same reference when below 950K threshold", () => {
-    const msgs = makeMessages(10);
-    const result = trimGeminiContext(msgs, 700_000);
-    expect(result).toBe(msgs);
-  });
-  it("trims when above 950K threshold", () => {
-    const msgs = makeMessages(10);
-    const result = trimGeminiContext(msgs, 960_000);
-    expect(result).not.toBe(msgs);
-    let trimmedCount = 0;
-    for (const msg of result) {
-      if (Array.isArray(msg.content)) {
-        for (const block of msg.content) {
-          if (block.type === "tool_result" && block.content === "[trimmed]") trimmedCount++;
-        }
-      }
-    }
-    expect(trimmedCount).toBe(5); // 10 - 5 (keepRecent default) = 5
-  });
-});

package/dist/shared/api-client-trimming.d.ts

@@ -1,36 +0,0 @@
-/**
- * Context Trimming and Compaction — extracted from api-client.ts
- *
- * Client-side context trimming for non-Anthropic providers (Gemini, OpenAI)
- * and server-side compaction request.
- */
-import type { CompactableMessage } from "./compaction.js";
-/**
- * Client-side context trimming for Gemini (no server-side context management).
- * When estimated input tokens exceed threshold, replaces old tool result contents
- * with "[trimmed]" to stay within Gemini's 1M context window.
- */
-export declare function trimGeminiContext<T extends CompactableMessage>(messages: T[], estimatedTokens: number, threshold?: number, keepRecent?: number): T[];
-export declare function trimOpenAIContext<T extends CompactableMessage>(messages: T[], estimatedTokens: number, threshold?: number, keepRecent?: number): T[];
-/**
- * Call server transcription endpoint (OpenAI Whisper via our Fly.io proxy).
- * Used by tool-dispatch when __AUDIO__ marker is detected in tool results.
- */
-export declare function callTranscribe(opts: {
-    proxyUrl: string;
-    token: string;
-    storeId: string;
-    audioBase64: string;
-    mediaType: string;
-}): Promise<string>;
-/**
- * Request conversation compaction from the server (Haiku summarization).
- * Used by non-Anthropic providers (OpenAI, Gemini) that lack native compaction.
- * Returns summary string or null on failure (caller keeps unsummarized history).
- */
-export declare function requestProviderCompaction(opts: {
-    proxyUrl: string;
-    token: string;
-    messages: object[];
-    systemPrompt: string;
-}): Promise<string | null>;