mirror of
https://github.com/claude-code-best/claude-code.git
synced 2026-06-18 14:25:51 +00:00
feat: 重构供应商层次 (#286)
* refactor: 创建 @anthropic-ai/model-provider 包骨架与类型定义
- 新建 workspace 包 packages/@anthropic-ai/model-provider
- 定义 ModelProviderHooks 接口（依赖注入：分析、成本、日志等）
- 定义 ClientFactories 接口（Anthropic/OpenAI/Gemini/Grok 客户端工厂）
- 搬入核心类型：Message 体系、NonNullableUsage、EMPTY_USAGE、SystemPrompt、错误常量
- 主项目 src/types/message.ts 等改为 re-export，保持向后兼容
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: 提升 OpenAI 转换器和模型映射到 model-provider 包
- 搬入 OpenAI 消息转换（convertMessages）、工具转换（convertTools）、流适配（streamAdapter）
- 搬入 OpenAI 和 Grok 模型映射（resolveOpenAIModel、resolveGrokModel）
- 主项目文件改为 thin re-export proxy
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: 搬入 Gemini 兼容层到 model-provider 包
- 搬入 Gemini 类型定义、消息转换、工具转换、流适配、模型映射
- 主项目 gemini/ 目录下文件改为 thin re-export proxy
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: 搬入 errorUtils 并迁移消费者导入到 model-provider
- 搬入 formatAPIError、extractConnectionErrorDetails 等 errorUtils
- 迁移 10 个消费者文件直接从 @anthropic-ai/model-provider 导入
- 更新 emptyUsage、sdkUtilityTypes、systemPromptType 为 re-export proxy
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: compact 模型降级为 -1 模式（Opus→Sonnet, Sonnet→Haiku）
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: 添加 agent-loop 绘图
* Revert "feat: compact 模型降级为 -1 模式（Opus→Sonnet, Sonnet→Haiku）"
This reverts commit e458d6391d.
* docs: 添加简化版 agent loop
* fix: 修复 n 快捷键导致关闭的问题
* fix: 修复 node 下 ws 没打包问题
* docs: 修复链接
* test: 添加测试支持
* fix: 修复类型问题(#267) (#271)
* fix: 修复 Bun 的 polyfill 问题
* fix: 类型修复完成
* feat: 统一所有包的类型文件
* fix: 修复构建问题
* test: 修复类型校验 (#279)
* fix: 修复 Bun 的 polyfill 问题
* fix: 类型修复完成
* feat: 统一所有包的类型文件
* fix: 修复构建问题
* fix(remote-control): harden self-hosted session flows (#278)
Co-authored-by: chengzifeng <chengzifeng@meituan.com>
* docs: update contributors
* build: 新增 vite 构建流程
* feat: 添加环境变量支持以覆盖 max_tokens 设置
* feat(langfuse): LLM generation 记录工具定义
将 Anthropic 格式的工具定义转换为 Langfuse 兼容的 OpenAI 格式，
并在 generation 的 input 中以 { messages, tools } 结构传入，
以便在 Langfuse UI 中查看完整的工具定义信息。
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: 添加对 ACP 协议的支持 (#284)
* feat: 适配 zed acp 协议
* docs: 完善 acp 文档
* chore: 1.4.0
* conflict: 解决冲突
* feat: 添加测试覆盖率上报
* style: 改名加移动文件夹位置
* refactor: 移动测试用例及实现
* test: 修复测试用例完成
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Cheng Zi Feng <1154238323@qq.com>
Co-authored-by: chengzifeng <chengzifeng@meituan.com>
Co-authored-by: claude-code-best <272536312+claude-code-best@users.noreply.github.com>
This commit is contained in:
claude-code-best
@@ -1,457 +0,0 @@
|
||||
import { describe, expect, test } from 'bun:test'
|
||||
import { anthropicMessagesToOpenAI } from '../convertMessages.js'
|
||||
import type { UserMessage, AssistantMessage } from '../../../../types/message.js'
|
||||
|
||||
// Helpers to create internal-format messages
|
||||
function makeUserMsg(content: string | any[]): UserMessage {
|
||||
return {
|
||||
type: 'user',
|
||||
uuid: '00000000-0000-0000-0000-000000000000',
|
||||
message: { role: 'user', content },
|
||||
} as UserMessage
|
||||
}
|
||||
|
||||
function makeAssistantMsg(content: string | any[]): AssistantMessage {
|
||||
return {
|
||||
type: 'assistant',
|
||||
uuid: '00000000-0000-0000-0000-000000000001',
|
||||
message: { role: 'assistant', content },
|
||||
} as AssistantMessage
|
||||
}
|
||||
|
||||
describe('anthropicMessagesToOpenAI', () => {
|
||||
test('converts system prompt to system message', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg('hello')],
|
||||
['You are helpful.'] as any,
|
||||
)
|
||||
expect(result[0]).toEqual({ role: 'system', content: 'You are helpful.' })
|
||||
})
|
||||
|
||||
test('joins multiple system prompt strings', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg('hi')],
|
||||
['Part 1', 'Part 2'] as any,
|
||||
)
|
||||
expect(result[0]).toEqual({ role: 'system', content: 'Part 1\n\nPart 2' })
|
||||
})
|
||||
|
||||
test('skips empty system prompt', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg('hi')],
|
||||
[] as any,
|
||||
)
|
||||
expect(result[0].role).toBe('user')
|
||||
})
|
||||
|
||||
test('converts simple user text message', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg('hello world')],
|
||||
[] as any,
|
||||
)
|
||||
expect(result).toEqual([{ role: 'user', content: 'hello world' }])
|
||||
})
|
||||
|
||||
test('converts user message with content array', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg([
|
||||
{ type: 'text', text: 'line 1' },
|
||||
{ type: 'text', text: 'line 2' },
|
||||
])],
|
||||
[] as any,
|
||||
)
|
||||
expect(result).toEqual([{ role: 'user', content: 'line 1\nline 2' }])
|
||||
})
|
||||
|
||||
test('converts assistant message with text', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeAssistantMsg('response text')],
|
||||
[] as any,
|
||||
)
|
||||
expect(result).toEqual([{ role: 'assistant', content: 'response text' }])
|
||||
})
|
||||
|
||||
test('converts assistant message with tool_use', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeAssistantMsg([
|
||||
{ type: 'text', text: 'Let me help.' },
|
||||
{
|
||||
type: 'tool_use' as const,
|
||||
id: 'toolu_123',
|
||||
name: 'bash',
|
||||
input: { command: 'ls' },
|
||||
},
|
||||
])],
|
||||
[] as any,
|
||||
)
|
||||
expect(result).toEqual([{
|
||||
role: 'assistant',
|
||||
content: 'Let me help.',
|
||||
tool_calls: [{
|
||||
id: 'toolu_123',
|
||||
type: 'function',
|
||||
function: { name: 'bash', arguments: '{"command":"ls"}' },
|
||||
}],
|
||||
}])
|
||||
})
|
||||
|
||||
test('converts tool_result to tool message', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg([
|
||||
{
|
||||
type: 'tool_result' as const,
|
||||
tool_use_id: 'toolu_123',
|
||||
content: 'file1.txt\nfile2.txt',
|
||||
},
|
||||
])],
|
||||
[] as any,
|
||||
)
|
||||
expect(result).toEqual([{
|
||||
role: 'tool',
|
||||
tool_call_id: 'toolu_123',
|
||||
content: 'file1.txt\nfile2.txt',
|
||||
}])
|
||||
})
|
||||
|
||||
test('strips thinking blocks', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'internal thoughts...' },
|
||||
{ type: 'text', text: 'visible response' },
|
||||
])],
|
||||
[] as any,
|
||||
)
|
||||
expect(result).toEqual([{ role: 'assistant', content: 'visible response' }])
|
||||
})
|
||||
|
||||
test('handles full conversation with tools', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[
|
||||
makeUserMsg('list files'),
|
||||
makeAssistantMsg([
|
||||
{
|
||||
type: 'tool_use' as const,
|
||||
id: 'toolu_abc',
|
||||
name: 'bash',
|
||||
input: { command: 'ls' },
|
||||
},
|
||||
]),
|
||||
makeUserMsg([
|
||||
{
|
||||
type: 'tool_result' as const,
|
||||
tool_use_id: 'toolu_abc',
|
||||
content: 'file.txt',
|
||||
},
|
||||
]),
|
||||
],
|
||||
['You are helpful.'] as any,
|
||||
)
|
||||
|
||||
expect(result).toHaveLength(4)
|
||||
expect(result[0].role).toBe('system')
|
||||
expect(result[1].role).toBe('user')
|
||||
expect(result[2].role).toBe('assistant')
|
||||
expect((result[2] as any).tool_calls).toBeDefined()
|
||||
expect(result[3].role).toBe('tool')
|
||||
})
|
||||
|
||||
test('converts base64 image to image_url', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg([
|
||||
{ type: 'text', text: 'what is this?' },
|
||||
{
|
||||
type: 'image' as const,
|
||||
source: {
|
||||
type: 'base64',
|
||||
media_type: 'image/png',
|
||||
data: 'iVBORw0KGgo=',
|
||||
},
|
||||
},
|
||||
])],
|
||||
[] as any,
|
||||
)
|
||||
expect(result).toEqual([{
|
||||
role: 'user',
|
||||
content: [
|
||||
{ type: 'text', text: 'what is this?' },
|
||||
{
|
||||
type: 'image_url',
|
||||
image_url: { url: 'data:image/png;base64,iVBORw0KGgo=' },
|
||||
},
|
||||
],
|
||||
}])
|
||||
})
|
||||
|
||||
test('converts url image to image_url', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg([
|
||||
{
|
||||
type: 'image' as const,
|
||||
source: {
|
||||
type: 'url',
|
||||
url: 'https://example.com/img.png',
|
||||
},
|
||||
},
|
||||
])],
|
||||
[] as any,
|
||||
)
|
||||
expect(result).toEqual([{
|
||||
role: 'user',
|
||||
content: [
|
||||
{
|
||||
type: 'image_url',
|
||||
image_url: { url: 'https://example.com/img.png' },
|
||||
},
|
||||
],
|
||||
}])
|
||||
})
|
||||
|
||||
test('converts image-only message without text', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg([
|
||||
{
|
||||
type: 'image' as const,
|
||||
source: {
|
||||
type: 'base64',
|
||||
media_type: 'image/jpeg',
|
||||
data: '/9j/4AAQ',
|
||||
},
|
||||
},
|
||||
])],
|
||||
[] as any,
|
||||
)
|
||||
expect(result).toEqual([{
|
||||
role: 'user',
|
||||
content: [
|
||||
{
|
||||
type: 'image_url',
|
||||
image_url: { url: 'data:image/jpeg;base64,/9j/4AAQ' },
|
||||
},
|
||||
],
|
||||
}])
|
||||
})
|
||||
|
||||
test('defaults to image/png when media_type is missing', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg([
|
||||
{
|
||||
type: 'image' as const,
|
||||
source: {
|
||||
type: 'base64',
|
||||
data: 'ABC123',
|
||||
},
|
||||
},
|
||||
])],
|
||||
[] as any,
|
||||
)
|
||||
expect((result[0].content as any[])[0].image_url.url).toBe(
|
||||
'data:image/png;base64,ABC123',
|
||||
)
|
||||
})
|
||||
})
|
||||
|
||||
describe('DeepSeek thinking mode (enableThinking)', () => {
|
||||
test('preserves thinking block as reasoning_content when enabled', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg('question'), makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'Let me reason about this...' },
|
||||
{ type: 'text', text: 'The answer is 42.' },
|
||||
])],
|
||||
[] as any,
|
||||
{ enableThinking: true },
|
||||
)
|
||||
// Should have: user, assistant with reasoning_content
|
||||
expect(result).toHaveLength(2)
|
||||
expect(result[0].role).toBe('user')
|
||||
const assistant = result[1] as any
|
||||
expect(assistant.role).toBe('assistant')
|
||||
expect(assistant.content).toBe('The answer is 42.')
|
||||
expect(assistant.reasoning_content).toBe('Let me reason about this...')
|
||||
})
|
||||
|
||||
test('drops thinking block when enableThinking is false (default)', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'internal thoughts...' },
|
||||
{ type: 'text', text: 'visible response' },
|
||||
])],
|
||||
[] as any,
|
||||
)
|
||||
const assistant = result[0] as any
|
||||
expect(assistant.content).toBe('visible response')
|
||||
expect(assistant.reasoning_content).toBeUndefined()
|
||||
})
|
||||
|
||||
test('preserves reasoning_content with tool_calls in same turn', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[
|
||||
makeUserMsg('what is the weather?'),
|
||||
makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'I need to call the weather tool.' },
|
||||
{ type: 'text', text: '' },
|
||||
{
|
||||
type: 'tool_use' as const,
|
||||
id: 'toolu_001',
|
||||
name: 'get_weather',
|
||||
input: { location: 'Hangzhou' },
|
||||
},
|
||||
]),
|
||||
makeUserMsg([
|
||||
{
|
||||
type: 'tool_result' as const,
|
||||
tool_use_id: 'toolu_001',
|
||||
content: 'Cloudy 7~13°C',
|
||||
},
|
||||
]),
|
||||
],
|
||||
[] as any,
|
||||
{ enableThinking: true },
|
||||
)
|
||||
|
||||
// Find the assistant message
|
||||
const assistants = result.filter(m => m.role === 'assistant')
|
||||
expect(assistants.length).toBe(1)
|
||||
const assistant = assistants[0] as any
|
||||
expect(assistant.reasoning_content).toBe('I need to call the weather tool.')
|
||||
expect(assistant.tool_calls).toBeDefined()
|
||||
expect(assistant.tool_calls[0].function.name).toBe('get_weather')
|
||||
})
|
||||
|
||||
test('strips reasoning_content from previous turns', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[
|
||||
// Turn 1: user → assistant (with thinking)
|
||||
makeUserMsg('question 1'),
|
||||
makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'Turn 1 reasoning...' },
|
||||
{ type: 'text', text: 'Turn 1 answer' },
|
||||
]),
|
||||
// Turn 2: new user message → previous reasoning should be stripped
|
||||
makeUserMsg('question 2'),
|
||||
makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'Turn 2 reasoning...' },
|
||||
{ type: 'text', text: 'Turn 2 answer' },
|
||||
]),
|
||||
],
|
||||
[] as any,
|
||||
{ enableThinking: true },
|
||||
)
|
||||
|
||||
const assistants = result.filter(m => m.role === 'assistant')
|
||||
// Turn 1 assistant: reasoning should be stripped (previous turn)
|
||||
expect((assistants[0] as any).reasoning_content).toBeUndefined()
|
||||
expect((assistants[0] as any).content).toBe('Turn 1 answer')
|
||||
// Turn 2 assistant: reasoning should be preserved (current turn)
|
||||
expect((assistants[1] as any).reasoning_content).toBe('Turn 2 reasoning...')
|
||||
expect((assistants[1] as any).content).toBe('Turn 2 answer')
|
||||
})
|
||||
|
||||
test('preserves reasoning_content in multi-iteration tool call within same turn', () => {
|
||||
// Simulates a full DeepSeek tool call iteration:
|
||||
// user → assistant(thinking+tool_call) → tool_result → assistant(thinking+tool_call) → tool_result → assistant(thinking+text)
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[
|
||||
makeUserMsg("tomorrow's weather in Hangzhou"),
|
||||
// Iteration 1: thinking + tool call
|
||||
makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'I need the date first.' },
|
||||
{
|
||||
type: 'tool_use' as const,
|
||||
id: 'toolu_001',
|
||||
name: 'get_date',
|
||||
input: {},
|
||||
},
|
||||
]),
|
||||
makeUserMsg([
|
||||
{
|
||||
type: 'tool_result' as const,
|
||||
tool_use_id: 'toolu_001',
|
||||
content: '2026-04-08',
|
||||
},
|
||||
]),
|
||||
// Iteration 2: thinking + tool call
|
||||
makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'Now I can get the weather.' },
|
||||
{
|
||||
type: 'tool_use' as const,
|
||||
id: 'toolu_002',
|
||||
name: 'get_weather',
|
||||
input: { location: 'Hangzhou', date: '2026-04-08' },
|
||||
},
|
||||
]),
|
||||
makeUserMsg([
|
||||
{
|
||||
type: 'tool_result' as const,
|
||||
tool_use_id: 'toolu_002',
|
||||
content: 'Cloudy 7~13°C',
|
||||
},
|
||||
]),
|
||||
// Iteration 3: thinking + final answer
|
||||
makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'I have the info now.' },
|
||||
{ type: 'text', text: 'Tomorrow will be cloudy, 7-13°C.' },
|
||||
]),
|
||||
],
|
||||
[] as any,
|
||||
{ enableThinking: true },
|
||||
)
|
||||
|
||||
// All 3 assistant messages are in the current turn (after last user msg is the last tool_result,
|
||||
// but the "last user message" boundary logic finds the last user-typed message).
|
||||
// Actually, tool_result messages are also UserMessage type, so the last user message
|
||||
// is the one with tool_result for toolu_002. All assistant messages after that should have reasoning.
|
||||
const assistants = result.filter(m => m.role === 'assistant')
|
||||
expect(assistants.length).toBe(3)
|
||||
// All iterations within the same turn preserve reasoning
|
||||
expect((assistants[0] as any).reasoning_content).toBe('I need the date first.')
|
||||
expect((assistants[1] as any).reasoning_content).toBe('Now I can get the weather.')
|
||||
expect((assistants[2] as any).reasoning_content).toBe('I have the info now.')
|
||||
})
|
||||
|
||||
test('handles multiple thinking blocks in single assistant message', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg('question'), makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'First thought.' },
|
||||
{ type: 'thinking' as const, thinking: 'Second thought.' },
|
||||
{ type: 'text', text: 'Final answer.' },
|
||||
])],
|
||||
[] as any,
|
||||
{ enableThinking: true },
|
||||
)
|
||||
const assistant = result.filter(m => m.role === 'assistant')[0] as any
|
||||
expect(assistant.reasoning_content).toBe('First thought.\nSecond thought.')
|
||||
})
|
||||
|
||||
test('skips empty thinking blocks', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg('question'), makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: '' },
|
||||
{ type: 'text', text: 'Answer.' },
|
||||
])],
|
||||
[] as any,
|
||||
{ enableThinking: true },
|
||||
)
|
||||
const assistant = result.filter(m => m.role === 'assistant')[0] as any
|
||||
expect(assistant.reasoning_content).toBeUndefined()
|
||||
})
|
||||
|
||||
test('sets content to null when only thinking and tool_calls present', () => {
|
||||
const result = anthropicMessagesToOpenAI(
|
||||
[makeUserMsg('question'), makeAssistantMsg([
|
||||
{ type: 'thinking' as const, thinking: 'Reasoning only.' },
|
||||
{
|
||||
type: 'tool_use' as const,
|
||||
id: 'toolu_001',
|
||||
name: 'bash',
|
||||
input: { command: 'ls' },
|
||||
},
|
||||
])],
|
||||
[] as any,
|
||||
{ enableThinking: true },
|
||||
)
|
||||
const assistant = result.filter(m => m.role === 'assistant')[0] as any
|
||||
expect(assistant.content).toBeNull()
|
||||
expect(assistant.reasoning_content).toBe('Reasoning only.')
|
||||
expect(assistant.tool_calls).toHaveLength(1)
|
||||
})
|
||||
})
|
||||
@@ -1,167 +0,0 @@
|
||||
import { describe, expect, test } from 'bun:test'
|
||||
import { anthropicToolsToOpenAI, anthropicToolChoiceToOpenAI } from '../convertTools.js'
|
||||
|
||||
describe('anthropicToolsToOpenAI', () => {
|
||||
test('converts basic tool', () => {
|
||||
const tools = [
|
||||
{
|
||||
type: 'custom',
|
||||
name: 'bash',
|
||||
description: 'Run a bash command',
|
||||
input_schema: {
|
||||
type: 'object',
|
||||
properties: { command: { type: 'string' } },
|
||||
required: ['command'],
|
||||
},
|
||||
},
|
||||
]
|
||||
|
||||
const result = anthropicToolsToOpenAI(tools as any)
|
||||
|
||||
expect(result).toEqual([{
|
||||
type: 'function',
|
||||
function: {
|
||||
name: 'bash',
|
||||
description: 'Run a bash command',
|
||||
parameters: {
|
||||
type: 'object',
|
||||
properties: { command: { type: 'string' } },
|
||||
required: ['command'],
|
||||
},
|
||||
},
|
||||
}])
|
||||
})
|
||||
|
||||
test('uses empty schema when input_schema missing', () => {
|
||||
const tools = [{ type: 'custom', name: 'noop', description: 'no-op' }]
|
||||
const result = anthropicToolsToOpenAI(tools as any)
|
||||
|
||||
expect((result[0] as { function: { parameters: unknown } }).function.parameters).toEqual({ type: 'object', properties: {} })
|
||||
})
|
||||
|
||||
test('strips Anthropic-specific fields', () => {
|
||||
const tools = [
|
||||
{
|
||||
type: 'custom',
|
||||
name: 'bash',
|
||||
description: 'Run bash',
|
||||
input_schema: { type: 'object', properties: {} },
|
||||
cache_control: { type: 'ephemeral' },
|
||||
defer_loading: true,
|
||||
},
|
||||
]
|
||||
const result = anthropicToolsToOpenAI(tools as any)
|
||||
|
||||
expect((result[0] as any).cache_control).toBeUndefined()
|
||||
expect((result[0] as any).defer_loading).toBeUndefined()
|
||||
})
|
||||
|
||||
test('handles empty tools array', () => {
|
||||
expect(anthropicToolsToOpenAI([])).toEqual([])
|
||||
})
|
||||
|
||||
test('sanitizes const to enum in tool schema', () => {
|
||||
const tools = [
|
||||
{
|
||||
type: 'custom',
|
||||
name: 'test',
|
||||
description: 'test tool',
|
||||
input_schema: {
|
||||
type: 'object',
|
||||
properties: {
|
||||
mode: { const: 'read' },
|
||||
name: { type: 'string' },
|
||||
},
|
||||
},
|
||||
},
|
||||
]
|
||||
const result = anthropicToolsToOpenAI(tools as any)
|
||||
const props = (result[0] as { function: { parameters: any } }).function.parameters as any
|
||||
expect(props.properties.mode).toEqual({ enum: ['read'] })
|
||||
expect(props.properties.mode.const).toBeUndefined()
|
||||
expect(props.properties.name).toEqual({ type: 'string' })
|
||||
})
|
||||
|
||||
test('sanitizes const in deeply nested schemas', () => {
|
||||
const tools = [
|
||||
{
|
||||
type: 'custom',
|
||||
name: 'deep',
|
||||
description: 'nested const',
|
||||
input_schema: {
|
||||
type: 'object',
|
||||
properties: {
|
||||
outer: {
|
||||
type: 'object',
|
||||
properties: {
|
||||
inner: { const: 'fixed' },
|
||||
},
|
||||
},
|
||||
},
|
||||
definitions: {
|
||||
MyType: {
|
||||
type: 'object',
|
||||
properties: {
|
||||
field: { const: 42 },
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
]
|
||||
const result = anthropicToolsToOpenAI(tools as any)
|
||||
const params = (result[0] as { function: { parameters: any } }).function.parameters as any
|
||||
expect(params.properties.outer.properties.inner).toEqual({ enum: ['fixed'] })
|
||||
expect(params.definitions.MyType.properties.field).toEqual({ enum: [42] })
|
||||
})
|
||||
|
||||
test('sanitizes const in anyOf/oneOf/allOf', () => {
|
||||
const tools = [
|
||||
{
|
||||
type: 'custom',
|
||||
name: 'union',
|
||||
description: 'union test',
|
||||
input_schema: {
|
||||
type: 'object',
|
||||
properties: {
|
||||
val: {
|
||||
anyOf: [
|
||||
{ const: 'a' },
|
||||
{ const: 'b' },
|
||||
{ type: 'string' },
|
||||
],
|
||||
},
|
||||
},
|
||||
},
|
||||
},
|
||||
]
|
||||
const result = anthropicToolsToOpenAI(tools as any)
|
||||
const anyOf = ((result[0] as { function: { parameters: any } }).function.parameters as any).properties.val.anyOf
|
||||
expect(anyOf[0]).toEqual({ enum: ['a'] })
|
||||
expect(anyOf[1]).toEqual({ enum: ['b'] })
|
||||
expect(anyOf[2]).toEqual({ type: 'string' })
|
||||
})
|
||||
})
|
||||
|
||||
describe('anthropicToolChoiceToOpenAI', () => {
|
||||
test('maps auto', () => {
|
||||
expect(anthropicToolChoiceToOpenAI({ type: 'auto' })).toBe('auto')
|
||||
})
|
||||
|
||||
test('maps any to required', () => {
|
||||
expect(anthropicToolChoiceToOpenAI({ type: 'any' })).toBe('required')
|
||||
})
|
||||
|
||||
test('maps tool to function', () => {
|
||||
const result = anthropicToolChoiceToOpenAI({ type: 'tool', name: 'bash' })
|
||||
expect(result).toEqual({ type: 'function', function: { name: 'bash' } })
|
||||
})
|
||||
|
||||
test('returns undefined for undefined input', () => {
|
||||
expect(anthropicToolChoiceToOpenAI(undefined)).toBeUndefined()
|
||||
})
|
||||
|
||||
test('returns undefined for unknown type', () => {
|
||||
expect(anthropicToolChoiceToOpenAI({ type: 'unknown' })).toBeUndefined()
|
||||
})
|
||||
})
|
||||
@@ -1,68 +0,0 @@
|
||||
import { describe, expect, test, beforeEach, afterEach } from 'bun:test'
|
||||
import { resolveOpenAIModel } from '../modelMapping.js'
|
||||
|
||||
describe('resolveOpenAIModel', () => {
|
||||
const originalEnv = {
|
||||
OPENAI_MODEL: process.env.OPENAI_MODEL,
|
||||
OPENAI_DEFAULT_HAIKU_MODEL: process.env.OPENAI_DEFAULT_HAIKU_MODEL,
|
||||
OPENAI_DEFAULT_SONNET_MODEL: process.env.OPENAI_DEFAULT_SONNET_MODEL,
|
||||
OPENAI_DEFAULT_OPUS_MODEL: process.env.OPENAI_DEFAULT_OPUS_MODEL,
|
||||
ANTHROPIC_DEFAULT_HAIKU_MODEL: process.env.ANTHROPIC_DEFAULT_HAIKU_MODEL,
|
||||
ANTHROPIC_DEFAULT_SONNET_MODEL: process.env.ANTHROPIC_DEFAULT_SONNET_MODEL,
|
||||
ANTHROPIC_DEFAULT_OPUS_MODEL: process.env.ANTHROPIC_DEFAULT_OPUS_MODEL,
|
||||
}
|
||||
|
||||
beforeEach(() => {
|
||||
delete process.env.OPENAI_MODEL
|
||||
delete process.env.OPENAI_DEFAULT_HAIKU_MODEL
|
||||
delete process.env.OPENAI_DEFAULT_SONNET_MODEL
|
||||
delete process.env.OPENAI_DEFAULT_OPUS_MODEL
|
||||
delete process.env.ANTHROPIC_DEFAULT_HAIKU_MODEL
|
||||
delete process.env.ANTHROPIC_DEFAULT_SONNET_MODEL
|
||||
delete process.env.ANTHROPIC_DEFAULT_OPUS_MODEL
|
||||
})
|
||||
|
||||
afterEach(() => {
|
||||
Object.assign(process.env, originalEnv)
|
||||
})
|
||||
|
||||
test('OPENAI_MODEL env var overrides all', () => {
|
||||
process.env.OPENAI_MODEL = 'my-custom-model'
|
||||
expect(resolveOpenAIModel('claude-sonnet-4-6')).toBe('my-custom-model')
|
||||
})
|
||||
|
||||
test('ANTHROPIC_DEFAULT_SONNET_MODEL overrides default map', () => {
|
||||
process.env.ANTHROPIC_DEFAULT_SONNET_MODEL = 'my-sonnet'
|
||||
expect(resolveOpenAIModel('claude-sonnet-4-6')).toBe('my-sonnet')
|
||||
})
|
||||
|
||||
test('ANTHROPIC_DEFAULT_HAIKU_MODEL overrides default map', () => {
|
||||
process.env.ANTHROPIC_DEFAULT_HAIKU_MODEL = 'my-haiku'
|
||||
expect(resolveOpenAIModel('claude-haiku-4-5-20251001')).toBe('my-haiku')
|
||||
})
|
||||
|
||||
test('ANTHROPIC_DEFAULT_OPUS_MODEL overrides default map', () => {
|
||||
process.env.ANTHROPIC_DEFAULT_OPUS_MODEL = 'my-opus'
|
||||
expect(resolveOpenAIModel('claude-opus-4-6')).toBe('my-opus')
|
||||
})
|
||||
|
||||
test('maps known Anthropic model via DEFAULT_MODEL_MAP', () => {
|
||||
expect(resolveOpenAIModel('claude-sonnet-4-6')).toBe('gpt-4o')
|
||||
})
|
||||
|
||||
test('maps haiku model', () => {
|
||||
expect(resolveOpenAIModel('claude-haiku-4-5-20251001')).toBe('gpt-4o-mini')
|
||||
})
|
||||
|
||||
test('maps opus model', () => {
|
||||
expect(resolveOpenAIModel('claude-opus-4-6')).toBe('o3')
|
||||
})
|
||||
|
||||
test('passes through unknown model name', () => {
|
||||
expect(resolveOpenAIModel('some-random-model')).toBe('some-random-model')
|
||||
})
|
||||
|
||||
test('strips [1m] suffix', () => {
|
||||
expect(resolveOpenAIModel('claude-sonnet-4-6[1m]')).toBe('gpt-4o')
|
||||
})
|
||||
})
|
||||
@@ -1,487 +0,0 @@
|
||||
/**
|
||||
* Tests for queryModelOpenAI in index.ts.
|
||||
*
|
||||
* Focused on the two bugs fixed:
|
||||
* 1. stop_reason was always null in the assembled AssistantMessage because
|
||||
* partialMessage (from message_start) has stop_reason: null, and the
|
||||
* stop_reason captured from message_delta was never applied.
|
||||
* 2. partialMessage was not reset to null after message_stop, so the safety
|
||||
* fallback at the end of the loop would yield a second identical
|
||||
* AssistantMessage (causing doubled content in the next API request).
|
||||
*
|
||||
* Strategy: mock getOpenAIClient + adaptOpenAIStreamToAnthropic so we can
|
||||
* feed pre-built Anthropic events directly into queryModelOpenAI and inspect
|
||||
* what it emits — without any real HTTP calls.
|
||||
*/
|
||||
import { describe, expect, test, mock, beforeEach, afterEach } from 'bun:test'
|
||||
import type { BetaRawMessageStreamEvent } from '@anthropic-ai/sdk/resources/beta/messages/messages.mjs'
|
||||
import type { AssistantMessage, StreamEvent } from '../../../../types/message.js'
|
||||
|
||||
// ─── helpers ─────────────────────────────────────────────────────────────────
|
||||
|
||||
/** Build a minimal message_start event */
|
||||
function makeMessageStart(overrides: Record<string, any> = {}): BetaRawMessageStreamEvent {
|
||||
return {
|
||||
type: 'message_start',
|
||||
message: {
|
||||
id: 'msg_test',
|
||||
type: 'message',
|
||||
role: 'assistant',
|
||||
content: [],
|
||||
model: 'test-model',
|
||||
stop_reason: null,
|
||||
stop_sequence: null,
|
||||
usage: { input_tokens: 0, output_tokens: 0, cache_creation_input_tokens: 0, cache_read_input_tokens: 0 },
|
||||
...overrides,
|
||||
},
|
||||
} as any
|
||||
}
|
||||
|
||||
/** Build a content_block_start event for the given block type */
|
||||
function makeContentBlockStart(index: number, type: 'text' | 'tool_use' | 'thinking', extra: Record<string, any> = {}): BetaRawMessageStreamEvent {
|
||||
const block =
|
||||
type === 'text'
|
||||
? { type: 'text', text: '' }
|
||||
: type === 'tool_use'
|
||||
? { type: 'tool_use', id: 'toolu_test', name: 'bash', input: {} }
|
||||
: { type: 'thinking', thinking: '', signature: '' }
|
||||
return { type: 'content_block_start', index, content_block: { ...block, ...extra } } as any
|
||||
}
|
||||
|
||||
/** Build a text_delta content_block_delta event */
|
||||
function makeTextDelta(index: number, text: string): BetaRawMessageStreamEvent {
|
||||
return { type: 'content_block_delta', index, delta: { type: 'text_delta', text } } as any
|
||||
}
|
||||
|
||||
/** Build an input_json_delta content_block_delta event */
|
||||
function makeInputJsonDelta(index: number, json: string): BetaRawMessageStreamEvent {
|
||||
return { type: 'content_block_delta', index, delta: { type: 'input_json_delta', partial_json: json } } as any
|
||||
}
|
||||
|
||||
/** Build a thinking_delta content_block_delta event */
|
||||
function makeThinkingDelta(index: number, thinking: string): BetaRawMessageStreamEvent {
|
||||
return { type: 'content_block_delta', index, delta: { type: 'thinking_delta', thinking } } as any
|
||||
}
|
||||
|
||||
/** Build a content_block_stop event */
|
||||
function makeContentBlockStop(index: number): BetaRawMessageStreamEvent {
|
||||
return { type: 'content_block_stop', index } as any
|
||||
}
|
||||
|
||||
/** Build a message_delta event with stop_reason and output_tokens */
|
||||
function makeMessageDelta(stopReason: string, outputTokens: number): BetaRawMessageStreamEvent {
|
||||
return {
|
||||
type: 'message_delta',
|
||||
delta: { stop_reason: stopReason, stop_sequence: null },
|
||||
usage: { output_tokens: outputTokens },
|
||||
} as any
|
||||
}
|
||||
|
||||
/** Build a message_stop event */
|
||||
function makeMessageStop(): BetaRawMessageStreamEvent {
|
||||
return { type: 'message_stop' } as any
|
||||
}
|
||||
|
||||
/** Async generator from a fixed array of events */
|
||||
async function* eventStream(events: BetaRawMessageStreamEvent[]) {
|
||||
for (const e of events) yield e
|
||||
}
|
||||
|
||||
/** Collect all outputs from queryModelOpenAI into typed buckets */
|
||||
async function runQueryModel(
|
||||
events: BetaRawMessageStreamEvent[],
|
||||
envOverrides: Record<string, string | undefined> = {},
|
||||
) {
|
||||
// Wire events into the mocked stream adapter
|
||||
_nextEvents = events
|
||||
// Save + apply env overrides
|
||||
const saved: Record<string, string | undefined> = {}
|
||||
for (const [k, v] of Object.entries(envOverrides)) {
|
||||
saved[k] = process.env[k]
|
||||
if (v === undefined) delete process.env[k]
|
||||
else process.env[k] = v
|
||||
}
|
||||
|
||||
try {
|
||||
// We inline mock.module inside the try block.
|
||||
// Bun resolves mock.module at the call site synchronously (hoisted),
|
||||
// so we register once per test file, then re-import each time.
|
||||
const { queryModelOpenAI } = await import('../index.js')
|
||||
|
||||
const assistantMessages: AssistantMessage[] = []
|
||||
const streamEvents: StreamEvent[] = []
|
||||
const otherOutputs: any[] = []
|
||||
|
||||
const minimalOptions: any = {
|
||||
model: 'test-model',
|
||||
tools: [],
|
||||
agents: [],
|
||||
querySource: 'main_loop',
|
||||
getToolPermissionContext: async () => ({
|
||||
alwaysAllow: [],
|
||||
alwaysDeny: [],
|
||||
needsPermission: [],
|
||||
mode: 'default',
|
||||
isBypassingPermissions: false,
|
||||
}),
|
||||
}
|
||||
|
||||
for await (const item of queryModelOpenAI(
|
||||
[],
|
||||
{ type: 'text', text: '' } as any,
|
||||
[],
|
||||
new AbortController().signal,
|
||||
minimalOptions,
|
||||
)) {
|
||||
if (item.type === 'assistant') {
|
||||
assistantMessages.push(item as AssistantMessage)
|
||||
} else if (item.type === 'stream_event') {
|
||||
streamEvents.push(item as StreamEvent)
|
||||
} else {
|
||||
otherOutputs.push(item)
|
||||
}
|
||||
}
|
||||
|
||||
return { assistantMessages, streamEvents, otherOutputs }
|
||||
} finally {
|
||||
// Restore env
|
||||
for (const [k, v] of Object.entries(saved)) {
|
||||
if (v === undefined) delete process.env[k]
|
||||
else process.env[k] = v
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// ─── mock setup ──────────────────────────────────────────────────────────────
|
||||
|
||||
// We mock at module level. Bun's mock.module replaces the module for the
|
||||
// entire file, so we configure the stream per-test via a shared variable.
|
||||
let _nextEvents: BetaRawMessageStreamEvent[] = []
|
||||
|
||||
/** Captured arguments from the last chat.completions.create() call */
|
||||
let _lastCreateArgs: Record<string, any> | null = null
|
||||
|
||||
mock.module('../client.js', () => ({
|
||||
getOpenAIClient: () => ({
|
||||
chat: {
|
||||
completions: {
|
||||
create: async (args: Record<string, any>) => {
|
||||
_lastCreateArgs = args
|
||||
return { [Symbol.asyncIterator]: async function* () {} }
|
||||
},
|
||||
},
|
||||
},
|
||||
}),
|
||||
}))
|
||||
|
||||
mock.module('../streamAdapter.js', () => ({
|
||||
adaptOpenAIStreamToAnthropic: (_stream: any, _model: string) => eventStream(_nextEvents),
|
||||
}))
|
||||
|
||||
mock.module('../modelMapping.js', () => ({
|
||||
resolveOpenAIModel: (m: string) => m,
|
||||
}))
|
||||
|
||||
mock.module('../convertMessages.js', () => ({
|
||||
anthropicMessagesToOpenAI: () => [],
|
||||
}))
|
||||
|
||||
mock.module('../convertTools.js', () => ({
|
||||
anthropicToolsToOpenAI: () => [],
|
||||
anthropicToolChoiceToOpenAI: () => undefined,
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/context.js', () => ({
|
||||
MODEL_CONTEXT_WINDOW_DEFAULT: 200_000,
|
||||
COMPACT_MAX_OUTPUT_TOKENS: 20_000,
|
||||
CAPPED_DEFAULT_MAX_TOKENS: 8_000,
|
||||
ESCALATED_MAX_TOKENS: 64_000,
|
||||
is1mContextDisabled: () => false,
|
||||
has1mContext: () => false,
|
||||
modelSupports1M: () => false,
|
||||
getModelMaxOutputTokens: () => ({ upperLimit: 8192, default: 8192 }),
|
||||
getContextWindowForModel: () => 200_000,
|
||||
getSonnet1mExpTreatmentEnabled: () => false,
|
||||
calculateContextPercentages: () => ({ usedPercent: 0, remainingPercent: 100 }),
|
||||
getMaxThinkingTokensForModel: () => 0,
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/messages.js', () => ({
|
||||
normalizeMessagesForAPI: (msgs: any) => msgs,
|
||||
normalizeContentFromAPI: (blocks: any[]) => blocks,
|
||||
createAssistantAPIErrorMessage: (opts: any) => ({
|
||||
type: 'assistant',
|
||||
message: { content: [{ type: 'text', text: opts.content }], apiError: opts.apiError },
|
||||
uuid: 'error-uuid',
|
||||
timestamp: new Date().toISOString(),
|
||||
}),
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/api.js', () => ({
|
||||
toolToAPISchema: async (t: any) => t,
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/toolSearch.js', () => ({
|
||||
isToolSearchEnabled: async () => false,
|
||||
extractDiscoveredToolNames: () => new Set(),
|
||||
}))
|
||||
|
||||
mock.module('../../../../tools/ToolSearchTool/prompt.js', () => ({
|
||||
isDeferredTool: () => false,
|
||||
TOOL_SEARCH_TOOL_NAME: '__tool_search__',
|
||||
}))
|
||||
|
||||
mock.module('../../../../cost-tracker.js', () => ({
|
||||
addToTotalSessionCost: () => {},
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/modelCost.js', () => ({
|
||||
COST_TIER_3_15: {},
|
||||
COST_TIER_15_75: {},
|
||||
COST_TIER_5_25: {},
|
||||
COST_TIER_30_150: {},
|
||||
COST_HAIKU_35: {},
|
||||
COST_HAIKU_45: {},
|
||||
getOpus46CostTier: () => ({}),
|
||||
MODEL_COSTS: {},
|
||||
getModelCosts: () => ({}),
|
||||
calculateUSDCost: () => 0,
|
||||
calculateCostFromTokens: () => 0,
|
||||
formatModelPricing: () => '',
|
||||
getModelPricingString: () => undefined,
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/debug.js', () => ({
|
||||
logForDebugging: () => {},
|
||||
logAntError: () => {},
|
||||
isDebugMode: () => false,
|
||||
isDebugToStdErr: () => false,
|
||||
getDebugFilePath: () => null,
|
||||
getDebugLogPath: () => '',
|
||||
getDebugFilter: () => null,
|
||||
getMinDebugLogLevel: () => 'debug',
|
||||
enableDebugLogging: () => false,
|
||||
setHasFormattedOutput: () => {},
|
||||
getHasFormattedOutput: () => false,
|
||||
flushDebugLogs: async () => {},
|
||||
}))
|
||||
|
||||
// ─── tests ───────────────────────────────────────────────────────────────────
|
||||
|
||||
describe('queryModelOpenAI — stop_reason propagation', () => {
|
||||
test('assembled AssistantMessage has stop_reason end_turn (not null)', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'Hello'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 10),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
expect(assistantMessages[0]!.message.stop_reason).toBe('end_turn')
|
||||
})
|
||||
|
||||
test('assembled AssistantMessage has stop_reason tool_use', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'tool_use'),
|
||||
makeInputJsonDelta(0, '{"cmd":"ls"}'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('tool_use', 20),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
expect(assistantMessages[0]!.message.stop_reason).toBe('tool_use')
|
||||
})
|
||||
|
||||
test('assembled AssistantMessage has stop_reason max_tokens', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'truncated'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('max_tokens', 8192),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
// Two assistant-typed items: the content message + the max_output_tokens error signal.
|
||||
// The error signal is emitted as a synthetic assistant message by createAssistantAPIErrorMessage.
|
||||
expect(assistantMessages).toHaveLength(2)
|
||||
const contentMsg = assistantMessages[0]!
|
||||
expect(contentMsg.message.stop_reason).toBe('max_tokens')
|
||||
// Second item is the error signal (has apiError set)
|
||||
const errorMsg = assistantMessages[1]!.message as any
|
||||
expect(errorMsg.apiError).toBe('max_output_tokens')
|
||||
})
|
||||
|
||||
test('stop_reason is null when no message_delta was received (safety fallback path)', async () => {
|
||||
// Stream ends without message_stop — triggers the safety fallback branch.
|
||||
// stop_reason stays null since no message_delta was ever seen.
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'partial'),
|
||||
makeContentBlockStop(0),
|
||||
// No message_delta / message_stop
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
// Safety fallback should yield the partial content
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
expect(assistantMessages[0]!.message.stop_reason).toBeNull()
|
||||
})
|
||||
})
|
||||
|
||||
describe('queryModelOpenAI — usage accumulation', () => {
|
||||
test('usage in assembled message reflects all four fields from message_delta', async () => {
|
||||
// message_start has all fields=0 (trailing-chunk pattern: usage not yet available).
|
||||
// message_delta carries the real values after stream ends.
|
||||
// The spread in the message_delta handler must override all zeros from message_start,
|
||||
// including cache_read_input_tokens which was previously missing from message_delta.
|
||||
_nextEvents = [
|
||||
makeMessageStart({ usage: { input_tokens: 0, output_tokens: 0, cache_creation_input_tokens: 0, cache_read_input_tokens: 0 } }),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'response'),
|
||||
makeContentBlockStop(0),
|
||||
// message_delta carries all four Anthropic usage fields (as emitted by the fixed streamAdapter)
|
||||
{
|
||||
type: 'message_delta',
|
||||
delta: { stop_reason: 'end_turn', stop_sequence: null },
|
||||
usage: { input_tokens: 30011, output_tokens: 190, cache_read_input_tokens: 19904, cache_creation_input_tokens: 0 },
|
||||
} as any,
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
const usage = assistantMessages[0]!.message.usage as any
|
||||
expect(usage.input_tokens).toBe(30011)
|
||||
expect(usage.output_tokens).toBe(190)
|
||||
// cache_read_input_tokens from message_delta overrides the 0 from message_start
|
||||
expect(usage.cache_read_input_tokens).toBe(19904)
|
||||
expect(usage.cache_creation_input_tokens).toBe(0)
|
||||
})
|
||||
|
||||
test('usage is zero when no usage events arrive (prevents false autocompact)', async () => {
|
||||
// If usage stays 0, tokenCountWithEstimation will undercount — so at least
|
||||
// verify the field exists and is numeric (to detect regressions).
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'hi'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 0),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
const usage = assistantMessages[0]!.message.usage as any
|
||||
expect(typeof usage.input_tokens).toBe('number')
|
||||
expect(typeof usage.output_tokens).toBe('number')
|
||||
})
|
||||
})
|
||||
|
||||
describe('queryModelOpenAI — no duplicate AssistantMessage (partialMessage reset)', () => {
|
||||
test('yields exactly one AssistantMessage per message_stop when content is present', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'only once'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 5),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
// Before the fix, partialMessage was not reset to null, so the safety
|
||||
// fallback at the end of the loop would yield a second message with the
|
||||
// same message.id — causing mergeAssistantMessages to concatenate content.
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
})
|
||||
|
||||
test('thinking + text response yields exactly one AssistantMessage', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'thinking'),
|
||||
makeThinkingDelta(0, 'let me think'),
|
||||
makeContentBlockStop(0),
|
||||
makeContentBlockStart(1, 'text'),
|
||||
makeTextDelta(1, 'answer'),
|
||||
makeContentBlockStop(1),
|
||||
makeMessageDelta('end_turn', 30),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
})
|
||||
|
||||
test('safety fallback path still yields message when stream ends without message_stop', async () => {
|
||||
// Simulates a stream that cuts off without the normal termination sequence.
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'abrupt end'),
|
||||
// No content_block_stop, no message_delta, no message_stop
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
})
|
||||
})
|
||||
|
||||
describe('queryModelOpenAI — stream_events forwarded', () => {
|
||||
test('every adapted event is also yielded as stream_event for real-time display', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'hello'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 5),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { streamEvents } = await runQueryModel(_nextEvents)
|
||||
|
||||
const eventTypes = streamEvents.map(e => (e as any).event?.type)
|
||||
expect(eventTypes).toContain('message_start')
|
||||
expect(eventTypes).toContain('content_block_start')
|
||||
expect(eventTypes).toContain('content_block_delta')
|
||||
expect(eventTypes).toContain('content_block_stop')
|
||||
expect(eventTypes).toContain('message_delta')
|
||||
expect(eventTypes).toContain('message_stop')
|
||||
})
|
||||
})
|
||||
|
||||
describe('queryModelOpenAI — max_tokens forwarded to request', () => {
|
||||
test('buildOpenAIRequestBody includes max_tokens in the request payload', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'hi'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 5),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
await runQueryModel(_nextEvents)
|
||||
|
||||
expect(_lastCreateArgs).not.toBeNull()
|
||||
expect(_lastCreateArgs!.max_tokens).toBe(8192)
|
||||
})
|
||||
})
|
||||
@@ -1,559 +0,0 @@
|
||||
/**
|
||||
* Tests for queryModelOpenAI in index.ts.
|
||||
*
|
||||
* Focused on the two bugs fixed:
|
||||
* 1. stop_reason was always null in the assembled AssistantMessage because
|
||||
* partialMessage (from message_start) has stop_reason: null, and the
|
||||
* stop_reason captured from message_delta was never applied.
|
||||
* 2. partialMessage was not reset to null after message_stop, so the safety
|
||||
* fallback at the end of the loop would yield a second identical
|
||||
* AssistantMessage (causing doubled content in the next API request).
|
||||
*
|
||||
* Strategy: mock getOpenAIClient + adaptOpenAIStreamToAnthropic so we can
|
||||
* feed pre-built Anthropic events directly into queryModelOpenAI and inspect
|
||||
* what it emits — without any real HTTP calls.
|
||||
*/
|
||||
import { describe, expect, test, mock, beforeEach, afterEach } from 'bun:test'
|
||||
import type { BetaRawMessageStreamEvent } from '@anthropic-ai/sdk/resources/beta/messages/messages.mjs'
|
||||
import type { AssistantMessage, StreamEvent } from '../../../../types/message.js'
|
||||
|
||||
// ─── helpers ─────────────────────────────────────────────────────────────────
|
||||
|
||||
/** Build a minimal message_start event */
|
||||
function makeMessageStart(overrides: Record<string, any> = {}): BetaRawMessageStreamEvent {
|
||||
return {
|
||||
type: 'message_start',
|
||||
message: {
|
||||
id: 'msg_test',
|
||||
type: 'message',
|
||||
role: 'assistant',
|
||||
content: [],
|
||||
model: 'test-model',
|
||||
stop_reason: null,
|
||||
stop_sequence: null,
|
||||
usage: { input_tokens: 0, output_tokens: 0, cache_creation_input_tokens: 0, cache_read_input_tokens: 0 },
|
||||
...overrides,
|
||||
},
|
||||
} as any
|
||||
}
|
||||
|
||||
/** Build a content_block_start event for the given block type */
|
||||
function makeContentBlockStart(index: number, type: 'text' | 'tool_use' | 'thinking', extra: Record<string, any> = {}): BetaRawMessageStreamEvent {
|
||||
const block =
|
||||
type === 'text'
|
||||
? { type: 'text', text: '' }
|
||||
: type === 'tool_use'
|
||||
? { type: 'tool_use', id: 'toolu_test', name: 'bash', input: {} }
|
||||
: { type: 'thinking', thinking: '', signature: '' }
|
||||
return { type: 'content_block_start', index, content_block: { ...block, ...extra } } as any
|
||||
}
|
||||
|
||||
/** Build a text_delta content_block_delta event */
|
||||
function makeTextDelta(index: number, text: string): BetaRawMessageStreamEvent {
|
||||
return { type: 'content_block_delta', index, delta: { type: 'text_delta', text } } as any
|
||||
}
|
||||
|
||||
/** Build an input_json_delta content_block_delta event */
|
||||
function makeInputJsonDelta(index: number, json: string): BetaRawMessageStreamEvent {
|
||||
return { type: 'content_block_delta', index, delta: { type: 'input_json_delta', partial_json: json } } as any
|
||||
}
|
||||
|
||||
/** Build a thinking_delta content_block_delta event */
|
||||
function makeThinkingDelta(index: number, thinking: string): BetaRawMessageStreamEvent {
|
||||
return { type: 'content_block_delta', index, delta: { type: 'thinking_delta', thinking } } as any
|
||||
}
|
||||
|
||||
/** Build a content_block_stop event */
|
||||
function makeContentBlockStop(index: number): BetaRawMessageStreamEvent {
|
||||
return { type: 'content_block_stop', index } as any
|
||||
}
|
||||
|
||||
/** Build a message_delta event with stop_reason and output_tokens */
|
||||
function makeMessageDelta(stopReason: string, outputTokens: number): BetaRawMessageStreamEvent {
|
||||
return {
|
||||
type: 'message_delta',
|
||||
delta: { stop_reason: stopReason, stop_sequence: null },
|
||||
usage: { output_tokens: outputTokens },
|
||||
} as any
|
||||
}
|
||||
|
||||
/** Build a message_stop event */
|
||||
function makeMessageStop(): BetaRawMessageStreamEvent {
|
||||
return { type: 'message_stop' } as any
|
||||
}
|
||||
|
||||
/** Async generator from a fixed array of events */
|
||||
async function* eventStream(events: BetaRawMessageStreamEvent[]) {
|
||||
for (const e of events) yield e
|
||||
}
|
||||
|
||||
/** Collect all outputs from queryModelOpenAI into typed buckets */
|
||||
async function runQueryModel(
|
||||
events: BetaRawMessageStreamEvent[],
|
||||
envOverrides: Record<string, string | undefined> = {},
|
||||
) {
|
||||
// Wire events into the mocked stream adapter
|
||||
_nextEvents = events
|
||||
// Save + apply env overrides
|
||||
const saved: Record<string, string | undefined> = {}
|
||||
for (const [k, v] of Object.entries(envOverrides)) {
|
||||
saved[k] = process.env[k]
|
||||
if (v === undefined) delete process.env[k]
|
||||
else process.env[k] = v
|
||||
}
|
||||
|
||||
try {
|
||||
// We inline mock.module inside the try block.
|
||||
// Bun resolves mock.module at the call site synchronously (hoisted),
|
||||
// so we register once per test file, then re-import each time.
|
||||
const { queryModelOpenAI } = await import('../index.js')
|
||||
|
||||
const assistantMessages: AssistantMessage[] = []
|
||||
const streamEvents: StreamEvent[] = []
|
||||
const otherOutputs: any[] = []
|
||||
|
||||
const minimalOptions: any = {
|
||||
model: 'test-model',
|
||||
tools: [],
|
||||
agents: [],
|
||||
querySource: 'main_loop',
|
||||
getToolPermissionContext: async () => ({
|
||||
alwaysAllow: [],
|
||||
alwaysDeny: [],
|
||||
needsPermission: [],
|
||||
mode: 'default',
|
||||
isBypassingPermissions: false,
|
||||
}),
|
||||
}
|
||||
|
||||
for await (const item of queryModelOpenAI(
|
||||
[],
|
||||
{ type: 'text', text: '' } as any,
|
||||
[],
|
||||
new AbortController().signal,
|
||||
minimalOptions,
|
||||
)) {
|
||||
if (item.type === 'assistant') {
|
||||
assistantMessages.push(item as AssistantMessage)
|
||||
} else if (item.type === 'stream_event') {
|
||||
streamEvents.push(item as StreamEvent)
|
||||
} else {
|
||||
otherOutputs.push(item)
|
||||
}
|
||||
}
|
||||
|
||||
return { assistantMessages, streamEvents, otherOutputs }
|
||||
} finally {
|
||||
// Restore env
|
||||
for (const [k, v] of Object.entries(saved)) {
|
||||
if (v === undefined) delete process.env[k]
|
||||
else process.env[k] = v
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// ─── mock setup ──────────────────────────────────────────────────────────────
|
||||
|
||||
// We mock at module level. Bun's mock.module replaces the module for the
|
||||
// entire file, so we configure the stream per-test via a shared variable.
|
||||
let _nextEvents: BetaRawMessageStreamEvent[] = []
|
||||
|
||||
/** Captured arguments from the last chat.completions.create() call */
|
||||
let _lastCreateArgs: Record<string, any> | null = null
|
||||
|
||||
mock.module('../client.js', () => ({
|
||||
getOpenAIClient: () => ({
|
||||
chat: {
|
||||
completions: {
|
||||
create: async (args: Record<string, any>) => {
|
||||
_lastCreateArgs = args
|
||||
return { [Symbol.asyncIterator]: async function* () {} }
|
||||
},
|
||||
},
|
||||
},
|
||||
}),
|
||||
}))
|
||||
|
||||
mock.module('../streamAdapter.js', () => ({
|
||||
adaptOpenAIStreamToAnthropic: (_stream: any, _model: string) => eventStream(_nextEvents),
|
||||
}))
|
||||
|
||||
mock.module('../modelMapping.js', () => ({
|
||||
resolveOpenAIModel: (m: string) => m,
|
||||
}))
|
||||
|
||||
mock.module('../convertMessages.js', () => ({
|
||||
anthropicMessagesToOpenAI: () => [],
|
||||
}))
|
||||
|
||||
mock.module('../convertTools.js', () => ({
|
||||
anthropicToolsToOpenAI: () => [],
|
||||
anthropicToolChoiceToOpenAI: () => undefined,
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/context.js', () => ({
|
||||
getModelMaxOutputTokens: () => ({ upperLimit: 8192, default: 8192 }),
|
||||
getContextWindowForModel: () => 200_000,
|
||||
modelSupports1M: () => false,
|
||||
has1mContext: () => false,
|
||||
is1mContextDisabled: () => false,
|
||||
getSonnet1mExpTreatmentEnabled: () => false,
|
||||
MODEL_CONTEXT_WINDOW_DEFAULT: 200_000,
|
||||
COMPACT_MAX_OUTPUT_TOKENS: 20_000,
|
||||
CAPPED_DEFAULT_MAX_TOKENS: 8_000,
|
||||
ESCALATED_MAX_TOKENS: 64_000,
|
||||
calculateContextPercentages: () => ({ used: null, remaining: null }),
|
||||
getMaxThinkingTokensForModel: () => 8191,
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/messages.js', () => ({
|
||||
normalizeMessagesForAPI: (msgs: any) => msgs,
|
||||
normalizeContentFromAPI: (blocks: any[]) => blocks,
|
||||
createAssistantAPIErrorMessage: (opts: any) => ({
|
||||
type: 'assistant',
|
||||
message: { content: [{ type: 'text', text: opts.content }], apiError: opts.apiError },
|
||||
uuid: 'error-uuid',
|
||||
timestamp: new Date().toISOString(),
|
||||
}),
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/api.js', () => ({
|
||||
toolToAPISchema: async (t: any) => t,
|
||||
}))
|
||||
|
||||
mock.module('../../../../Tool.js', () => ({
|
||||
getEmptyToolPermissionContext: () => ({
|
||||
alwaysAllow: [],
|
||||
alwaysDeny: [],
|
||||
needsPermission: [],
|
||||
mode: 'default',
|
||||
isBypassingPermissions: false,
|
||||
}),
|
||||
toolMatchesName: () => false,
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/envUtils.js', () => ({
|
||||
isEnvTruthy: (v: string | undefined) => v === '1' || v === 'true',
|
||||
isEnvDefinedFalsy: (v: string | undefined) => v === '0' || v === 'false' || v === 'no' || v === 'off',
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/toolSearch.js', () => ({
|
||||
isToolSearchEnabled: async () => false,
|
||||
extractDiscoveredToolNames: () => new Set(),
|
||||
}))
|
||||
|
||||
mock.module('../../../../tools/ToolSearchTool/prompt.js', () => ({
|
||||
isDeferredTool: () => false,
|
||||
TOOL_SEARCH_TOOL_NAME: '__tool_search__',
|
||||
}))
|
||||
|
||||
mock.module('../../../../cost-tracker.js', () => ({
|
||||
addToTotalSessionCost: () => {},
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/modelCost.js', () => ({
|
||||
calculateUSDCost: () => 0,
|
||||
}))
|
||||
|
||||
mock.module('../../../../utils/debug.js', () => ({
|
||||
logForDebugging: () => {},
|
||||
}))
|
||||
|
||||
// ─── tests ───────────────────────────────────────────────────────────────────
|
||||
|
||||
describe('queryModelOpenAI — stop_reason propagation', () => {
|
||||
test('assembled AssistantMessage has stop_reason end_turn (not null)', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'Hello'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 10),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
expect(assistantMessages[0]!.message.stop_reason).toBe('end_turn')
|
||||
})
|
||||
|
||||
test('assembled AssistantMessage has stop_reason tool_use', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'tool_use'),
|
||||
makeInputJsonDelta(0, '{"cmd":"ls"}'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('tool_use', 20),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
expect(assistantMessages[0]!.message.stop_reason).toBe('tool_use')
|
||||
})
|
||||
|
||||
test('assembled AssistantMessage has stop_reason max_tokens', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'truncated'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('max_tokens', 8192),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
// Two assistant-typed items: the content message + the max_output_tokens error signal.
|
||||
// The error signal is emitted as a synthetic assistant message by createAssistantAPIErrorMessage.
|
||||
expect(assistantMessages).toHaveLength(2)
|
||||
const contentMsg = assistantMessages[0]!
|
||||
expect(contentMsg.message.stop_reason).toBe('max_tokens')
|
||||
// Second item is the error signal (has apiError set)
|
||||
const errorMsg = assistantMessages[1]!.message as any
|
||||
expect(errorMsg.apiError).toBe('max_output_tokens')
|
||||
})
|
||||
|
||||
test('stop_reason is null when no message_delta was received (safety fallback path)', async () => {
|
||||
// Stream ends without message_stop — triggers the safety fallback branch.
|
||||
// stop_reason stays null since no message_delta was ever seen.
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'partial'),
|
||||
makeContentBlockStop(0),
|
||||
// No message_delta / message_stop
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
// Safety fallback should yield the partial content
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
expect(assistantMessages[0]!.message.stop_reason).toBeNull()
|
||||
})
|
||||
})
|
||||
|
||||
describe('queryModelOpenAI — usage accumulation', () => {
|
||||
test('usage in assembled message reflects all four fields from message_delta', async () => {
|
||||
// message_start has all fields=0 (trailing-chunk pattern: usage not yet available).
|
||||
// message_delta carries the real values after stream ends.
|
||||
// The spread in the message_delta handler must override all zeros from message_start,
|
||||
// including cache_read_input_tokens which was previously missing from message_delta.
|
||||
_nextEvents = [
|
||||
makeMessageStart({ usage: { input_tokens: 0, output_tokens: 0, cache_creation_input_tokens: 0, cache_read_input_tokens: 0 } }),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'response'),
|
||||
makeContentBlockStop(0),
|
||||
// message_delta carries all four Anthropic usage fields (as emitted by the fixed streamAdapter)
|
||||
{
|
||||
type: 'message_delta',
|
||||
delta: { stop_reason: 'end_turn', stop_sequence: null },
|
||||
usage: { input_tokens: 30011, output_tokens: 190, cache_read_input_tokens: 19904, cache_creation_input_tokens: 0 },
|
||||
} as any,
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
const usage = assistantMessages[0]!.message.usage as any
|
||||
expect(usage.input_tokens).toBe(30011)
|
||||
expect(usage.output_tokens).toBe(190)
|
||||
// cache_read_input_tokens from message_delta overrides the 0 from message_start
|
||||
expect(usage.cache_read_input_tokens).toBe(19904)
|
||||
expect(usage.cache_creation_input_tokens).toBe(0)
|
||||
})
|
||||
|
||||
test('usage is zero when no usage events arrive (prevents false autocompact)', async () => {
|
||||
// If usage stays 0, tokenCountWithEstimation will undercount — so at least
|
||||
// verify the field exists and is numeric (to detect regressions).
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'hi'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 0),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
const usage = assistantMessages[0]!.message.usage as any
|
||||
expect(typeof usage.input_tokens).toBe('number')
|
||||
expect(typeof usage.output_tokens).toBe('number')
|
||||
})
|
||||
})
|
||||
|
||||
describe('queryModelOpenAI — no duplicate AssistantMessage (partialMessage reset)', () => {
|
||||
test('yields exactly one AssistantMessage per message_stop when content is present', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'only once'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 5),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
// Before the fix, partialMessage was not reset to null, so the safety
|
||||
// fallback at the end of the loop would yield a second message with the
|
||||
// same message.id — causing mergeAssistantMessages to concatenate content.
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
})
|
||||
|
||||
test('thinking + text response yields exactly one AssistantMessage', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'thinking'),
|
||||
makeThinkingDelta(0, 'let me think'),
|
||||
makeContentBlockStop(0),
|
||||
makeContentBlockStart(1, 'text'),
|
||||
makeTextDelta(1, 'answer'),
|
||||
makeContentBlockStop(1),
|
||||
makeMessageDelta('end_turn', 30),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
})
|
||||
|
||||
test('safety fallback path still yields message when stream ends without message_stop', async () => {
|
||||
// Simulates a stream that cuts off without the normal termination sequence.
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'abrupt end'),
|
||||
// No content_block_stop, no message_delta, no message_stop
|
||||
]
|
||||
|
||||
const { assistantMessages } = await runQueryModel(_nextEvents)
|
||||
|
||||
expect(assistantMessages).toHaveLength(1)
|
||||
})
|
||||
})
|
||||
|
||||
describe('queryModelOpenAI — stream_events forwarded', () => {
|
||||
test('every adapted event is also yielded as stream_event for real-time display', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'hello'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 5),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
const { streamEvents } = await runQueryModel(_nextEvents)
|
||||
|
||||
const eventTypes = streamEvents.map(e => (e as any).event?.type)
|
||||
expect(eventTypes).toContain('message_start')
|
||||
expect(eventTypes).toContain('content_block_start')
|
||||
expect(eventTypes).toContain('content_block_delta')
|
||||
expect(eventTypes).toContain('content_block_stop')
|
||||
expect(eventTypes).toContain('message_delta')
|
||||
expect(eventTypes).toContain('message_stop')
|
||||
})
|
||||
})
|
||||
|
||||
describe('queryModelOpenAI — max_tokens forwarded to request', () => {
|
||||
test('buildOpenAIRequestBody includes max_tokens in the request payload', async () => {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'hi'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 5),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
await runQueryModel(_nextEvents)
|
||||
|
||||
expect(_lastCreateArgs).not.toBeNull()
|
||||
expect(_lastCreateArgs!.max_tokens).toBe(8192)
|
||||
})
|
||||
|
||||
test('OPENAI_MAX_TOKENS env var overrides max_tokens', async () => {
|
||||
const original = process.env.OPENAI_MAX_TOKENS
|
||||
process.env.OPENAI_MAX_TOKENS = '4096'
|
||||
try {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'hi'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 5),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
await runQueryModel(_nextEvents)
|
||||
|
||||
expect(_lastCreateArgs).not.toBeNull()
|
||||
expect(_lastCreateArgs!.max_tokens).toBe(4096)
|
||||
} finally {
|
||||
if (original === undefined) {
|
||||
delete process.env.OPENAI_MAX_TOKENS
|
||||
} else {
|
||||
process.env.OPENAI_MAX_TOKENS = original
|
||||
}
|
||||
}
|
||||
})
|
||||
|
||||
test('CLAUDE_CODE_MAX_OUTPUT_TOKENS env var overrides max_tokens', async () => {
|
||||
const original = process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS
|
||||
process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS = '2048'
|
||||
try {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'hi'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 5),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
await runQueryModel(_nextEvents)
|
||||
|
||||
expect(_lastCreateArgs).not.toBeNull()
|
||||
expect(_lastCreateArgs!.max_tokens).toBe(2048)
|
||||
} finally {
|
||||
if (original === undefined) {
|
||||
delete process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS
|
||||
} else {
|
||||
process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS = original
|
||||
}
|
||||
}
|
||||
})
|
||||
|
||||
test('OPENAI_MAX_TOKENS takes priority over CLAUDE_CODE_MAX_OUTPUT_TOKENS', async () => {
|
||||
const origOpenai = process.env.OPENAI_MAX_TOKENS
|
||||
const origClaude = process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS
|
||||
process.env.OPENAI_MAX_TOKENS = '4096'
|
||||
process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS = '2048'
|
||||
try {
|
||||
_nextEvents = [
|
||||
makeMessageStart(),
|
||||
makeContentBlockStart(0, 'text'),
|
||||
makeTextDelta(0, 'hi'),
|
||||
makeContentBlockStop(0),
|
||||
makeMessageDelta('end_turn', 5),
|
||||
makeMessageStop(),
|
||||
]
|
||||
|
||||
await runQueryModel(_nextEvents)
|
||||
|
||||
expect(_lastCreateArgs).not.toBeNull()
|
||||
expect(_lastCreateArgs!.max_tokens).toBe(4096)
|
||||
} finally {
|
||||
if (origOpenai === undefined) delete process.env.OPENAI_MAX_TOKENS
|
||||
else process.env.OPENAI_MAX_TOKENS = origOpenai
|
||||
if (origClaude === undefined) delete process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS
|
||||
else process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS = origClaude
|
||||
}
|
||||
})
|
||||
})
|
||||
@@ -1,679 +0,0 @@
|
||||
import { describe, expect, test } from 'bun:test'
|
||||
import type { ChatCompletionChunk } from 'openai/resources/chat/completions/completions.mjs'
|
||||
import { join, dirname } from 'path'
|
||||
import { fileURLToPath } from 'url'
|
||||
import { readFileSync, writeFileSync, mkdirSync } from 'fs'
|
||||
import { tmpdir } from 'os'
|
||||
|
||||
// Guard against mock pollution from queryModelOpenAI.test.ts which replaces
|
||||
// ../streamAdapter.js process-wide via mock.module (bun has no un-mock API).
|
||||
// We copy the source to a unique temp path so the import bypasses bun's
|
||||
// module mock cache completely.
|
||||
const _testDir = dirname(fileURLToPath(import.meta.url))
|
||||
const _realSource = readFileSync(join(_testDir, '..', 'streamAdapter.ts'), 'utf-8')
|
||||
const _tempDir = join(tmpdir(), `stream-adapter-test-${Date.now()}`)
|
||||
mkdirSync(_tempDir, { recursive: true })
|
||||
const _tempFile = join(_tempDir, 'streamAdapter.ts')
|
||||
writeFileSync(_tempFile, _realSource, 'utf-8')
|
||||
const { adaptOpenAIStreamToAnthropic } = await import(_tempFile)
|
||||
|
||||
/** Helper to create a mock async iterable from chunk array */
|
||||
function mockStream(chunks: ChatCompletionChunk[]): AsyncIterable<ChatCompletionChunk> {
|
||||
return {
|
||||
[Symbol.asyncIterator]() {
|
||||
let i = 0
|
||||
return {
|
||||
async next() {
|
||||
if (i >= chunks.length) return { done: true, value: undefined }
|
||||
return { done: false, value: chunks[i++] }
|
||||
},
|
||||
}
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
/** Create a minimal ChatCompletionChunk */
|
||||
function makeChunk(overrides: Partial<ChatCompletionChunk> & any = {}): ChatCompletionChunk {
|
||||
return {
|
||||
id: 'chatcmpl-test',
|
||||
object: 'chat.completion.chunk',
|
||||
created: 1234567890,
|
||||
model: 'gpt-4o',
|
||||
choices: [],
|
||||
...overrides,
|
||||
} as ChatCompletionChunk
|
||||
}
|
||||
|
||||
/** Collect all emitted Anthropic events from the stream adapter for assertion */
|
||||
async function collectEvents(chunks: ChatCompletionChunk[]) {
|
||||
const realModuleUrl = new URL(
|
||||
`../streamAdapter.js?real=${Date.now()}-${Math.random().toString(36).slice(2)}`,
|
||||
import.meta.url,
|
||||
).href
|
||||
const { adaptOpenAIStreamToAnthropic } = await import(realModuleUrl)
|
||||
const events: any[] = []
|
||||
for await (const event of adaptOpenAIStreamToAnthropic(mockStream(chunks), 'gpt-4o')) {
|
||||
events.push(event)
|
||||
}
|
||||
return events
|
||||
}
|
||||
|
||||
describe('adaptOpenAIStreamToAnthropic', () => {
|
||||
test('emits message_start on first chunk', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { role: 'assistant', content: '' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { content: 'hello' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: {},
|
||||
finish_reason: 'stop',
|
||||
}],
|
||||
usage: { prompt_tokens: 10, completion_tokens: 5, total_tokens: 15 },
|
||||
}),
|
||||
])
|
||||
|
||||
expect(events[0].type).toBe('message_start')
|
||||
expect(events[0].message.role).toBe('assistant')
|
||||
expect(events[0].message.model).toBe('gpt-4o')
|
||||
})
|
||||
|
||||
test('converts text content stream', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'Hello' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: ' world' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const types = events.map(e => e.type)
|
||||
expect(types).toContain('message_start')
|
||||
expect(types).toContain('content_block_start')
|
||||
expect(types.filter(t => t === 'content_block_delta').length).toBe(2)
|
||||
expect(types).toContain('content_block_stop')
|
||||
expect(types).toContain('message_delta')
|
||||
expect(types).toContain('message_stop')
|
||||
|
||||
const textDeltas = events.filter(e => e.type === 'content_block_delta') as any[]
|
||||
expect(textDeltas[0].delta.text).toBe('Hello')
|
||||
expect(textDeltas[1].delta.text).toBe(' world')
|
||||
})
|
||||
|
||||
test('converts tool_calls stream', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: {
|
||||
tool_calls: [{
|
||||
index: 0,
|
||||
id: 'call_abc',
|
||||
type: 'function',
|
||||
function: { name: 'bash', arguments: '' },
|
||||
}],
|
||||
},
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: {
|
||||
tool_calls: [{
|
||||
index: 0,
|
||||
function: { arguments: '{"comm' },
|
||||
}],
|
||||
},
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: {
|
||||
tool_calls: [{
|
||||
index: 0,
|
||||
function: { arguments: 'and":"ls"}' },
|
||||
}],
|
||||
},
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'tool_calls' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const blockStart = events.find(e => e.type === 'content_block_start') as any
|
||||
expect(blockStart.content_block.type).toBe('tool_use')
|
||||
expect(blockStart.content_block.name).toBe('bash')
|
||||
|
||||
const jsonDeltas = events.filter(
|
||||
e => e.type === 'content_block_delta' && e.delta.type === 'input_json_delta',
|
||||
) as any[]
|
||||
const fullArgs = jsonDeltas.map(d => d.delta.partial_json).join('')
|
||||
expect(fullArgs).toBe('{"command":"ls"}')
|
||||
})
|
||||
|
||||
test('maps finish_reason stop to end_turn', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'hi' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.delta.stop_reason).toBe('end_turn')
|
||||
})
|
||||
|
||||
test('forces tool_use stop_reason when tool_calls present but finish_reason is stop', async () => {
|
||||
// Some backends (e.g., certain OpenAI-compatible endpoints) incorrectly
|
||||
// return finish_reason "stop" when they actually made tool calls.
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: {
|
||||
tool_calls: [{ index: 0, id: 'call_1', function: { name: 'bash', arguments: '{"cmd":"ls"}' } }],
|
||||
},
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.delta.stop_reason).toBe('tool_use')
|
||||
})
|
||||
|
||||
test('maps finish_reason tool_calls to tool_use', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: {
|
||||
tool_calls: [{ index: 0, id: 'call_1', function: { name: 'bash', arguments: '{}' } }],
|
||||
},
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'tool_calls' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.delta.stop_reason).toBe('tool_use')
|
||||
})
|
||||
|
||||
test('maps finish_reason length to max_tokens', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'truncated' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'length' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.delta.stop_reason).toBe('max_tokens')
|
||||
})
|
||||
|
||||
test('handles mixed text and tool_calls', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'Thinking...' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: {
|
||||
tool_calls: [{ index: 0, id: 'call_1', function: { name: 'grep', arguments: '{"p":"test"}' } }],
|
||||
},
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'tool_calls' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const blockStarts = events.filter(e => e.type === 'content_block_start') as any[]
|
||||
expect(blockStarts.length).toBe(2)
|
||||
expect(blockStarts[0].content_block.type).toBe('text')
|
||||
expect(blockStarts[1].content_block.type).toBe('tool_use')
|
||||
})
|
||||
})
|
||||
|
||||
describe('thinking support (reasoning_content)', () => {
|
||||
test('converts reasoning_content to thinking block', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { reasoning_content: 'Let me analyze this...' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { reasoning_content: ' step by step.' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
])
|
||||
|
||||
// Should have a thinking content block
|
||||
const blockStart = events.find(e => e.type === 'content_block_start') as any
|
||||
expect(blockStart.content_block.type).toBe('thinking')
|
||||
expect(blockStart.content_block.signature).toBe('')
|
||||
|
||||
// Should have thinking_delta events
|
||||
const thinkingDeltas = events.filter(
|
||||
e => e.type === 'content_block_delta' && e.delta.type === 'thinking_delta',
|
||||
) as any[]
|
||||
expect(thinkingDeltas.length).toBe(2)
|
||||
expect(thinkingDeltas[0].delta.thinking).toBe('Let me analyze this...')
|
||||
expect(thinkingDeltas[1].delta.thinking).toBe(' step by step.')
|
||||
})
|
||||
|
||||
test('converts reasoning then content (DeepSeek-style)', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { reasoning_content: 'Thinking about the answer...' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { content: 'Here is my answer.' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
])
|
||||
|
||||
// Should have two content blocks: thinking + text
|
||||
const blockStarts = events.filter(e => e.type === 'content_block_start') as any[]
|
||||
expect(blockStarts.length).toBe(2)
|
||||
expect(blockStarts[0].content_block.type).toBe('thinking')
|
||||
expect(blockStarts[1].content_block.type).toBe('text')
|
||||
|
||||
// Thinking block should be closed before text block starts
|
||||
const blockStops = events.filter(e => e.type === 'content_block_stop') as any[]
|
||||
expect(blockStops[0].index).toBe(0) // thinking block closed at index 0
|
||||
expect(blockStarts[1].index).toBe(1) // text block starts at index 1
|
||||
|
||||
// Verify text delta
|
||||
const textDelta = events.find(
|
||||
e => e.type === 'content_block_delta' && e.delta.type === 'text_delta',
|
||||
) as any
|
||||
expect(textDelta.delta.text).toBe('Here is my answer.')
|
||||
})
|
||||
|
||||
test('handles reasoning then tool_calls', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { reasoning_content: 'I need to run a command.' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: {
|
||||
tool_calls: [{ index: 0, id: 'call_1', function: { name: 'bash', arguments: '{"c":"ls"}' } }],
|
||||
},
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'tool_calls' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const blockStarts = events.filter(e => e.type === 'content_block_start') as any[]
|
||||
expect(blockStarts.length).toBe(2)
|
||||
expect(blockStarts[0].content_block.type).toBe('thinking')
|
||||
expect(blockStarts[1].content_block.type).toBe('tool_use')
|
||||
})
|
||||
|
||||
test('thinking block index is 0, text block index is 1', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { reasoning_content: 'reason' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { content: 'answer' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const blockStarts = events.filter(e => e.type === 'content_block_start') as any[]
|
||||
expect(blockStarts[0].index).toBe(0)
|
||||
expect(blockStarts[1].index).toBe(1)
|
||||
})
|
||||
})
|
||||
|
||||
describe('prompt caching support', () => {
|
||||
test('maps cached_tokens to cache_read_input_tokens', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: { content: 'hi' },
|
||||
finish_reason: null,
|
||||
}],
|
||||
usage: {
|
||||
prompt_tokens: 1000,
|
||||
completion_tokens: 0,
|
||||
total_tokens: 1000,
|
||||
prompt_tokens_details: { cached_tokens: 800 },
|
||||
} as any,
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
usage: {
|
||||
prompt_tokens: 1000,
|
||||
completion_tokens: 50,
|
||||
total_tokens: 1050,
|
||||
prompt_tokens_details: { cached_tokens: 800 },
|
||||
} as any,
|
||||
}),
|
||||
])
|
||||
|
||||
const msgStart = events.find(e => e.type === 'message_start') as any
|
||||
expect(msgStart.message.usage.cache_read_input_tokens).toBe(800)
|
||||
expect(msgStart.message.usage.input_tokens).toBe(1000)
|
||||
})
|
||||
|
||||
test('defaults cache_read_input_tokens to 0 when no cached_tokens', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'hi' }, finish_reason: null }],
|
||||
usage: { prompt_tokens: 100, completion_tokens: 0, total_tokens: 100 },
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
])
|
||||
|
||||
const msgStart = events.find(e => e.type === 'message_start') as any
|
||||
expect(msgStart.message.usage.cache_read_input_tokens).toBe(0)
|
||||
expect(msgStart.message.usage.cache_creation_input_tokens).toBe(0)
|
||||
})
|
||||
|
||||
test('updates cached_tokens from later chunks', async () => {
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'hi' }, finish_reason: null }],
|
||||
usage: {
|
||||
prompt_tokens: 500,
|
||||
completion_tokens: 0,
|
||||
total_tokens: 500,
|
||||
} as any,
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
usage: {
|
||||
prompt_tokens: 500,
|
||||
completion_tokens: 10,
|
||||
total_tokens: 510,
|
||||
prompt_tokens_details: { cached_tokens: 300 },
|
||||
} as any,
|
||||
}),
|
||||
])
|
||||
|
||||
const msgStart = events.find(e => e.type === 'message_start') as any
|
||||
// First chunk had no cached_tokens, so initially 0
|
||||
// But the message_start usage reflects the first chunk's data
|
||||
expect(msgStart.message.usage.cache_read_input_tokens).toBe(0)
|
||||
expect(msgStart.message.usage.input_tokens).toBe(500)
|
||||
})
|
||||
|
||||
test('captures output_tokens and input_tokens from trailing chunk sent after finish_reason', async () => {
|
||||
// Many OpenAI-compatible endpoints (e.g. DeepSeek) send usage in a separate
|
||||
// final chunk AFTER the finish_reason chunk, with choices: [].
|
||||
// message_delta must carry both input_tokens and output_tokens so that
|
||||
// queryModelOpenAI's spread can override the zeros from message_start — which is
|
||||
// emitted before the trailing chunk and always has input_tokens=0.
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'hello' }, finish_reason: null }],
|
||||
}),
|
||||
// finish_reason chunk — usage not yet available
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
// trailing usage-only chunk (choices: [])
|
||||
makeChunk({
|
||||
choices: [],
|
||||
usage: { prompt_tokens: 123, completion_tokens: 45, total_tokens: 168 },
|
||||
}),
|
||||
])
|
||||
|
||||
// message_start emits on the first chunk before trailing usage arrives
|
||||
const msgStart = events.find(e => e.type === 'message_start') as any
|
||||
expect(msgStart.message.usage.input_tokens).toBe(0)
|
||||
|
||||
// message_delta is emitted after stream loop ends with final real values
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.usage.input_tokens).toBe(123)
|
||||
expect(msgDelta.usage.output_tokens).toBe(45)
|
||||
expect(msgDelta.delta.stop_reason).toBe('end_turn')
|
||||
})
|
||||
|
||||
test('captures input_tokens from trailing chunk (used by tokenCountWithEstimation for autocompact)', async () => {
|
||||
// input_tokens is the dominant term in tokenCountWithEstimation. Without it,
|
||||
// getTokenCountFromUsage returns only output_tokens (~100-700), which is far below
|
||||
// the autocompact threshold (~33k), so compaction never fires.
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'answer' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [],
|
||||
usage: { prompt_tokens: 800, completion_tokens: 200, total_tokens: 1000 },
|
||||
}),
|
||||
])
|
||||
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.usage.input_tokens).toBe(800)
|
||||
expect(msgDelta.usage.output_tokens).toBe(200)
|
||||
})
|
||||
|
||||
test('trailing usage chunk with tool_calls: stop_reason stays tool_use', async () => {
|
||||
// Verifies that deferring message_delta does not break stop_reason mapping
|
||||
// when the model made tool calls and usage arrives in a trailing chunk.
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{
|
||||
index: 0,
|
||||
delta: {
|
||||
tool_calls: [{ index: 0, id: 'call_x', function: { name: 'bash', arguments: '{"cmd":"ls"}' } }],
|
||||
},
|
||||
finish_reason: null,
|
||||
}],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'tool_calls' }],
|
||||
}),
|
||||
// trailing usage-only chunk
|
||||
makeChunk({
|
||||
choices: [],
|
||||
usage: { prompt_tokens: 500, completion_tokens: 30, total_tokens: 530 },
|
||||
}),
|
||||
])
|
||||
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.delta.stop_reason).toBe('tool_use')
|
||||
expect(msgDelta.usage.output_tokens).toBe(30)
|
||||
})
|
||||
|
||||
test('message_delta always comes before message_stop', async () => {
|
||||
// Verifies event ordering is preserved after deferring to post-loop emission.
|
||||
const events = await collectEvents([
|
||||
makeChunk({ choices: [{ index: 0, delta: { content: 'x' }, finish_reason: null }] }),
|
||||
makeChunk({ choices: [{ index: 0, delta: {}, finish_reason: 'stop' }] }),
|
||||
makeChunk({ choices: [], usage: { prompt_tokens: 10, completion_tokens: 5, total_tokens: 15 } }),
|
||||
])
|
||||
|
||||
const types = events.map(e => e.type)
|
||||
const deltaIdx = types.lastIndexOf('message_delta')
|
||||
const stopIdx = types.lastIndexOf('message_stop')
|
||||
expect(deltaIdx).toBeGreaterThanOrEqual(0)
|
||||
expect(stopIdx).toBeGreaterThan(deltaIdx)
|
||||
})
|
||||
|
||||
// ── cache_read_input_tokens in message_delta (the core bug fix) ──────────
|
||||
|
||||
test('message_delta carries cache_read_input_tokens from trailing usage chunk', async () => {
|
||||
// Real-world case: DeepSeek-V3 returns cached_tokens=19904
|
||||
// in a trailing chunk with choices:[]. Previously message_delta only carried
|
||||
// input_tokens and output_tokens, so cache_read_input_tokens stayed 0 after
|
||||
// queryModelOpenAI's spread — even though cachedTokens was captured internally.
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'answer' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
// trailing usage chunk matching the observed server response format
|
||||
makeChunk({
|
||||
choices: [],
|
||||
usage: {
|
||||
prompt_tokens: 30011,
|
||||
completion_tokens: 190,
|
||||
total_tokens: 30201,
|
||||
prompt_tokens_details: { audio_tokens: 0, cached_tokens: 19904 },
|
||||
} as any,
|
||||
}),
|
||||
])
|
||||
|
||||
// message_start is emitted before trailing chunk — cache fields are 0
|
||||
const msgStart = events.find(e => e.type === 'message_start') as any
|
||||
expect(msgStart.message.usage.cache_read_input_tokens).toBe(0)
|
||||
|
||||
// message_delta carries the real values from the trailing chunk
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.usage.input_tokens).toBe(30011)
|
||||
expect(msgDelta.usage.output_tokens).toBe(190)
|
||||
expect(msgDelta.usage.cache_read_input_tokens).toBe(19904)
|
||||
expect(msgDelta.usage.cache_creation_input_tokens).toBe(0)
|
||||
})
|
||||
|
||||
test('cache_read_input_tokens=0 in message_delta when cached_tokens is absent', async () => {
|
||||
// Non-caching requests should still have the field present and zero.
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'hi' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [],
|
||||
usage: { prompt_tokens: 100, completion_tokens: 20, total_tokens: 120 },
|
||||
}),
|
||||
])
|
||||
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.usage.cache_read_input_tokens).toBe(0)
|
||||
expect(msgDelta.usage.cache_creation_input_tokens).toBe(0)
|
||||
})
|
||||
|
||||
test('cache_read_input_tokens=0 in message_delta when cached_tokens is 0', async () => {
|
||||
// Explicit cached_tokens:0 should not be treated differently from absent.
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'hi' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [],
|
||||
usage: {
|
||||
prompt_tokens: 500,
|
||||
completion_tokens: 50,
|
||||
total_tokens: 550,
|
||||
prompt_tokens_details: { cached_tokens: 0 },
|
||||
} as any,
|
||||
}),
|
||||
])
|
||||
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.usage.cache_read_input_tokens).toBe(0)
|
||||
})
|
||||
|
||||
test('cache_read_input_tokens updated when cached_tokens arrives in same chunk as finish_reason', async () => {
|
||||
// Some endpoints send usage in the finish_reason chunk instead of a trailing chunk.
|
||||
const events = await collectEvents([
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: { content: 'result' }, finish_reason: null }],
|
||||
}),
|
||||
makeChunk({
|
||||
choices: [{ index: 0, delta: {}, finish_reason: 'stop' }],
|
||||
usage: {
|
||||
prompt_tokens: 2000,
|
||||
completion_tokens: 100,
|
||||
total_tokens: 2100,
|
||||
prompt_tokens_details: { cached_tokens: 1500 },
|
||||
} as any,
|
||||
}),
|
||||
])
|
||||
|
||||
const msgDelta = events.find(e => e.type === 'message_delta') as any
|
||||
expect(msgDelta.usage.cache_read_input_tokens).toBe(1500)
|
||||
expect(msgDelta.usage.input_tokens).toBe(2000)
|
||||
expect(msgDelta.usage.output_tokens).toBe(100)
|
||||
})
|
||||
})
|
||||
@@ -1,5 +1,5 @@
|
||||
import { describe, expect, test, beforeEach, afterEach } from 'bun:test'
|
||||
import { isOpenAIThinkingEnabled, buildOpenAIRequestBody } from '../index.js'
|
||||
import { isOpenAIThinkingEnabled, buildOpenAIRequestBody } from '../requestBody.js'
|
||||
|
||||
describe('isOpenAIThinkingEnabled', () => {
|
||||
const originalEnv = {
|
||||
|
||||
@@ -1,305 +0,0 @@
|
||||
import type {
|
||||
BetaContentBlockParam,
|
||||
BetaToolResultBlockParam,
|
||||
BetaToolUseBlock,
|
||||
} from '@anthropic-ai/sdk/resources/beta/messages/messages.mjs'
|
||||
import type {
|
||||
ChatCompletionAssistantMessageParam,
|
||||
ChatCompletionMessageParam,
|
||||
ChatCompletionSystemMessageParam,
|
||||
ChatCompletionToolMessageParam,
|
||||
ChatCompletionUserMessageParam,
|
||||
} from 'openai/resources/chat/completions/completions.mjs'
|
||||
import type { AssistantMessage, UserMessage } from '../../../types/message.js'
|
||||
import type { SystemPrompt } from '../../../utils/systemPromptType.js'
|
||||
|
||||
export interface ConvertMessagesOptions {
|
||||
/** When true, preserve thinking blocks as reasoning_content on assistant messages
|
||||
* (required for DeepSeek thinking mode with tool calls). */
|
||||
enableThinking?: boolean
|
||||
}
|
||||
|
||||
/**
|
||||
* Convert internal (UserMessage | AssistantMessage)[] to OpenAI-format messages.
|
||||
*
|
||||
* Key conversions:
|
||||
* - system prompt → role: "system" message prepended
|
||||
* - tool_use blocks → tool_calls[] on assistant message
|
||||
* - tool_result blocks → role: "tool" messages
|
||||
* - thinking blocks → silently dropped (or preserved as reasoning_content when enableThinking=true)
|
||||
* - cache_control → stripped
|
||||
*/
|
||||
export function anthropicMessagesToOpenAI(
|
||||
messages: (UserMessage | AssistantMessage)[],
|
||||
systemPrompt: SystemPrompt,
|
||||
options?: ConvertMessagesOptions,
|
||||
): ChatCompletionMessageParam[] {
|
||||
const result: ChatCompletionMessageParam[] = []
|
||||
const enableThinking = options?.enableThinking ?? false
|
||||
|
||||
// Prepend system prompt as system message
|
||||
const systemText = systemPromptToText(systemPrompt)
|
||||
if (systemText) {
|
||||
result.push({
|
||||
role: 'system',
|
||||
content: systemText,
|
||||
} satisfies ChatCompletionSystemMessageParam)
|
||||
}
|
||||
|
||||
// When thinking mode is on, detect turn boundaries so that reasoning_content
|
||||
// from *previous* user turns is stripped (saves bandwidth; DeepSeek ignores it).
|
||||
// A "new turn" starts when a user text message appears after at least one assistant response.
|
||||
const turnBoundaries = new Set<number>()
|
||||
if (enableThinking) {
|
||||
let hasSeenAssistant = false
|
||||
for (let i = 0; i < messages.length; i++) {
|
||||
const msg = messages[i]
|
||||
if (msg.type === 'assistant') {
|
||||
hasSeenAssistant = true
|
||||
}
|
||||
if (msg.type === 'user' && hasSeenAssistant) {
|
||||
const content = msg.message.content
|
||||
// A user message starts a new turn if it contains any non-tool_result content
|
||||
// (text, image, or other media). Tool results alone do NOT start a new turn
|
||||
// because they are continuations of the previous assistant tool call.
|
||||
const startsNewUserTurn = typeof content === 'string'
|
||||
? content.length > 0
|
||||
: Array.isArray(content) && content.some(
|
||||
(b: any) =>
|
||||
typeof b === 'string' ||
|
||||
(b &&
|
||||
typeof b === 'object' &&
|
||||
'type' in b &&
|
||||
b.type !== 'tool_result'),
|
||||
)
|
||||
if (startsNewUserTurn) {
|
||||
turnBoundaries.add(i)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
for (let i = 0; i < messages.length; i++) {
|
||||
const msg = messages[i]
|
||||
switch (msg.type) {
|
||||
case 'user':
|
||||
result.push(...convertInternalUserMessage(msg))
|
||||
break
|
||||
case 'assistant':
|
||||
// Preserve reasoning_content unless we're before a turn boundary
|
||||
// (i.e., from a previous user Q&A round)
|
||||
const preserveReasoning = enableThinking && !isBeforeAnyTurnBoundary(i, turnBoundaries)
|
||||
result.push(...convertInternalAssistantMessage(msg, preserveReasoning))
|
||||
break
|
||||
default:
|
||||
break
|
||||
}
|
||||
}
|
||||
|
||||
return result
|
||||
}
|
||||
|
||||
function systemPromptToText(systemPrompt: SystemPrompt): string {
|
||||
if (!systemPrompt || systemPrompt.length === 0) return ''
|
||||
return systemPrompt
|
||||
.filter(Boolean)
|
||||
.join('\n\n')
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if index `i` falls before any turn boundary (i.e. it belongs to a previous turn).
|
||||
* A message at index i is "before" a boundary if there exists a boundary j where i < j.
|
||||
*/
|
||||
function isBeforeAnyTurnBoundary(i: number, boundaries: Set<number>): boolean {
|
||||
for (const b of boundaries) {
|
||||
if (i < b) return true
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
function convertInternalUserMessage(
|
||||
msg: UserMessage,
|
||||
): ChatCompletionMessageParam[] {
|
||||
const result: ChatCompletionMessageParam[] = []
|
||||
const content = msg.message.content
|
||||
|
||||
if (typeof content === 'string') {
|
||||
result.push({
|
||||
role: 'user',
|
||||
content,
|
||||
} satisfies ChatCompletionUserMessageParam)
|
||||
} else if (Array.isArray(content)) {
|
||||
const textParts: string[] = []
|
||||
const toolResults: BetaToolResultBlockParam[] = []
|
||||
const imageParts: Array<{ type: 'image_url'; image_url: { url: string } }> = []
|
||||
|
||||
for (const block of content) {
|
||||
if (typeof block === 'string') {
|
||||
textParts.push(block)
|
||||
} else if (block.type === 'text') {
|
||||
textParts.push(block.text)
|
||||
} else if (block.type === 'tool_result') {
|
||||
toolResults.push(block as BetaToolResultBlockParam)
|
||||
} else if (block.type === 'image') {
|
||||
const imagePart = convertImageBlockToOpenAI(block as unknown as Record<string, unknown>)
|
||||
if (imagePart) {
|
||||
imageParts.push(imagePart)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// CRITICAL: tool messages must come BEFORE any user message in the result.
|
||||
// OpenAI API requires that a tool message immediately follows the assistant
|
||||
// message with tool_calls. If we emit a user message first, the API will
|
||||
// reject the request with "insufficient tool messages following tool_calls".
|
||||
// See: https://github.com/anthropics/claude-code/issues/xxx
|
||||
for (const tr of toolResults) {
|
||||
result.push(convertToolResult(tr))
|
||||
}
|
||||
|
||||
// 如果有图片，构建多模态 content 数组
|
||||
if (imageParts.length > 0) {
|
||||
const multiContent: Array<{ type: 'text'; text: string } | { type: 'image_url'; image_url: { url: string } }> = []
|
||||
if (textParts.length > 0) {
|
||||
multiContent.push({ type: 'text', text: textParts.join('\n') })
|
||||
}
|
||||
multiContent.push(...imageParts)
|
||||
result.push({
|
||||
role: 'user',
|
||||
content: multiContent,
|
||||
} satisfies ChatCompletionUserMessageParam)
|
||||
} else if (textParts.length > 0) {
|
||||
result.push({
|
||||
role: 'user',
|
||||
content: textParts.join('\n'),
|
||||
} satisfies ChatCompletionUserMessageParam)
|
||||
}
|
||||
}
|
||||
|
||||
return result
|
||||
}
|
||||
|
||||
function convertToolResult(
|
||||
block: BetaToolResultBlockParam,
|
||||
): ChatCompletionToolMessageParam {
|
||||
let content: string
|
||||
if (typeof block.content === 'string') {
|
||||
content = block.content
|
||||
} else if (Array.isArray(block.content)) {
|
||||
content = block.content
|
||||
.map(c => {
|
||||
if (typeof c === 'string') return c
|
||||
if ('text' in c) return c.text
|
||||
return ''
|
||||
})
|
||||
.filter(Boolean)
|
||||
.join('\n')
|
||||
} else {
|
||||
content = ''
|
||||
}
|
||||
|
||||
return {
|
||||
role: 'tool',
|
||||
tool_call_id: block.tool_use_id,
|
||||
content,
|
||||
} satisfies ChatCompletionToolMessageParam
|
||||
}
|
||||
|
||||
function convertInternalAssistantMessage(
|
||||
msg: AssistantMessage,
|
||||
preserveReasoning = false,
|
||||
): ChatCompletionMessageParam[] {
|
||||
const content = msg.message.content
|
||||
|
||||
if (typeof content === 'string') {
|
||||
return [
|
||||
{
|
||||
role: 'assistant',
|
||||
content,
|
||||
} satisfies ChatCompletionAssistantMessageParam,
|
||||
]
|
||||
}
|
||||
|
||||
if (!Array.isArray(content)) {
|
||||
return [
|
||||
{
|
||||
role: 'assistant',
|
||||
content: '',
|
||||
} satisfies ChatCompletionAssistantMessageParam,
|
||||
]
|
||||
}
|
||||
|
||||
const textParts: string[] = []
|
||||
const toolCalls: NonNullable<ChatCompletionAssistantMessageParam['tool_calls']> = []
|
||||
const reasoningParts: string[] = []
|
||||
|
||||
for (const block of content) {
|
||||
if (typeof block === 'string') {
|
||||
textParts.push(block)
|
||||
} else if (block.type === 'text') {
|
||||
textParts.push(block.text)
|
||||
} else if (block.type === 'tool_use') {
|
||||
const tu = block as BetaToolUseBlock
|
||||
toolCalls.push({
|
||||
id: tu.id,
|
||||
type: 'function',
|
||||
function: {
|
||||
name: tu.name,
|
||||
arguments:
|
||||
typeof tu.input === 'string' ? tu.input : JSON.stringify(tu.input),
|
||||
},
|
||||
})
|
||||
} else if (block.type === 'thinking' && preserveReasoning) {
|
||||
// DeepSeek thinking mode: preserve reasoning_content for tool call iterations
|
||||
const thinkingText = (block as unknown as Record<string, unknown>).thinking
|
||||
if (typeof thinkingText === 'string' && thinkingText) {
|
||||
reasoningParts.push(thinkingText)
|
||||
}
|
||||
}
|
||||
// Skip redacted_thinking, server_tool_use, etc.
|
||||
}
|
||||
|
||||
const result: ChatCompletionAssistantMessageParam = {
|
||||
role: 'assistant',
|
||||
content: textParts.length > 0 ? textParts.join('\n') : null,
|
||||
...(toolCalls.length > 0 && { tool_calls: toolCalls }),
|
||||
...(reasoningParts.length > 0 && { reasoning_content: reasoningParts.join('\n') }),
|
||||
}
|
||||
|
||||
return [result]
|
||||
}
|
||||
|
||||
/**
|
||||
* 将 Anthropic image 块转换为 OpenAI image_url 格式。
|
||||
*
|
||||
* Anthropic 格式: { type: "image", source: { type: "base64", media_type: "image/png", data: "..." } }
|
||||
* OpenAI 格式: { type: "image_url", image_url: { url: "data:image/png;base64,..." } }
|
||||
*/
|
||||
function convertImageBlockToOpenAI(
|
||||
block: Record<string, unknown>,
|
||||
): { type: 'image_url'; image_url: { url: string } } | null {
|
||||
const source = block.source as Record<string, unknown> | undefined
|
||||
if (!source) return null
|
||||
|
||||
if (source.type === 'base64' && typeof source.data === 'string') {
|
||||
const mediaType = (source.media_type as string) || 'image/png'
|
||||
return {
|
||||
type: 'image_url',
|
||||
image_url: {
|
||||
url: `data:${mediaType};base64,${source.data}`,
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
// url 类型的图片直接传递
|
||||
if (source.type === 'url' && typeof source.url === 'string') {
|
||||
return {
|
||||
type: 'image_url',
|
||||
image_url: {
|
||||
url: source.url,
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
return null
|
||||
}
|
||||
@@ -1,123 +0,0 @@
|
||||
import type { BetaToolUnion } from '@anthropic-ai/sdk/resources/beta/messages/messages.mjs'
|
||||
import type { ChatCompletionTool } from 'openai/resources/chat/completions/completions.mjs'
|
||||
|
||||
/**
|
||||
* Convert Anthropic tool schemas to OpenAI function calling format.
|
||||
*
|
||||
* Anthropic: { name, description, input_schema }
|
||||
* OpenAI: { type: "function", function: { name, description, parameters } }
|
||||
*
|
||||
* Anthropic-specific fields (cache_control, defer_loading, etc.) are stripped.
|
||||
*/
|
||||
export function anthropicToolsToOpenAI(
|
||||
tools: BetaToolUnion[],
|
||||
): ChatCompletionTool[] {
|
||||
return tools
|
||||
.filter(tool => {
|
||||
// Only convert standard tools (skip server tools like computer_use, etc.)
|
||||
const toolType = (tool as unknown as { type?: string }).type
|
||||
return tool.type === 'custom' || !('type' in tool) || toolType !== 'server'
|
||||
})
|
||||
.map(tool => {
|
||||
// Handle the various tool shapes from Anthropic SDK
|
||||
const anyTool = tool as unknown as Record<string, unknown>
|
||||
const name = (anyTool.name as string) || ''
|
||||
const description = (anyTool.description as string) || ''
|
||||
const inputSchema = anyTool.input_schema as Record<string, unknown> | undefined
|
||||
|
||||
return {
|
||||
type: 'function' as const,
|
||||
function: {
|
||||
name,
|
||||
description,
|
||||
parameters: sanitizeJsonSchema(inputSchema || { type: 'object', properties: {} }),
|
||||
},
|
||||
} satisfies ChatCompletionTool
|
||||
})
|
||||
}
|
||||
|
||||
/**
|
||||
* Recursively sanitize a JSON Schema for OpenAI-compatible providers.
|
||||
*
|
||||
* Many OpenAI-compatible endpoints (Ollama, DeepSeek, vLLM, etc.) do not
|
||||
* support the `const` keyword in JSON Schema. Convert it to `enum` with a
|
||||
* single-element array, which is semantically equivalent.
|
||||
*/
|
||||
function sanitizeJsonSchema(schema: Record<string, unknown>): Record<string, unknown> {
|
||||
if (!schema || typeof schema !== 'object') return schema
|
||||
|
||||
const result = { ...schema }
|
||||
|
||||
// Convert `const` → `enum: [value]`
|
||||
if ('const' in result) {
|
||||
result.enum = [result.const]
|
||||
delete result.const
|
||||
}
|
||||
|
||||
// Recursively process nested schemas
|
||||
const objectKeys = ['properties', 'definitions', '$defs', 'patternProperties'] as const
|
||||
for (const key of objectKeys) {
|
||||
const nested = result[key]
|
||||
if (nested && typeof nested === 'object') {
|
||||
const sanitized: Record<string, unknown> = {}
|
||||
for (const [k, v] of Object.entries(nested as Record<string, unknown>)) {
|
||||
sanitized[k] = v && typeof v === 'object' ? sanitizeJsonSchema(v as Record<string, unknown>) : v
|
||||
}
|
||||
result[key] = sanitized
|
||||
}
|
||||
}
|
||||
|
||||
// Recursively process single-schema keys
|
||||
const singleKeys = ['items', 'additionalProperties', 'not', 'if', 'then', 'else', 'contains', 'propertyNames'] as const
|
||||
for (const key of singleKeys) {
|
||||
const nested = result[key]
|
||||
if (nested && typeof nested === 'object' && !Array.isArray(nested)) {
|
||||
result[key] = sanitizeJsonSchema(nested as Record<string, unknown>)
|
||||
}
|
||||
}
|
||||
|
||||
// Recursively process array-of-schemas keys
|
||||
const arrayKeys = ['anyOf', 'oneOf', 'allOf'] as const
|
||||
for (const key of arrayKeys) {
|
||||
const nested = result[key]
|
||||
if (Array.isArray(nested)) {
|
||||
result[key] = nested.map(item =>
|
||||
item && typeof item === 'object' ? sanitizeJsonSchema(item as Record<string, unknown>) : item
|
||||
)
|
||||
}
|
||||
}
|
||||
|
||||
return result
|
||||
}
|
||||
|
||||
/**
|
||||
* Map Anthropic tool_choice to OpenAI tool_choice format.
|
||||
*
|
||||
* Anthropic → OpenAI:
|
||||
* - { type: "auto" } → "auto"
|
||||
* - { type: "any" } → "required"
|
||||
* - { type: "tool", name } → { type: "function", function: { name } }
|
||||
* - undefined → undefined (use provider default)
|
||||
*/
|
||||
export function anthropicToolChoiceToOpenAI(
|
||||
toolChoice: unknown,
|
||||
): string | { type: 'function'; function: { name: string } } | undefined {
|
||||
if (!toolChoice || typeof toolChoice !== 'object') return undefined
|
||||
|
||||
const tc = toolChoice as Record<string, unknown>
|
||||
const type = tc.type as string
|
||||
|
||||
switch (type) {
|
||||
case 'auto':
|
||||
return 'auto'
|
||||
case 'any':
|
||||
return 'required'
|
||||
case 'tool':
|
||||
return {
|
||||
type: 'function',
|
||||
function: { name: tc.name as string },
|
||||
}
|
||||
default:
|
||||
return undefined
|
||||
}
|
||||
}
|
||||
@@ -10,17 +10,10 @@ import type { AgentId } from '../../../types/ids.js'
|
||||
import type { Tools } from '../../../Tool.js'
|
||||
import type { Stream } from 'openai/streaming.mjs'
|
||||
import type {
|
||||
ChatCompletionChunk,
|
||||
ChatCompletionCreateParamsStreaming,
|
||||
} from 'openai/resources/chat/completions/completions.mjs'
|
||||
import { getOpenAIClient } from './client.js'
|
||||
import { anthropicMessagesToOpenAI } from './convertMessages.js'
|
||||
import {
|
||||
anthropicToolsToOpenAI,
|
||||
anthropicToolChoiceToOpenAI,
|
||||
} from './convertTools.js'
|
||||
import { adaptOpenAIStreamToAnthropic } from './streamAdapter.js'
|
||||
import { resolveOpenAIModel } from './modelMapping.js'
|
||||
import { anthropicMessagesToOpenAI, resolveOpenAIModel, adaptOpenAIStreamToAnthropic, anthropicToolsToOpenAI, anthropicToolChoiceToOpenAI } from '@ant/model-provider'
|
||||
import { normalizeMessagesForAPI } from '../../../utils/messages.js'
|
||||
import { toolToAPISchema } from '../../../utils/api.js'
|
||||
import {
|
||||
@@ -30,7 +23,8 @@ import {
|
||||
import { logForDebugging } from '../../../utils/debug.js'
|
||||
import { addToTotalSessionCost } from '../../../cost-tracker.js'
|
||||
import { calculateUSDCost } from '../../../utils/modelCost.js'
|
||||
import { isEnvTruthy, isEnvDefinedFalsy } from '../../../utils/envUtils.js'
|
||||
import { isOpenAIThinkingEnabled, resolveOpenAIMaxTokens, buildOpenAIRequestBody } from './requestBody.js'
|
||||
export { isOpenAIThinkingEnabled, resolveOpenAIMaxTokens, buildOpenAIRequestBody }
|
||||
import { getModelMaxOutputTokens } from '../../../utils/context.js'
|
||||
import type { Options } from '../claude.js'
|
||||
import { randomUUID } from 'crypto'
|
||||
@@ -48,104 +42,6 @@ import {
|
||||
TOOL_SEARCH_TOOL_NAME,
|
||||
} from '@claude-code-best/builtin-tools/tools/ToolSearchTool/prompt.js'
|
||||
|
||||
/**
|
||||
* Detect whether DeepSeek-style thinking mode should be enabled.
|
||||
*
|
||||
* Enabled when:
|
||||
* 1. OPENAI_ENABLE_THINKING=1 is set (explicit enable), OR
|
||||
* 2. Model name contains "deepseek-reasoner" OR "DeepSeek-V3.2" (auto-detect, case-insensitive)
|
||||
*
|
||||
* Disabled when:
|
||||
* - OPENAI_ENABLE_THINKING=0/false/no/off is explicitly set (overrides model detection)
|
||||
*
|
||||
* @param model - The resolved OpenAI model name
|
||||
* @internal Exported for testing purposes only
|
||||
*/
|
||||
export function isOpenAIThinkingEnabled(model: string): boolean {
|
||||
// Explicit disable takes priority (overrides model auto-detect)
|
||||
if (isEnvDefinedFalsy(process.env.OPENAI_ENABLE_THINKING)) return false
|
||||
// Explicit enable
|
||||
if (isEnvTruthy(process.env.OPENAI_ENABLE_THINKING)) return true
|
||||
// Auto-detect from model name (deepseek-reasoner and DeepSeek-V3.2 support thinking mode)
|
||||
const modelLower = model.toLowerCase()
|
||||
return modelLower.includes('deepseek-reasoner') || modelLower.includes('deepseek-v3.2')
|
||||
}
|
||||
|
||||
/**
|
||||
* Resolve max output tokens for the OpenAI-compatible path.
|
||||
*
|
||||
* Override priority:
|
||||
* 1. maxOutputTokensOverride (programmatic, from query pipeline)
|
||||
* 2. OPENAI_MAX_TOKENS env var (OpenAI-specific, useful for local models
|
||||
* with small context windows, e.g. RTX 3060 12GB running 65536-token models)
|
||||
* 3. CLAUDE_CODE_MAX_OUTPUT_TOKENS env var (generic override)
|
||||
* 4. upperLimit default (64000)
|
||||
*
|
||||
* @internal Exported for testing purposes only
|
||||
*/
|
||||
export function resolveOpenAIMaxTokens(
|
||||
upperLimit: number,
|
||||
maxOutputTokensOverride?: number,
|
||||
): number {
|
||||
return maxOutputTokensOverride
|
||||
?? (process.env.OPENAI_MAX_TOKENS ? parseInt(process.env.OPENAI_MAX_TOKENS, 10) || undefined : undefined)
|
||||
?? (process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS ? parseInt(process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS, 10) || undefined : undefined)
|
||||
?? upperLimit
|
||||
}
|
||||
|
||||
/**
|
||||
* Build the request body for OpenAI chat.completions.create().
|
||||
* Extracted for testability — the thinking mode params are injected here.
|
||||
*
|
||||
* DeepSeek thinking mode: inject thinking params via request body.
|
||||
* Two formats are added simultaneously to support different deployments:
|
||||
* - Official DeepSeek API: `thinking: { type: 'enabled' }`
|
||||
* - Self-hosted DeepSeek-V3.2: `enable_thinking: true` + `chat_template_kwargs: { thinking: true }`
|
||||
* OpenAI SDK passes unknown keys through to the HTTP body.
|
||||
* Each endpoint will use the format it recognizes and ignore the others.
|
||||
* @internal Exported for testing purposes only
|
||||
*/
|
||||
export function buildOpenAIRequestBody(params: {
|
||||
model: string
|
||||
messages: any[]
|
||||
tools: any[]
|
||||
toolChoice: any
|
||||
enableThinking: boolean
|
||||
maxTokens: number
|
||||
temperatureOverride?: number
|
||||
}): ChatCompletionCreateParamsStreaming & {
|
||||
thinking?: { type: string }
|
||||
enable_thinking?: boolean
|
||||
chat_template_kwargs?: { thinking: boolean }
|
||||
} {
|
||||
const { model, messages, tools, toolChoice, enableThinking, maxTokens, temperatureOverride } = params
|
||||
return {
|
||||
model,
|
||||
messages,
|
||||
max_tokens: maxTokens,
|
||||
...(tools.length > 0 && {
|
||||
tools,
|
||||
...(toolChoice && { tool_choice: toolChoice }),
|
||||
}),
|
||||
stream: true,
|
||||
stream_options: { include_usage: true },
|
||||
// DeepSeek thinking mode: enable chain-of-thought output.
|
||||
// When active, temperature/top_p/presence_penalty/frequency_penalty are ignored by DeepSeek.
|
||||
...(enableThinking && {
|
||||
// Official DeepSeek API format
|
||||
thinking: { type: 'enabled' },
|
||||
// Self-hosted DeepSeek-V3.2 format
|
||||
enable_thinking: true,
|
||||
chat_template_kwargs: { thinking: true },
|
||||
}),
|
||||
// Only send temperature when thinking mode is off (DeepSeek ignores it anyway,
|
||||
// but other providers may respect it)
|
||||
...(!enableThinking && temperatureOverride !== undefined && {
|
||||
temperature: temperatureOverride,
|
||||
}),
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Assemble the final AssistantMessage (and optional max_tokens error) from
|
||||
* accumulated stream state. Extracted to avoid duplication between the
|
||||
|
||||
@@ -1,63 +0,0 @@
|
||||
/**
|
||||
* Default mapping from Anthropic model names to OpenAI model names.
|
||||
* Used only when ANTHROPIC_DEFAULT_*_MODEL env vars are not set.
|
||||
*/
|
||||
const DEFAULT_MODEL_MAP: Record<string, string> = {
|
||||
'claude-sonnet-4-20250514': 'gpt-4o',
|
||||
'claude-sonnet-4-5-20250929': 'gpt-4o',
|
||||
'claude-sonnet-4-6': 'gpt-4o',
|
||||
'claude-opus-4-20250514': 'o3',
|
||||
'claude-opus-4-1-20250805': 'o3',
|
||||
'claude-opus-4-5-20251101': 'o3',
|
||||
'claude-opus-4-6': 'o3',
|
||||
'claude-haiku-4-5-20251001': 'gpt-4o-mini',
|
||||
'claude-3-5-haiku-20241022': 'gpt-4o-mini',
|
||||
'claude-3-7-sonnet-20250219': 'gpt-4o',
|
||||
'claude-3-5-sonnet-20241022': 'gpt-4o',
|
||||
}
|
||||
|
||||
/**
|
||||
* Determine the model family (haiku / sonnet / opus) from an Anthropic model ID.
|
||||
*/
|
||||
function getModelFamily(model: string): 'haiku' | 'sonnet' | 'opus' | null {
|
||||
if (/haiku/i.test(model)) return 'haiku'
|
||||
if (/opus/i.test(model)) return 'opus'
|
||||
if (/sonnet/i.test(model)) return 'sonnet'
|
||||
return null
|
||||
}
|
||||
|
||||
/**
|
||||
* Resolve the OpenAI model name for a given Anthropic model.
|
||||
*
|
||||
* Priority:
|
||||
* 1. OPENAI_MODEL env var (override all)
|
||||
* 2. OPENAI_DEFAULT_{FAMILY}_MODEL env var (e.g. OPENAI_DEFAULT_SONNET_MODEL)
|
||||
* 3. ANTHROPIC_DEFAULT_{FAMILY}_MODEL env var (backward compatibility)
|
||||
* 4. DEFAULT_MODEL_MAP lookup
|
||||
* 5. Pass through original model name
|
||||
*/
|
||||
export function resolveOpenAIModel(anthropicModel: string): string {
|
||||
// Highest priority: explicit override
|
||||
if (process.env.OPENAI_MODEL) {
|
||||
return process.env.OPENAI_MODEL
|
||||
}
|
||||
|
||||
// Strip [1m] suffix if present (Claude-specific modifier)
|
||||
const cleanModel = anthropicModel.replace(/\[1m\]$/, '')
|
||||
|
||||
// Check family-specific overrides
|
||||
const family = getModelFamily(cleanModel)
|
||||
if (family) {
|
||||
// OpenAI-specific family override (preferred for openai provider)
|
||||
const openaiEnvVar = `OPENAI_DEFAULT_${family.toUpperCase()}_MODEL`
|
||||
const openaiOverride = process.env[openaiEnvVar]
|
||||
if (openaiOverride) return openaiOverride
|
||||
|
||||
// Anthropic env var (backward compatibility)
|
||||
const anthropicEnvVar = `ANTHROPIC_DEFAULT_${family.toUpperCase()}_MODEL`
|
||||
const anthropicOverride = process.env[anthropicEnvVar]
|
||||
if (anthropicOverride) return anthropicOverride
|
||||
}
|
||||
|
||||
return DEFAULT_MODEL_MAP[cleanModel] ?? cleanModel
|
||||
}
|
||||
103
src/services/api/openai/requestBody.ts
Normal file
103
src/services/api/openai/requestBody.ts
Normal file
@@ -0,0 +1,103 @@
|
||||
/**
|
||||
* Pure utility functions for building OpenAI request bodies and detecting
|
||||
* thinking mode. Extracted from index.ts so tests can import them without
|
||||
* triggering heavy module side-effects (OpenAI client, stream adapter, etc.).
|
||||
*/
|
||||
import type {
|
||||
ChatCompletionCreateParamsStreaming,
|
||||
} from 'openai/resources/chat/completions/completions.mjs'
|
||||
import { isEnvTruthy, isEnvDefinedFalsy } from '../../../utils/envUtils.js'
|
||||
|
||||
/**
|
||||
* Detect whether DeepSeek-style thinking mode should be enabled.
|
||||
*
|
||||
* Enabled when:
|
||||
* 1. OPENAI_ENABLE_THINKING=1 is set (explicit enable), OR
|
||||
* 2. Model name contains "deepseek-reasoner" OR "DeepSeek-V3.2" (auto-detect, case-insensitive)
|
||||
*
|
||||
* Disabled when:
|
||||
* - OPENAI_ENABLE_THINKING=0/false/no/off is explicitly set (overrides model detection)
|
||||
*
|
||||
* @param model - The resolved OpenAI model name
|
||||
*/
|
||||
export function isOpenAIThinkingEnabled(model: string): boolean {
|
||||
// Explicit disable takes priority (overrides model auto-detect)
|
||||
if (isEnvDefinedFalsy(process.env.OPENAI_ENABLE_THINKING)) return false
|
||||
// Explicit enable
|
||||
if (isEnvTruthy(process.env.OPENAI_ENABLE_THINKING)) return true
|
||||
// Auto-detect from model name (deepseek-reasoner and DeepSeek-V3.2 support thinking mode)
|
||||
const modelLower = model.toLowerCase()
|
||||
return modelLower.includes('deepseek-reasoner') || modelLower.includes('deepseek-v3.2')
|
||||
}
|
||||
|
||||
/**
|
||||
* Resolve max output tokens for the OpenAI-compatible path.
|
||||
*
|
||||
* Override priority:
|
||||
* 1. maxOutputTokensOverride (programmatic, from query pipeline)
|
||||
* 2. OPENAI_MAX_TOKENS env var (OpenAI-specific, useful for local models
|
||||
* with small context windows, e.g. RTX 3060 12GB running 65536-token models)
|
||||
* 3. CLAUDE_CODE_MAX_OUTPUT_TOKENS env var (generic override)
|
||||
* 4. upperLimit default (64000)
|
||||
*/
|
||||
export function resolveOpenAIMaxTokens(
|
||||
upperLimit: number,
|
||||
maxOutputTokensOverride?: number,
|
||||
): number {
|
||||
return maxOutputTokensOverride
|
||||
?? (process.env.OPENAI_MAX_TOKENS ? parseInt(process.env.OPENAI_MAX_TOKENS, 10) || undefined : undefined)
|
||||
?? (process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS ? parseInt(process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS, 10) || undefined : undefined)
|
||||
?? upperLimit
|
||||
}
|
||||
|
||||
/**
|
||||
* Build the request body for OpenAI chat.completions.create().
|
||||
* Extracted for testability — the thinking mode params are injected here.
|
||||
*
|
||||
* DeepSeek thinking mode: inject thinking params via request body.
|
||||
* Two formats are added simultaneously to support different deployments:
|
||||
* - Official DeepSeek API: `thinking: { type: 'enabled' }`
|
||||
* - Self-hosted DeepSeek-V3.2: `enable_thinking: true` + `chat_template_kwargs: { thinking: true }`
|
||||
* OpenAI SDK passes unknown keys through to the HTTP body.
|
||||
* Each endpoint will use the format it recognizes and ignore the others.
|
||||
*/
|
||||
export function buildOpenAIRequestBody(params: {
|
||||
model: string
|
||||
messages: any[]
|
||||
tools: any[]
|
||||
toolChoice: any
|
||||
enableThinking: boolean
|
||||
maxTokens: number
|
||||
temperatureOverride?: number
|
||||
}): ChatCompletionCreateParamsStreaming & {
|
||||
thinking?: { type: string }
|
||||
enable_thinking?: boolean
|
||||
chat_template_kwargs?: { thinking: boolean }
|
||||
} {
|
||||
const { model, messages, tools, toolChoice, enableThinking, maxTokens, temperatureOverride } = params
|
||||
return {
|
||||
model,
|
||||
messages,
|
||||
max_tokens: maxTokens,
|
||||
...(tools.length > 0 && {
|
||||
tools,
|
||||
...(toolChoice && { tool_choice: toolChoice }),
|
||||
}),
|
||||
stream: true,
|
||||
stream_options: { include_usage: true },
|
||||
// DeepSeek thinking mode: enable chain-of-thought output.
|
||||
// When active, temperature/top_p/presence_penalty/frequency_penalty are ignored by DeepSeek.
|
||||
...(enableThinking && {
|
||||
// Official DeepSeek API format
|
||||
thinking: { type: 'enabled' },
|
||||
// Self-hosted DeepSeek-V3.2 format
|
||||
enable_thinking: true,
|
||||
chat_template_kwargs: { thinking: true },
|
||||
}),
|
||||
// Only send temperature when thinking mode is off (DeepSeek ignores it anyway,
|
||||
// but other providers may respect it)
|
||||
...(!enableThinking && temperatureOverride !== undefined && {
|
||||
temperature: temperatureOverride,
|
||||
}),
|
||||
}
|
||||
}
|
||||
@@ -1,375 +0,0 @@
|
||||
import type { BetaRawMessageStreamEvent } from '@anthropic-ai/sdk/resources/beta/messages/messages.mjs'
|
||||
import type { ChatCompletionChunk } from 'openai/resources/chat/completions/completions.mjs'
|
||||
import { randomUUID } from 'crypto'
|
||||
|
||||
/**
|
||||
* Adapt an OpenAI streaming response into Anthropic BetaRawMessageStreamEvent.
|
||||
*
|
||||
* Mapping:
|
||||
* First chunk → message_start
|
||||
* delta.reasoning_content → content_block_start(thinking) + thinking_delta + content_block_stop
|
||||
* delta.content → content_block_start(text) + text_delta + content_block_stop
|
||||
* delta.tool_calls → content_block_start(tool_use) + input_json_delta + content_block_stop
|
||||
* finish_reason → message_delta(stop_reason) + message_stop
|
||||
*
|
||||
* Usage field mapping (OpenAI → Anthropic):
|
||||
* prompt_tokens → input_tokens
|
||||
* completion_tokens → output_tokens
|
||||
* prompt_tokens_details.cached_tokens → cache_read_input_tokens
|
||||
* (no OpenAI equivalent) → cache_creation_input_tokens (always 0)
|
||||
*
|
||||
* All four fields are emitted in the post-loop message_delta (not message_start)
|
||||
* so that trailing usage chunks (sent after finish_reason by some
|
||||
* OpenAI-compatible endpoints) are fully captured before the final counts are reported.
|
||||
*
|
||||
* Thinking support:
|
||||
* DeepSeek and compatible providers send `delta.reasoning_content` for chain-of-thought.
|
||||
* This is mapped to Anthropic's `thinking` content blocks:
|
||||
* content_block_start: { type: 'thinking', thinking: '', signature: '' }
|
||||
* content_block_delta: { type: 'thinking_delta', thinking: '...' }
|
||||
*
|
||||
* Prompt caching:
|
||||
* OpenAI reports cached tokens in usage.prompt_tokens_details.cached_tokens.
|
||||
* This is mapped to Anthropic's cache_read_input_tokens.
|
||||
*/
|
||||
export async function* adaptOpenAIStreamToAnthropic(
|
||||
stream: AsyncIterable<ChatCompletionChunk>,
|
||||
model: string,
|
||||
): AsyncGenerator<BetaRawMessageStreamEvent, void> {
|
||||
const messageId = `msg_${randomUUID().replace(/-/g, '').slice(0, 24)}`
|
||||
|
||||
let started = false
|
||||
let currentContentIndex = -1
|
||||
|
||||
// Track tool_use blocks: tool_calls index → { contentIndex, id, name, arguments }
|
||||
const toolBlocks = new Map<number, { contentIndex: number; id: string; name: string; arguments: string }>()
|
||||
|
||||
// Track thinking block state
|
||||
let thinkingBlockOpen = false
|
||||
|
||||
// Track text block state
|
||||
let textBlockOpen = false
|
||||
|
||||
// Track usage — all four Anthropic fields, populated from OpenAI usage fields:
|
||||
// prompt_tokens → input_tokens
|
||||
// completion_tokens → output_tokens
|
||||
// prompt_tokens_details.cached_tokens → cache_read_input_tokens
|
||||
// (no standard OpenAI equivalent) → cache_creation_input_tokens (always 0)
|
||||
let inputTokens = 0
|
||||
let outputTokens = 0
|
||||
let cachedReadTokens = 0
|
||||
|
||||
// Track all open content block indices (for cleanup)
|
||||
const openBlockIndices = new Set<number>()
|
||||
|
||||
// Deferred finish state: populated when finish_reason is encountered so that
|
||||
// message_delta / message_stop are emitted AFTER the stream loop ends.
|
||||
// This ensures usage chunks that arrive after the finish_reason chunk are
|
||||
// captured before we emit the final token counts.
|
||||
let pendingFinishReason: string | null = null
|
||||
let pendingHasToolCalls = false
|
||||
|
||||
for await (const chunk of stream) {
|
||||
const choice = chunk.choices?.[0]
|
||||
const delta = choice?.delta
|
||||
|
||||
// Extract usage from any chunk that carries it.
|
||||
// Many OpenAI-compatible endpoints (e.g. DeepSeek) send usage in a separate
|
||||
// final chunk that arrives AFTER the finish_reason chunk. Reading it here
|
||||
// (before emitting message_delta) ensures the token counts are available
|
||||
// when we later emit message_delta.
|
||||
if (chunk.usage) {
|
||||
inputTokens = chunk.usage.prompt_tokens ?? inputTokens
|
||||
outputTokens = chunk.usage.completion_tokens ?? outputTokens
|
||||
// OpenAI prompt caching: prompt_tokens_details.cached_tokens
|
||||
// → Anthropic cache_read_input_tokens
|
||||
// Note: OpenAI has no equivalent for cache_creation_input_tokens.
|
||||
const details = (chunk.usage as any).prompt_tokens_details
|
||||
if (details?.cached_tokens != null) {
|
||||
cachedReadTokens = details.cached_tokens
|
||||
}
|
||||
}
|
||||
|
||||
// Emit message_start on first chunk
|
||||
if (!started) {
|
||||
started = true
|
||||
|
||||
yield {
|
||||
type: 'message_start',
|
||||
message: {
|
||||
id: messageId,
|
||||
type: 'message',
|
||||
role: 'assistant',
|
||||
content: [],
|
||||
model,
|
||||
stop_reason: null,
|
||||
stop_sequence: null,
|
||||
usage: {
|
||||
input_tokens: inputTokens,
|
||||
output_tokens: 0,
|
||||
cache_creation_input_tokens: 0,
|
||||
cache_read_input_tokens: cachedReadTokens,
|
||||
},
|
||||
},
|
||||
} as unknown as BetaRawMessageStreamEvent
|
||||
}
|
||||
|
||||
// Skip chunks that carry only usage data (no delta content)
|
||||
if (!delta) continue
|
||||
|
||||
// Handle reasoning_content → Anthropic thinking block
|
||||
// DeepSeek and compatible providers send delta.reasoning_content
|
||||
const reasoningContent = (delta as any).reasoning_content
|
||||
if (reasoningContent != null && reasoningContent !== '') {
|
||||
if (!thinkingBlockOpen) {
|
||||
currentContentIndex++
|
||||
thinkingBlockOpen = true
|
||||
openBlockIndices.add(currentContentIndex)
|
||||
|
||||
yield {
|
||||
type: 'content_block_start',
|
||||
index: currentContentIndex,
|
||||
content_block: {
|
||||
type: 'thinking',
|
||||
thinking: '',
|
||||
signature: '',
|
||||
},
|
||||
} as BetaRawMessageStreamEvent
|
||||
}
|
||||
|
||||
yield {
|
||||
type: 'content_block_delta',
|
||||
index: currentContentIndex,
|
||||
delta: {
|
||||
type: 'thinking_delta',
|
||||
thinking: reasoningContent,
|
||||
},
|
||||
} as BetaRawMessageStreamEvent
|
||||
}
|
||||
|
||||
// Handle text content
|
||||
if (delta.content != null && delta.content !== '') {
|
||||
if (!textBlockOpen) {
|
||||
// Close thinking block if still open (reasoning done, now generating answer)
|
||||
if (thinkingBlockOpen) {
|
||||
yield {
|
||||
type: 'content_block_stop',
|
||||
index: currentContentIndex,
|
||||
} as BetaRawMessageStreamEvent
|
||||
openBlockIndices.delete(currentContentIndex)
|
||||
thinkingBlockOpen = false
|
||||
}
|
||||
|
||||
currentContentIndex++
|
||||
textBlockOpen = true
|
||||
openBlockIndices.add(currentContentIndex)
|
||||
|
||||
yield {
|
||||
type: 'content_block_start',
|
||||
index: currentContentIndex,
|
||||
content_block: {
|
||||
type: 'text',
|
||||
text: '',
|
||||
},
|
||||
} as BetaRawMessageStreamEvent
|
||||
}
|
||||
|
||||
yield {
|
||||
type: 'content_block_delta',
|
||||
index: currentContentIndex,
|
||||
delta: {
|
||||
type: 'text_delta',
|
||||
text: delta.content,
|
||||
},
|
||||
} as BetaRawMessageStreamEvent
|
||||
}
|
||||
|
||||
// Handle tool calls
|
||||
if (delta.tool_calls) {
|
||||
for (const tc of delta.tool_calls) {
|
||||
const tcIndex = tc.index
|
||||
|
||||
if (!toolBlocks.has(tcIndex)) {
|
||||
// Close thinking block if open
|
||||
if (thinkingBlockOpen) {
|
||||
yield {
|
||||
type: 'content_block_stop',
|
||||
index: currentContentIndex,
|
||||
} as BetaRawMessageStreamEvent
|
||||
openBlockIndices.delete(currentContentIndex)
|
||||
thinkingBlockOpen = false
|
||||
}
|
||||
|
||||
// Close text block if open
|
||||
if (textBlockOpen) {
|
||||
yield {
|
||||
type: 'content_block_stop',
|
||||
index: currentContentIndex,
|
||||
} as BetaRawMessageStreamEvent
|
||||
openBlockIndices.delete(currentContentIndex)
|
||||
textBlockOpen = false
|
||||
}
|
||||
|
||||
// Start new tool_use block
|
||||
currentContentIndex++
|
||||
const toolId = tc.id || `toolu_${randomUUID().replace(/-/g, '').slice(0, 24)}`
|
||||
const toolName = tc.function?.name || ''
|
||||
|
||||
toolBlocks.set(tcIndex, {
|
||||
contentIndex: currentContentIndex,
|
||||
id: toolId,
|
||||
name: toolName,
|
||||
arguments: '',
|
||||
})
|
||||
openBlockIndices.add(currentContentIndex)
|
||||
|
||||
yield {
|
||||
type: 'content_block_start',
|
||||
index: currentContentIndex,
|
||||
content_block: {
|
||||
type: 'tool_use',
|
||||
id: toolId,
|
||||
name: toolName,
|
||||
input: {},
|
||||
},
|
||||
} as BetaRawMessageStreamEvent
|
||||
}
|
||||
|
||||
// Stream argument fragments
|
||||
const argFragment = tc.function?.arguments
|
||||
if (argFragment) {
|
||||
toolBlocks.get(tcIndex)!.arguments += argFragment
|
||||
yield {
|
||||
type: 'content_block_delta',
|
||||
index: toolBlocks.get(tcIndex)!.contentIndex,
|
||||
delta: {
|
||||
type: 'input_json_delta',
|
||||
partial_json: argFragment,
|
||||
},
|
||||
} as BetaRawMessageStreamEvent
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Handle finish: close all open content blocks and record the finish_reason.
|
||||
// message_delta + message_stop are emitted AFTER the stream loop so that any
|
||||
// trailing usage chunk (sent after the finish chunk by some endpoints)
|
||||
// is captured first — ensuring token counts are non-zero.
|
||||
if (choice?.finish_reason) {
|
||||
// Close thinking block if still open
|
||||
if (thinkingBlockOpen) {
|
||||
yield {
|
||||
type: 'content_block_stop',
|
||||
index: currentContentIndex,
|
||||
} as BetaRawMessageStreamEvent
|
||||
openBlockIndices.delete(currentContentIndex)
|
||||
thinkingBlockOpen = false
|
||||
}
|
||||
|
||||
// Close text block if still open
|
||||
if (textBlockOpen) {
|
||||
yield {
|
||||
type: 'content_block_stop',
|
||||
index: currentContentIndex,
|
||||
} as BetaRawMessageStreamEvent
|
||||
openBlockIndices.delete(currentContentIndex)
|
||||
textBlockOpen = false
|
||||
}
|
||||
|
||||
// Close all tool blocks that haven't been closed yet
|
||||
for (const [, block] of toolBlocks) {
|
||||
if (openBlockIndices.has(block.contentIndex)) {
|
||||
yield {
|
||||
type: 'content_block_stop',
|
||||
index: block.contentIndex,
|
||||
} as BetaRawMessageStreamEvent
|
||||
openBlockIndices.delete(block.contentIndex)
|
||||
}
|
||||
}
|
||||
|
||||
// Defer message_delta / message_stop until after the loop so that any
|
||||
// trailing usage chunk is processed before we emit the final token counts.
|
||||
pendingFinishReason = choice.finish_reason
|
||||
pendingHasToolCalls = toolBlocks.size > 0
|
||||
}
|
||||
}
|
||||
|
||||
// Safety: close any remaining open blocks if stream ended without finish_reason
|
||||
for (const idx of openBlockIndices) {
|
||||
yield {
|
||||
type: 'content_block_stop',
|
||||
index: idx,
|
||||
} as BetaRawMessageStreamEvent
|
||||
}
|
||||
|
||||
// Emit message_delta + message_stop now that the stream is fully consumed.
|
||||
// Usage values (inputTokens / outputTokens) reflect all chunks including any
|
||||
// trailing usage-only chunk sent after the finish_reason chunk.
|
||||
if (pendingFinishReason !== null) {
|
||||
// Map finish_reason to Anthropic stop_reason.
|
||||
// CRITICAL: When finish_reason is 'length' (token budget exhausted), always
|
||||
// report 'max_tokens' regardless of whether partial tool calls were received.
|
||||
// Otherwise the query loop would try to execute tool calls with incomplete
|
||||
// JSON arguments instead of triggering the max_tokens retry/recovery path.
|
||||
const stopReason =
|
||||
pendingFinishReason === 'length'
|
||||
? 'max_tokens'
|
||||
: pendingHasToolCalls
|
||||
? 'tool_use'
|
||||
: mapFinishReason(pendingFinishReason)
|
||||
|
||||
yield {
|
||||
type: 'message_delta',
|
||||
delta: {
|
||||
stop_reason: stopReason,
|
||||
stop_sequence: null,
|
||||
},
|
||||
// Carry all four Anthropic usage fields so queryModelOpenAI's message_delta
|
||||
// handler (which spreads this into the accumulated usage object) can override
|
||||
// every field that message_start emitted as 0. For endpoints that send usage
|
||||
// in a trailing chunk (e.g. DeepSeek), message_start is emitted on the first
|
||||
// content chunk before the trailing usage chunk arrives, so all four fields
|
||||
// start at 0. By the time we reach here (post-loop) the trailing chunk has
|
||||
// been processed and all values reflect the real counts.
|
||||
//
|
||||
// OpenAI → Anthropic field mapping:
|
||||
// prompt_tokens → input_tokens
|
||||
// completion_tokens → output_tokens
|
||||
// prompt_tokens_details.cached_tokens → cache_read_input_tokens
|
||||
// (no OpenAI equivalent) → cache_creation_input_tokens (stays 0)
|
||||
usage: {
|
||||
input_tokens: inputTokens,
|
||||
output_tokens: outputTokens,
|
||||
cache_read_input_tokens: cachedReadTokens,
|
||||
cache_creation_input_tokens: 0,
|
||||
},
|
||||
} as BetaRawMessageStreamEvent
|
||||
|
||||
yield {
|
||||
type: 'message_stop',
|
||||
} as BetaRawMessageStreamEvent
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Map OpenAI finish_reason to Anthropic stop_reason.
|
||||
*
|
||||
* stop → end_turn
|
||||
* tool_calls → tool_use
|
||||
* length → max_tokens
|
||||
* content_filter → end_turn
|
||||
*/
|
||||
function mapFinishReason(reason: string): string {
|
||||
switch (reason) {
|
||||
case 'stop':
|
||||
return 'end_turn'
|
||||
case 'tool_calls':
|
||||
return 'tool_use'
|
||||
case 'length':
|
||||
return 'max_tokens'
|
||||
case 'content_filter':
|
||||
return 'end_turn'
|
||||
default:
|
||||
return 'end_turn'
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user