mirror of
https://wget.la/https://github.com/Wxw-Gu/WechatExplorer
synced 2026-10-04 03:01:42 +08:00
fix: 完善查询agent时间与检索契约
增加调试与诊断
This commit is contained in:
Wxw-Gu
+9
-5
@@ -14,10 +14,14 @@ resources/connectors/wechat-personal/
|
||||
.omc
|
||||
.codex/
|
||||
skills-lock.json
|
||||
AGENTS.md
|
||||
*__screenshots__
|
||||
|
||||
/AGENTS.md
|
||||
/CLAUDE.md
|
||||
/.claude/
|
||||
/.workbuddy/
|
||||
/.ai-local/
|
||||
|
||||
findings.md
|
||||
progress.md
|
||||
task_plan.md
|
||||
.agents
|
||||
*__screenshots__
|
||||
.ai-local
|
||||
task_plan.md
|
||||
@@ -48,6 +48,7 @@ Agent Hub 让微信机器人调用本机 TraceMemo；Reader Skill / Local HTTP A
|
||||
## 开发文档
|
||||
|
||||
- [开发、测试与构建](./development/overview.md)
|
||||
- [Query Agent POC（开发测试入口）](./development/query-agent-poc.md)
|
||||
- [本地启动排障](./development/local-startup-troubleshooting.md)
|
||||
- [macOS 数据访问说明](./platform/macos.md)
|
||||
- [关闭 SIP 教程](./mac-disable-sip.md)
|
||||
|
||||
@@ -1,27 +0,0 @@
|
||||
# Query Agent POC
|
||||
|
||||
这是独立的开发测试入口，不会修改生产“问问微信”执行链。
|
||||
|
||||
先启动 TraceMemo，并在 API Center 开启 Local HTTP API。然后在仓库根目录运行 例：
|
||||
|
||||
```bash
|
||||
pnpm poc:query-agent -- "我和BOBO第一次聊了什么"
|
||||
```
|
||||
|
||||
也可以直接传入其他自然语言问题：
|
||||
|
||||
```bash
|
||||
pnpm poc:query-agent -- "BOBO上个月有没有给我发过文件"
|
||||
```
|
||||
|
||||
POC 使用设置页当前默认 AI Provider、模型、Base URL 和安全存储中的 API Key。Local Query API 仍使用现有 Bearer Token；POC 输出不会打印 Token、API Key、数据库路径或内部消息 ID。
|
||||
|
||||
输出为 JSON，包含：
|
||||
|
||||
- `question`、`provider`、`model`
|
||||
- `modelCallCount`、`toolCallCount`
|
||||
- `firstModelMs`、`toolTotalMs`、`finalModelMs`、`totalMs`
|
||||
- 每次工具调用的名称、脱敏参数、耗时、状态和结果数量
|
||||
- 最终 `answer` 或错误信息
|
||||
|
||||
工具调用最多 5 次，只允许 `query_messages`、`search_messages`、`message_context`、`conversation_overview`。未配置 AI Provider、Local Query API 未启动或当前 Provider 协议不支持 tools 时，POC 会直接返回错误，不会回退到另一套模型配置。
|
||||
@@ -52,7 +52,7 @@ pnpm test:e2e:build
|
||||
|
||||
```bash
|
||||
git diff --check
|
||||
rg -n "v2\.1\.7|TraceMemo|迹忆|mcpServers|无鉴权" README.md docs --glob '*.md' --glob '!DOCUMENTATION_AUDIT.md' --glob '!development/overview.md'
|
||||
rg -n "v2\.1\.7|TraceMemo|迹忆|mcpServers|无鉴权" README.md docs --glob '*.md' --glob '!development/overview.md'
|
||||
```
|
||||
|
||||
历史迁移说明可以出现旧版本号；正式使用指南不要把过时版本写成当前版本。负向澄清“6131 不是 MCP Server”可以保留，以防用户照抄错误配置。
|
||||
|
||||
@@ -0,0 +1,84 @@
|
||||
# Query Agent POC
|
||||
|
||||
这是独立的开发测试入口，不会修改生产“问问微信”执行链。
|
||||
|
||||
先启动 TraceMemo，并在 API Center 开启 Local HTTP API。然后在仓库根目录运行：
|
||||
|
||||
```bash
|
||||
pnpm poc:query-agent "我和BOBO第一次聊了什么"
|
||||
```
|
||||
|
||||
## 两个入口
|
||||
|
||||
| 命令 | 行为 | 何时用 |
|
||||
| --- | --- | --- |
|
||||
| `pnpm poc:query-agent "问题"` | 先执行完整构建，再运行 | 首次运行，或刚改过代码 |
|
||||
| `pnpm poc:query-agent:run "问题"` | 直接运行已有构建，**不构建** | 连续迭代测试 |
|
||||
|
||||
`poc:query-agent:run` 在构建产物不存在时会明确提示先运行 `pnpm poc:query-agent`，**不会自动构建**。
|
||||
|
||||
注意：`poc:query-agent` 内部走的是完整 `electron-vite build`（main + preload + renderer），
|
||||
即使 POC 只需要一个 main entry。连续测试请使用 `poc:query-agent:run` 以免每次都重建整个 renderer。
|
||||
|
||||
## 传参
|
||||
|
||||
参数按原样转发给入口，可以被 `--` 分隔（`pnpm run` 惯例）：
|
||||
|
||||
```bash
|
||||
pnpm poc:query-agent -- "BOBO上个月有没有给我发过文件"
|
||||
pnpm poc:query-agent:run "BOBO上个月有没有给我发过文件"
|
||||
```
|
||||
|
||||
入口只会移除参数列表**开头**的一个独立 `--`；问题正文中的 `--` 会原样保留。
|
||||
|
||||
## Provider
|
||||
|
||||
POC 使用设置页当前默认 AI Provider、模型、Base URL 和安全存储中的 API Key。Local Query API 仍使用现有 Bearer Token；POC 输出不会打印 Token、API Key、数据库路径或内部消息 ID。
|
||||
|
||||
## 输出
|
||||
|
||||
**stdout 是 JSON**（`poc:query-agent` 会在它前面混入构建日志，`poc:query-agent:run` 只多两行 pnpm 横幅）。
|
||||
需要机器解析时用 `--silent` 拿到纯 JSON：
|
||||
|
||||
```bash
|
||||
pnpm --silent poc:query-agent:run "我和BOBO第一次聊了什么" > result.json
|
||||
```
|
||||
|
||||
JSON 字段：
|
||||
|
||||
- `question`、`provider`、`model`
|
||||
- `modelCallCount`、`toolCallCount`
|
||||
- `modelDurationsMs`（每次模型调用耗时，含失败的那次）
|
||||
- `modelDiagnostics`（每次模型调用的请求级诊断：HTTP status、content-type、是否返回 HTML、是否超时、耗时）
|
||||
- `firstModelMs`、`toolTotalMs`、`finalModelMs`、`totalMs`
|
||||
- 每次工具调用的名称、脱敏参数、耗时、状态和结果数量
|
||||
- 最终 `answer` 或错误信息
|
||||
|
||||
**stderr 是人类可读摘要**（不参与 JSON 解析）：
|
||||
|
||||
```text
|
||||
[Timing]
|
||||
Model #1 1315 ms
|
||||
TM Tools(1) 623 ms
|
||||
Model #2 1598 ms
|
||||
------------------------
|
||||
Total 3545 ms
|
||||
Model total 2913 ms (82.2%)
|
||||
TM tool total 623 ms (17.6%)
|
||||
|
||||
[Provider]
|
||||
provider DeepSeek
|
||||
model DeepSeek Chat
|
||||
host api.deepseek.com
|
||||
model calls 2
|
||||
tool calls 1
|
||||
attempt #1 elapsedMs=1298 status=200 contentType=application/json
|
||||
attempt #2 elapsedMs=1571 status=200 contentType=application/json
|
||||
```
|
||||
|
||||
`elapsedMs` 是 TTFB（收到响应头），`modelDurationsMs` 是整次调用（含读 body）；502 时两者接近，
|
||||
说明等待发生在上游网关，不是本地读 body 慢。诊断只记录 host，不记录完整 URL 或任何凭据。非 2xx 响应会先记录 status / content-type / elapsedMs，再返回安全错误（例如“模型服务返回了网页而不是 JSON（HTTP 502 Bad Gateway）”），不会把 HTML 正文丢给 JSON 解析器。
|
||||
|
||||
## 约束
|
||||
|
||||
工具调用最多 5 次，只允许 `query_messages`、`search_messages`、`message_context`、`conversation_overview`。未配置 AI Provider、Local Query API 未启动或当前 Provider 协议不支持 tools 时，POC 会直接返回错误，不会回退到另一套模型配置。
|
||||
+2
-1
@@ -41,7 +41,8 @@
|
||||
"predev": "node -e \"require('electron')\"",
|
||||
"dev": "node scripts/ensure-env.cjs && node scripts/build-wechat-connector.cjs && electron-vite dev",
|
||||
"dev:update": "cross-env TRACEMEMO_UPDATE_SIMULATION=true pnpm dev",
|
||||
"poc:query-agent": "electron-vite build && electron out/main/queryAgentPoc.js",
|
||||
"poc:query-agent": "electron-vite build && node scripts/run-query-agent-poc.cjs",
|
||||
"poc:query-agent:run": "node scripts/run-query-agent-poc.cjs",
|
||||
"test:wechat-connector": "go -C services/wechat-connector test ./... && go -C services/wechat-connector vet ./...",
|
||||
"test:unit": "vitest run --config vitest.unit.config.ts",
|
||||
"test:component": "vitest run --config vitest.component.config.ts",
|
||||
|
||||
@@ -0,0 +1,54 @@
|
||||
#!/usr/bin/env node
|
||||
/**
|
||||
* Query Agent POC 快速运行入口：直接执行已构建的 out/main/queryAgentPoc.js，不做任何构建。
|
||||
*
|
||||
* 与 `pnpm poc:query-agent` 的分工：
|
||||
* - poc:query-agent : 先 electron-vite build，再运行（代码改动后使用）
|
||||
* - poc:query-agent:run : 只运行现有构建产物（连续测试使用）
|
||||
*
|
||||
* 若构建产物不存在，给出明确提示；不会偷偷触发 full build，否则 fast-run 失去意义。
|
||||
*
|
||||
* 可选环境变量：
|
||||
* TRACEMEMO_POC_ELECTRON 指定 Electron 可执行文件（默认取 node_modules 中的 electron）
|
||||
*/
|
||||
const fs = require('fs')
|
||||
const path = require('path')
|
||||
const { spawnSync } = require('child_process')
|
||||
|
||||
const repoRoot = path.resolve(__dirname, '..')
|
||||
const pocEntry = path.join(repoRoot, 'out', 'main', 'queryAgentPoc.js')
|
||||
|
||||
if (!fs.existsSync(pocEntry)) {
|
||||
process.stderr.write(
|
||||
[
|
||||
'',
|
||||
'[poc] POC build 不存在，请先运行：',
|
||||
' pnpm poc:query-agent "你的问题"',
|
||||
'',
|
||||
` 预期构建产物：${path.relative(repoRoot, pocEntry)}`,
|
||||
' （本入口有意不自动构建，以免失去快速运行的意义）',
|
||||
''
|
||||
].join('\n')
|
||||
)
|
||||
process.exit(1)
|
||||
}
|
||||
|
||||
// 该入口的目标就是启动 Electron 主进程，因此必须清掉会让 Electron 退化成纯 Node 的标记。
|
||||
// 部分 IDE 集成终端会注入 ELECTRON_RUN_AS_NODE=1。
|
||||
const env = { ...process.env }
|
||||
delete env.ELECTRON_RUN_AS_NODE
|
||||
|
||||
// 在普通 Node 中 require('electron') 返回可执行文件路径。
|
||||
const electronBinary = env.TRACEMEMO_POC_ELECTRON || require('electron')
|
||||
|
||||
const result = spawnSync(electronBinary, [pocEntry, ...process.argv.slice(2)], {
|
||||
stdio: 'inherit',
|
||||
cwd: repoRoot,
|
||||
env
|
||||
})
|
||||
|
||||
if (result.error) {
|
||||
process.stderr.write(`[poc] 启动 Electron 失败：${result.error.message}\n`)
|
||||
process.exit(1)
|
||||
}
|
||||
process.exit(typeof result.status === 'number' ? result.status : 1)
|
||||
@@ -0,0 +1,13 @@
|
||||
/**
|
||||
* Query Agent POC 的 CLI 参数解析。
|
||||
*
|
||||
* 单独抽成纯函数，便于单测；避免 entry 文件的副作用（app bootstrap / app.whenReady）影响测试。
|
||||
*/
|
||||
export function parsePocQuestion(argv: readonly string[]): string {
|
||||
const args = [...argv]
|
||||
// npm / pnpm 在复合 script（`build && electron ...`）里会把 `--` 一并追加到命令末尾，
|
||||
// 于是 separator 会落到 POC 的 argv 里，污染问题正文。
|
||||
// 只移除开头的这一个独立 separator；问题正文中间的合法 `--` 必须保留。
|
||||
if (args[0] === '--') args.shift()
|
||||
return args.join(' ').trim()
|
||||
}
|
||||
@@ -3,9 +3,11 @@ import { app } from 'electron'
|
||||
import { apiTokenStore } from './api-token-store'
|
||||
import { AIProviderService } from './services/ai-provider-service'
|
||||
import { QueryAgentPocService, type QueryAgentToolResult } from './services/query-agent-poc-service'
|
||||
import { parsePocQuestion } from './query-agent-poc-cli'
|
||||
import { formatProviderDiagnostics, formatTiming } from './query-agent-poc-report'
|
||||
|
||||
const baseUrl = (process.env.TRACEMEMO_QUERY_API_BASE || 'http://127.0.0.1:6131/api/v1').replace(/\/+$/, '')
|
||||
const question = process.argv.slice(2).join(' ').trim()
|
||||
const question = parsePocQuestion(process.argv.slice(2))
|
||||
|
||||
async function callQueryApi(name: string, input: Record<string, unknown>): Promise<QueryAgentToolResult> {
|
||||
const paths: Record<string, string> = {
|
||||
@@ -26,9 +28,13 @@ async function callQueryApi(name: string, input: Record<string, unknown>): Promi
|
||||
|
||||
async function main(): Promise<void> {
|
||||
await app.whenReady()
|
||||
const service = new QueryAgentPocService(new AIProviderService(), callQueryApi)
|
||||
const provider = new AIProviderService()
|
||||
const service = new QueryAgentPocService(provider, callQueryApi)
|
||||
const result = await service.run(question)
|
||||
process.stdout.write(`${JSON.stringify(result, null, 2)}\n`)
|
||||
// 诊断摘要写 stderr，保持 stdout 仍是纯 JSON，方便管道与脚本消费。
|
||||
process.stderr.write(formatTiming(result))
|
||||
process.stderr.write(formatProviderDiagnostics(result, provider.getRuntimeEndpointHost()))
|
||||
app.quit()
|
||||
}
|
||||
|
||||
|
||||
@@ -0,0 +1,78 @@
|
||||
import type { QueryAgentPocResult } from './services/query-agent-poc-service'
|
||||
|
||||
const WIDTH = 24
|
||||
|
||||
function row(label: string, value: string): string {
|
||||
return `${label.padEnd(14)}${value.padStart(10)}`
|
||||
}
|
||||
|
||||
/**
|
||||
* 人类可读的耗时摘要。
|
||||
*
|
||||
* 只做展示，不改变 JSON 输出契约。需要能正确处理：
|
||||
* - 0 次模型调用（配置缺失等提前返回）
|
||||
* - 1 次模型调用 + 0 次工具
|
||||
* - 2 次模型调用 + N 次工具（此时可准确还原"模型 → 工具 → 模型"顺序）
|
||||
* - 3 次以上（无法逐段归属工具耗时时，明确标注为聚合）
|
||||
* - 首次模型调用失败 / 末尾模型调用失败
|
||||
*/
|
||||
export function formatTiming(result: QueryAgentPocResult): string {
|
||||
const durations = result.modelDurationsMs || []
|
||||
const toolTotalMs = result.toolTotalMs || 0
|
||||
const toolCount = result.toolCallCount || 0
|
||||
const lines: string[] = ['', '[Timing]']
|
||||
|
||||
if (durations.length === 0) {
|
||||
lines.push('（未发生模型调用）')
|
||||
} else if (durations.length <= 2) {
|
||||
lines.push(row('Model #1', `${durations[0]} ms`))
|
||||
if (toolCount > 0) lines.push(row(`TM Tools(${toolCount})`, `${toolTotalMs} ms`))
|
||||
if (durations[1] !== undefined) lines.push(row('Model #2', `${durations[1]} ms`))
|
||||
} else {
|
||||
durations.forEach((duration, index) => lines.push(row(`Model #${index + 1}`, `${duration} ms`)))
|
||||
lines.push(row(`TM Tools(${toolCount})`, `${toolTotalMs} ms`))
|
||||
lines.push('（3 次以上模型调用：工具耗时按总计展示）')
|
||||
}
|
||||
|
||||
const modelTotalMs = durations.reduce((sum, value) => sum + value, 0)
|
||||
const totalMs = result.totalMs || 0
|
||||
const share = (value: number): string =>
|
||||
totalMs > 0 ? `${((value / totalMs) * 100).toFixed(1)}%` : 'n/a'
|
||||
|
||||
lines.push('-'.repeat(WIDTH))
|
||||
lines.push(row('Total', `${totalMs} ms`))
|
||||
lines.push('')
|
||||
lines.push(row('Model total', `${modelTotalMs} ms (${share(modelTotalMs)})`))
|
||||
lines.push(row('TM tool total', `${toolTotalMs} ms (${share(toolTotalMs)})`))
|
||||
if (result.error) lines.push(`error: ${result.error}`)
|
||||
return `${lines.join('\n')}\n`
|
||||
}
|
||||
|
||||
/**
|
||||
* 请求级诊断。
|
||||
* 只输出 host 与状态字段；绝不输出 API key / Authorization / 完整 URL / 响应正文。
|
||||
*/
|
||||
export function formatProviderDiagnostics(result: QueryAgentPocResult, host?: string): string {
|
||||
const lines = [
|
||||
'',
|
||||
'[Provider]',
|
||||
`provider ${result.provider}`,
|
||||
`model ${result.model}`,
|
||||
`host ${host || '(unknown)'}`,
|
||||
`model calls ${result.modelCallCount}`,
|
||||
`tool calls ${result.toolCallCount}`
|
||||
]
|
||||
for (const entry of result.modelDiagnostics || []) {
|
||||
const parts = [`elapsedMs=${entry.elapsedMs}`]
|
||||
if (entry.status !== undefined) parts.push(`status=${entry.status}`)
|
||||
if (entry.contentType) parts.push(`contentType=${entry.contentType}`)
|
||||
if (entry.timedOut) parts.push('timedOut=true')
|
||||
if (entry.htmlInsteadOfJson) parts.push('htmlInsteadOfJson=true')
|
||||
if (entry.errorCode) parts.push(`code=${entry.errorCode}`)
|
||||
if (entry.errorType) parts.push(`type=${entry.errorType}`)
|
||||
if (entry.status !== undefined && [502, 503, 504].includes(entry.status)) parts.push('upstreamGateway=true')
|
||||
lines.push(`attempt #${entry.index} ${parts.join(' ')}`)
|
||||
if (entry.error) lines.push(` error: ${entry.error}`)
|
||||
}
|
||||
return `${lines.join('\n')}\n`
|
||||
}
|
||||
@@ -86,10 +86,16 @@ export class AIProviderRequestError extends Error {
|
||||
readonly code?: string
|
||||
readonly type?: string
|
||||
readonly responseBody?: unknown
|
||||
/** 诊断字段：上游响应 content-type（例如 502 返回的 text/html） */
|
||||
readonly contentType?: string
|
||||
/** 诊断字段：本次请求从发出到响应处理结束的耗时 */
|
||||
readonly elapsedMs?: number
|
||||
/** 诊断字段：上游返回网页而不是 JSON */
|
||||
readonly htmlInsteadOfJson?: boolean
|
||||
|
||||
constructor(
|
||||
message: string,
|
||||
details: { status?: number; code?: unknown; type?: unknown; responseBody?: unknown } = {}
|
||||
details: { status?: number; code?: unknown; type?: unknown; responseBody?: unknown; contentType?: string; elapsedMs?: number; htmlInsteadOfJson?: boolean } = {}
|
||||
) {
|
||||
super(message)
|
||||
this.name = 'AIProviderRequestError'
|
||||
@@ -97,6 +103,9 @@ export class AIProviderRequestError extends Error {
|
||||
this.code = typeof details.code === 'string' ? details.code : undefined
|
||||
this.type = typeof details.type === 'string' ? details.type : undefined
|
||||
this.responseBody = details.responseBody
|
||||
this.contentType = details.contentType
|
||||
this.elapsedMs = details.elapsedMs
|
||||
this.htmlInsteadOfJson = details.htmlInsteadOfJson
|
||||
}
|
||||
}
|
||||
|
||||
@@ -136,6 +145,19 @@ export class AIProviderService {
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* 诊断用：当前默认 Provider 的 endpoint host。
|
||||
* 只返回 hostname（不含路径与 query），避免把凭据或敏感 query 带进日志/终端。
|
||||
*/
|
||||
getRuntimeEndpointHost(): string | undefined {
|
||||
try {
|
||||
const resolved = this.resolveProvider(undefined)
|
||||
return endpointHost(resolved.provider.baseUrl)
|
||||
} catch {
|
||||
return undefined
|
||||
}
|
||||
}
|
||||
|
||||
getVisionRuntimeConfig(): AIVisionRuntimeConfig {
|
||||
const result = this.list()
|
||||
const defaultProvider = result.providers.find((item) => item.id === result.defaultProviderId)
|
||||
@@ -324,6 +346,13 @@ export class AIProviderService {
|
||||
toolCalls?: AIChatToolCall[]
|
||||
usage?: { input?: number; output?: number; total?: number; estimated?: boolean }
|
||||
error?: string
|
||||
errorStatus?: number
|
||||
errorCode?: string
|
||||
errorType?: string
|
||||
errorContentType?: string
|
||||
elapsedMs?: number
|
||||
timedOut?: boolean
|
||||
htmlInsteadOfJson?: boolean
|
||||
}> {
|
||||
try {
|
||||
if (options?.apiKey) throw new Error('Tool Calling 不支持 legacy provider 配置')
|
||||
@@ -342,7 +371,22 @@ export class AIProviderService {
|
||||
return { success: true, ...result }
|
||||
} catch (error) {
|
||||
if (signal?.aborted) throw error
|
||||
return { success: false, error: safeAIError(error) }
|
||||
const timedOut = error instanceof DOMException && error.name === 'TimeoutError'
|
||||
return {
|
||||
success: false,
|
||||
error: safeAIError(error),
|
||||
timedOut,
|
||||
htmlInsteadOfJson: error instanceof AIProviderRequestError ? Boolean(error.htmlInsteadOfJson) : false,
|
||||
...(error instanceof AIProviderRequestError
|
||||
? {
|
||||
...(error.status !== undefined ? { errorStatus: error.status } : {}),
|
||||
...(error.code ? { errorCode: error.code } : {}),
|
||||
...(error.type ? { errorType: error.type } : {}),
|
||||
...(error.contentType ? { errorContentType: error.contentType } : {}),
|
||||
...(error.elapsedMs !== undefined ? { elapsedMs: error.elapsedMs } : {})
|
||||
}
|
||||
: {})
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -598,6 +642,15 @@ function deepSeekProvider(baseUrl?: string, model?: string): AIProviderSummary {
|
||||
}
|
||||
}
|
||||
|
||||
/** 诊断用：只取 endpoint 的 hostname，不保留路径与 query。 */
|
||||
function endpointHost(baseUrl: string): string | undefined {
|
||||
try {
|
||||
return new URL(baseUrl).hostname || undefined
|
||||
} catch {
|
||||
return undefined
|
||||
}
|
||||
}
|
||||
|
||||
function stripRuntimeFields(
|
||||
provider: AIProviderSummary
|
||||
): Omit<AIProviderSummary, 'hasApiKey' | 'isDefault'> {
|
||||
@@ -912,7 +965,7 @@ async function requestOpenAICompatibleWithTools(
|
||||
messages: Array<Record<string, unknown>>,
|
||||
tools: AIChatToolDefinition[],
|
||||
signal?: AbortSignal
|
||||
): Promise<{ data: string; toolCalls: AIChatToolCall[]; usage?: AIRequestResult['usage'] }> {
|
||||
): Promise<{ data: string; toolCalls: AIChatToolCall[]; usage?: AIRequestResult['usage']; elapsedMs?: number }> {
|
||||
const endpoint = provider.baseUrl.endsWith('/chat/completions')
|
||||
? provider.baseUrl
|
||||
: `${provider.baseUrl.replace(/\/+$/, '')}/chat/completions`
|
||||
@@ -932,16 +985,33 @@ async function requestOpenAICompatibleWithTools(
|
||||
},
|
||||
provider.advanced.timeoutMs,
|
||||
signal,
|
||||
async (response) => {
|
||||
const payload = await parseJsonResponse<OpenAIResponsePayload>(response)
|
||||
if (!response.ok) {
|
||||
throw new AIProviderRequestError(payload.error?.message || `AI 请求失败 (${response.status})`, {
|
||||
status: response.status,
|
||||
code: payload.error?.code,
|
||||
type: payload.error?.type,
|
||||
responseBody: payload.error
|
||||
})
|
||||
async (response, timing) => {
|
||||
// 先尝试解析 body：非 2xx 时上游可能返回 JSON 错误体，也可能是 HTML 错误页。
|
||||
// 无论哪种，都要把 status / content-type / elapsed 带进错误，避免丢失上下文。
|
||||
let payload: OpenAIResponsePayload | undefined
|
||||
let parseError: unknown
|
||||
try {
|
||||
payload = await parseJsonResponse<OpenAIResponsePayload>(response)
|
||||
} catch (error) {
|
||||
parseError = error
|
||||
}
|
||||
if (!response.ok) {
|
||||
throw new AIProviderRequestError(
|
||||
payload?.error?.message || (parseError ? safeAIError(parseError) : `AI 请求失败 (${response.status})`),
|
||||
{
|
||||
status: timing.status,
|
||||
contentType: timing.contentType,
|
||||
elapsedMs: timing.elapsedMs,
|
||||
htmlInsteadOfJson: isHtmlParseError(parseError),
|
||||
code: payload?.error?.code,
|
||||
type: payload?.error?.type,
|
||||
responseBody: payload?.error
|
||||
}
|
||||
)
|
||||
}
|
||||
// 2xx 但 body 无法解析：抛出原始解析错误（含清晰的中文提示）。
|
||||
if (!payload) throw parseError ?? new Error('AI 响应格式异常')
|
||||
if (parseError) throw parseError
|
||||
const message = payload.choices?.[0]?.message
|
||||
return {
|
||||
data: openAIMessageText(message?.content),
|
||||
@@ -950,7 +1020,8 @@ async function requestOpenAICompatibleWithTools(
|
||||
if (!name) return []
|
||||
return [{ id: call.id || `tool-call-${index + 1}`, name, arguments: call.function?.arguments || '{}' }]
|
||||
}),
|
||||
usage: toOpenAIUsage(payload.usage)
|
||||
usage: toOpenAIUsage(payload.usage),
|
||||
elapsedMs: timing.elapsedMs
|
||||
}
|
||||
}
|
||||
)
|
||||
@@ -1029,12 +1100,19 @@ async function requestAnthropic(
|
||||
)
|
||||
}
|
||||
|
||||
/** 请求级诊断：不包含任何凭据，仅用于区分「上游慢 / 上游错误 / 本地解析」。 */
|
||||
interface AIRequestTiming {
|
||||
elapsedMs: number
|
||||
status: number
|
||||
contentType: string
|
||||
}
|
||||
|
||||
async function fetchWithTimeout<T>(
|
||||
url: string,
|
||||
init: RequestInit,
|
||||
timeoutMs: number,
|
||||
signal: AbortSignal | undefined,
|
||||
consume: (response: Response) => Promise<T>
|
||||
consume: (response: Response, timing: AIRequestTiming) => Promise<T>
|
||||
): Promise<T> {
|
||||
const controller = new AbortController()
|
||||
let timedOut = false
|
||||
@@ -1049,9 +1127,16 @@ async function fetchWithTimeout<T>(
|
||||
},
|
||||
Math.max(1_000, timeoutMs || 120_000)
|
||||
)
|
||||
const startedAt = Date.now()
|
||||
let responseReceivedAt = startedAt
|
||||
try {
|
||||
const response = await fetch(url, { ...init, signal: controller.signal })
|
||||
return await consume(response)
|
||||
responseReceivedAt = Date.now()
|
||||
return await consume(response, {
|
||||
elapsedMs: responseReceivedAt - startedAt,
|
||||
status: response.status,
|
||||
contentType: response.headers.get('content-type') || ''
|
||||
})
|
||||
} catch (error) {
|
||||
if (signal?.aborted) throw new DOMException('AI request cancelled', 'AbortError')
|
||||
if (timedOut) throw new DOMException('AI request timed out', 'TimeoutError')
|
||||
@@ -1295,6 +1380,11 @@ function safeAIError(error: unknown): string {
|
||||
return message.replace(/sk-[a-z0-9_-]+/gi, '***').slice(0, 300)
|
||||
}
|
||||
|
||||
/** 判断解析失败是否因为上游返回了 HTML（例如网关 502 错误页）。 */
|
||||
function isHtmlParseError(error: unknown): boolean {
|
||||
return error instanceof Error && error.message.includes('网页而不是 JSON')
|
||||
}
|
||||
|
||||
function aiProviderErrorDetails(error: unknown): {
|
||||
errorCode?: string
|
||||
errorStatus?: number
|
||||
|
||||
@@ -3,6 +3,14 @@ import { LOCAL_QUERY_TOOL_DEFINITIONS } from '../../shared/local-query-api'
|
||||
|
||||
const MAX_TOOL_CALLS = 5
|
||||
const FORBIDDEN_INPUT_KEYS = new Set(['apiKey', 'authorization', 'token', 'databasePath', 'sql', 'wxid', 'md5'])
|
||||
// 每个工具在“首次执行但结果为 0”之后允许的额外重试次数上限。
|
||||
const ZERO_RESULT_RETRY_LIMIT = 1
|
||||
// absolute 时间契约：LLM 只能给带时区的 ISO-8601 字符串，Host 负责换算成 Local Query API 的 epoch seconds。
|
||||
const ISO_ABSOLUTE_PATTERN = /^(\d{4})-(\d{2})-(\d{2})T(\d{2}):(\d{2}):(\d{2})(?:\.\d{1,3})?(Z|[+-]\d{2}:\d{2})$/
|
||||
// sanity 窗口：聊天记录不可能早于 2000 年，也不允许查询明显属于未来的区间。
|
||||
const ABSOLUTE_MIN_MS = Date.UTC(2000, 0, 1)
|
||||
const ABSOLUTE_MAX_FUTURE_MS = 366 * 24 * 60 * 60 * 1000
|
||||
const ISO_ABSOLUTE_HINT = '带时区偏移的 ISO-8601，例如 2026-08-01T00:00:00+08:00 或 2026-07-31T16:00:00Z'
|
||||
|
||||
interface ToolSchema {
|
||||
type?: string
|
||||
@@ -33,7 +41,21 @@ export interface QueryAgentProvider {
|
||||
chatWithTools(
|
||||
messages: Array<Record<string, unknown>>,
|
||||
tools: AIChatToolDefinition[]
|
||||
): Promise<{ success: boolean; data?: string; toolCalls?: AIChatToolCall[]; usage?: { input?: number; output?: number; total?: number; estimated?: boolean }; error?: string }>
|
||||
): Promise<{
|
||||
success: boolean
|
||||
data?: string
|
||||
toolCalls?: AIChatToolCall[]
|
||||
usage?: { input?: number; output?: number; total?: number; estimated?: boolean }
|
||||
error?: string
|
||||
/** 以下为诊断字段（additive，不参与业务语义） */
|
||||
elapsedMs?: number
|
||||
errorStatus?: number
|
||||
errorCode?: string
|
||||
errorType?: string
|
||||
errorContentType?: string
|
||||
timedOut?: boolean
|
||||
htmlInsteadOfJson?: boolean
|
||||
}>
|
||||
}
|
||||
|
||||
export interface QueryAgentToolResult {
|
||||
@@ -50,6 +72,18 @@ export interface QueryAgentTraceItem {
|
||||
evidenceCount?: number
|
||||
}
|
||||
|
||||
export interface QueryAgentModelCallDiagnostic {
|
||||
index: number
|
||||
elapsedMs: number
|
||||
status?: number
|
||||
contentType?: string
|
||||
timedOut?: boolean
|
||||
htmlInsteadOfJson?: boolean
|
||||
errorCode?: string
|
||||
errorType?: string
|
||||
error?: string
|
||||
}
|
||||
|
||||
export interface QueryAgentPocResult {
|
||||
question: string
|
||||
provider: string
|
||||
@@ -60,6 +94,10 @@ export interface QueryAgentPocResult {
|
||||
toolTotalMs: number
|
||||
finalModelMs?: number
|
||||
totalMs: number
|
||||
/** 每次模型调用的耗时（含失败的那次），按调用顺序；additive 诊断字段 */
|
||||
modelDurationsMs: number[]
|
||||
/** 每次模型调用的请求级诊断；additive 诊断字段 */
|
||||
modelDiagnostics: QueryAgentModelCallDiagnostic[]
|
||||
traces: QueryAgentTraceItem[]
|
||||
answer?: string
|
||||
error?: string
|
||||
@@ -74,11 +112,12 @@ const SYSTEM_PROMPT = `你是 TraceMemo 的本地聊天查询助手，只能使
|
||||
|
||||
规划原则：
|
||||
- 先判断问题需要哪种证据，再调用最少的 Tool。每次收到 Tool Result 后都判断“当前 Evidence 是否已经足以给出有边界的回答”；足够就立即回答，不为追求绝对完整继续调查。
|
||||
- query_messages 是精确事实查询，适用于能用联系人、时间、方向、消息类型、顺序等结构条件表达的问题。earliest/latest 等时间边界也是结构条件，必须使用 order 与 limit 精确查询，不能使用抽样 overview。结果已经回答问题时，不要追加 conversation_overview。
|
||||
- search_messages 是语义 Evidence 检索，适用于结构条件无法确定答案的问题。第一次尽量在一个调用中给出高质量 query 和最多 4 个 variants；调用前确认 variants 数组长度不超过 4。一次有效 search 后不得再次 search，已有相关 Evidence 就直接判断。
|
||||
- conversation_overview 只用于真正需要理解一个时间范围内整体聊了什么、主要话题或整体互动的 broad summary。它返回 temporal coverage sample，不代表完整聊天，也不是语义检索不足时的默认 fallback。
|
||||
- query_messages 是精确事实查询，适用于能用联系人、时间、方向、消息类型、顺序等结构条件表达的问题。earliest/latest 等时间边界也是结构条件，必须使用 order 与 limit 精确查询，不能使用抽样 overview。结果已经回答问题时，不要追加 conversation_overview。若返回 0 条且你判断是时间范围或结构条件不合适，允许再查一次并合理扩大或更换条件，但必须与上一次实质不同。
|
||||
- 需要绝对时间范围时，startTime/endTime 必须使用带时区偏移的 ISO-8601 字符串（例如 2026-08-01T00:00:00+08:00 或 2026-07-31T16:00:00Z）。不要传 epoch 数字，也不要传没有时区的裸本地时间。
|
||||
- search_messages 是关键词检索，适用于结构条件无法确定答案的问题。queries 的每一项都是一次独立的字面检索：一项只放一个简短关键词，不要把多个近义词或整句话塞进同一项。首次最多 4 项。检索到 Evidence 后直接判断；只有本次完全没有 Evidence 时，才允许再检索一次，且每一项都必须与上一次实质不同。
|
||||
- conversation_overview 只用于真正需要理解一个时间范围内整体聊了什么、主要话题或整体互动的 broad summary。它返回 temporal coverage sample，不代表完整聊天，也不是检索不足时的默认 fallback。
|
||||
- message_context 只用于已经找到一条有价值 Evidence、但单条内容缺少前后语境而无法判断真实含义的情况。不要把它当作默认确认步骤；上下文足够后立即回答。
|
||||
- 普通聊天查询不是 exhaustive investigation。经过合理的 search 或可选 context 仍不足以形成强结论时，直接说明证据范围和不确定性，不要循环调用 search、overview、context。
|
||||
- 普通聊天查询不是 exhaustive investigation。经过合理的检索或可选 context 仍不足以形成强结论时，直接说明证据范围和不确定性，不要循环调用 search、overview、context。
|
||||
|
||||
事实边界：不得编造未返回的消息、猜测联系人、修改 resolvedTimeRange，或把 partial/unknown 当作 complete。coverage complete 且结果为 0 时，可以说明当前可读取的完整范围没有找到；coverage partial/unknown 且结果为 0 时，必须说明无法确认绝对不存在。不要把 sampled Evidence 当作完整聊天，也不要把 source message count 和 selected evidence count 混为一谈。
|
||||
缺少必要信息时用自然语言澄清；超出工具能力时说明不能可靠完成，并给出当前工具可以执行的替代方向。`
|
||||
@@ -187,17 +226,147 @@ function validateSchema(value: unknown, schema: ToolSchema, field = '$'): ToolAr
|
||||
return undefined
|
||||
}
|
||||
|
||||
export function validateToolArguments(name: string, value: unknown): { input?: Record<string, unknown>; error?: ToolArgumentValidationError } {
|
||||
function argError(field: string, constraint: string, expected?: unknown, actual?: unknown, hint?: string): ToolArgumentValidationError {
|
||||
return { status: 'invalid_tool_arguments', field, constraint, ...(expected === undefined ? {} : { expected }), ...(actual === undefined ? {} : { actual }), ...(hint ? { hint } : {}) }
|
||||
}
|
||||
|
||||
/**
|
||||
* 解析 LLM 提供的绝对时间。只接受带显式时区偏移（或 Z）的 ISO-8601；
|
||||
* 无时区、epoch 数字、非法日历日一律返回 undefined，由调用方转成可修正的 invalid_tool_arguments。
|
||||
*/
|
||||
function parseIsoInstant(value: string): number | undefined {
|
||||
const match = ISO_ABSOLUTE_PATTERN.exec(value.trim())
|
||||
if (!match) return undefined
|
||||
const year = Number(match[1])
|
||||
const month = Number(match[2])
|
||||
const day = Number(match[3])
|
||||
const hour = Number(match[4])
|
||||
const minute = Number(match[5])
|
||||
const second = Number(match[6])
|
||||
if (month < 1 || month > 12 || day < 1 || day > 31 || hour > 23 || minute > 59 || second > 59) return undefined
|
||||
// 先在 UTC 语义下校验字面日期真实存在，再套用时区偏移，避免 2 月 31 日被静默进位。
|
||||
const naiveMs = Date.UTC(year, month - 1, day, hour, minute, second)
|
||||
const naive = new Date(naiveMs)
|
||||
if (naive.getUTCFullYear() !== year || naive.getUTCMonth() !== month - 1 || naive.getUTCDate() !== day) return undefined
|
||||
const offset = match[7]
|
||||
if (offset === 'Z') return naiveMs
|
||||
const sign = offset.startsWith('-') ? -1 : 1
|
||||
const offsetHour = Number(offset.slice(1, 3))
|
||||
const offsetMinute = Number(offset.slice(4, 6))
|
||||
if (offsetHour > 23 || offsetMinute > 59) return undefined
|
||||
return naiveMs - sign * (offsetHour * 60 + offsetMinute) * 60000
|
||||
}
|
||||
|
||||
function canonicalizeTimeRange(timeRange: Record<string, unknown>, now: Date): { value?: Record<string, unknown>; error?: ToolArgumentValidationError } {
|
||||
const startTime = timeRange.startTime
|
||||
const endTime = timeRange.endTime
|
||||
if (timeRange.kind !== 'absolute') {
|
||||
// 非 absolute 语义下 startTime/endTime 无意义，直接丢弃，避免残留数值进入 Query API。
|
||||
const { startTime: _startTime, endTime: _endTime, ...rest } = timeRange
|
||||
return { value: rest }
|
||||
}
|
||||
if (typeof startTime !== 'string' || typeof endTime !== 'string') {
|
||||
return { error: argError('timeRange.startTime', 'required', ISO_ABSOLUTE_HINT, actualType(startTime ?? endTime), ISO_ABSOLUTE_HINT) }
|
||||
}
|
||||
const startMs = parseIsoInstant(startTime)
|
||||
if (startMs === undefined) return { error: argError('timeRange.startTime', 'format', ISO_ABSOLUTE_HINT, startTime, ISO_ABSOLUTE_HINT) }
|
||||
const endMs = parseIsoInstant(endTime)
|
||||
if (endMs === undefined) return { error: argError('timeRange.endTime', 'format', ISO_ABSOLUTE_HINT, endTime, ISO_ABSOLUTE_HINT) }
|
||||
if (endMs < startMs) return { error: argError('timeRange.endTime', 'range_order', 'endTime 不得早于 startTime', endTime) }
|
||||
const latestMs = now.getTime() + ABSOLUTE_MAX_FUTURE_MS
|
||||
const sanityWindow = `2000-01-01 至 ${new Date(latestMs).toISOString()}`
|
||||
if (startMs < ABSOLUTE_MIN_MS || startMs > latestMs) return { error: argError('timeRange.startTime', 'range_sanity', sanityWindow, startTime) }
|
||||
if (endMs < ABSOLUTE_MIN_MS || endMs > latestMs) return { error: argError('timeRange.endTime', 'range_sanity', sanityWindow, endTime) }
|
||||
// 通过全部校验后才换算成 Local Query API 使用的 epoch seconds。
|
||||
return { value: { kind: 'absolute', startTime: Math.floor(startMs / 1000), endTime: Math.floor(endMs / 1000) } }
|
||||
}
|
||||
|
||||
/**
|
||||
* LLM Tool Adapter 的 canonicalization：
|
||||
* - absolute 的 ISO-8601 → Local Query API 的 epoch seconds
|
||||
* - search_messages 的 queries[] → Local Query API 的 query + variants
|
||||
*/
|
||||
function canonicalizeToolInput(name: string, input: Record<string, unknown>, now: Date): { input?: Record<string, unknown>; error?: ToolArgumentValidationError } {
|
||||
const output: Record<string, unknown> = { ...input }
|
||||
if (output.timeRange && typeof output.timeRange === 'object' && !Array.isArray(output.timeRange)) {
|
||||
const canonical = canonicalizeTimeRange(output.timeRange as Record<string, unknown>, now)
|
||||
if (canonical.error) return { error: canonical.error }
|
||||
output.timeRange = canonical.value
|
||||
}
|
||||
if (name === 'search_messages') {
|
||||
const raw = Array.isArray(output.queries) ? (output.queries as unknown[]) : []
|
||||
const probes = raw.filter((value): value is string => typeof value === 'string').map((value) => value.trim()).filter(Boolean)
|
||||
if (!probes.length) return { error: argError('queries', 'required', '至少一个非空检索项') }
|
||||
const [first, ...rest] = probes
|
||||
delete output.queries
|
||||
output.query = first
|
||||
if (rest.length) output.variants = rest
|
||||
}
|
||||
return { input: output }
|
||||
}
|
||||
|
||||
interface ZeroResultRetryState {
|
||||
searchAttempts: number
|
||||
searchSignatures: string[]
|
||||
queryAttempts: number
|
||||
querySignatures: string[]
|
||||
}
|
||||
|
||||
function newRetryState(): ZeroResultRetryState {
|
||||
return { searchAttempts: 0, searchSignatures: [], queryAttempts: 0, querySignatures: [] }
|
||||
}
|
||||
|
||||
function normalizedTarget(input: Record<string, unknown>): string {
|
||||
const target = input.target
|
||||
const query = target && typeof target === 'object' ? (target as Record<string, unknown>).query : undefined
|
||||
return typeof query === 'string' ? query.trim().toLowerCase() : ''
|
||||
}
|
||||
|
||||
/** 只覆盖“实质条件”：忽略 limit/order/excludeSystem 这类不改变检索语义的字段。 */
|
||||
function retrySignature(name: string, input: Record<string, unknown>): string {
|
||||
if (name === 'search_messages') {
|
||||
const probes = [input.query, ...(Array.isArray(input.variants) ? input.variants : [])]
|
||||
.filter((value): value is string => typeof value === 'string')
|
||||
.map((value) => value.trim().toLowerCase())
|
||||
return JSON.stringify({ target: normalizedTarget(input), timeRange: input.timeRange ?? null, probes: Array.from(new Set(probes)).sort() })
|
||||
}
|
||||
const messageTypes = Array.isArray(input.messageTypes) ? [...(input.messageTypes as string[])].sort() : []
|
||||
return JSON.stringify({ target: normalizedTarget(input), timeRange: input.timeRange ?? null, direction: input.direction ?? null, messageTypes })
|
||||
}
|
||||
|
||||
function duplicateRetry(name: string, input: Record<string, unknown>, state: ZeroResultRetryState): boolean {
|
||||
const signature = retrySignature(name, input)
|
||||
if (name === 'search_messages') return state.searchSignatures.includes(signature)
|
||||
if (name === 'query_messages') return state.querySignatures.includes(signature)
|
||||
return false
|
||||
}
|
||||
|
||||
function recordAttempt(name: string, input: Record<string, unknown>, state: ZeroResultRetryState): void {
|
||||
if (name === 'search_messages') { state.searchAttempts += 1; state.searchSignatures.push(retrySignature(name, input)) }
|
||||
else if (name === 'query_messages') { state.queryAttempts += 1; state.querySignatures.push(retrySignature(name, input)) }
|
||||
}
|
||||
|
||||
function retryNote(name: string, result: QueryAgentToolResult, state: ZeroResultRetryState): string | undefined {
|
||||
if (result.constraint === 'duplicate_retry') return '本次重试的条件与上一次完全相同，已被拒绝；请改用实质不同的条件，或直接基于现有结果作答。'
|
||||
if (result.status !== 'completed') return undefined
|
||||
const counts = resultCount(result)
|
||||
if (name === 'search_messages' && !counts.evidenceCount && state.searchAttempts <= ZERO_RESULT_RETRY_LIMIT) return '本次检索没有任何 Evidence。允许再执行一次 search_messages，但每一项都必须与上一次实质不同；完全相同的检索会被拒绝。'
|
||||
if (name === 'query_messages' && counts.resultCount === 0 && state.queryAttempts <= ZERO_RESULT_RETRY_LIMIT) return '本次精确查询返回 0 条。允许再执行一次 query_messages，用于合理扩大或更换时间范围、方向或消息类型；完全相同的条件会被拒绝。'
|
||||
return undefined
|
||||
}
|
||||
|
||||
export function validateToolArguments(name: string, value: unknown, now: Date = new Date()): { input?: Record<string, unknown>; error?: ToolArgumentValidationError } {
|
||||
const definition = LOCAL_QUERY_TOOL_DEFINITIONS.find((tool) => tool.name === name)
|
||||
if (!definition) return { error: { status: 'invalid_tool_arguments', field: '$', constraint: 'tool', expected: 'supported tool', actual: name } }
|
||||
if (!definition) return { error: argError('$', 'tool', 'supported tool', name) }
|
||||
if (!value || typeof value !== 'object' || Array.isArray(value)) {
|
||||
return { error: { status: 'invalid_tool_arguments', field: '$', constraint: 'type', expected: 'object', actual: actualType(value) } }
|
||||
return { error: argError('$', 'type', 'object', actualType(value)) }
|
||||
}
|
||||
if (containsForbiddenKey(value)) {
|
||||
return { error: { status: 'invalid_tool_arguments', field: '$', constraint: 'forbidden_field' } }
|
||||
return { error: argError('$', 'forbidden_field') }
|
||||
}
|
||||
const error = validateSchema(value, definition.parameters as ToolSchema)
|
||||
return error ? { error } : { input: value as Record<string, unknown> }
|
||||
if (error) return { error }
|
||||
return canonicalizeToolInput(name, value as Record<string, unknown>, now)
|
||||
}
|
||||
|
||||
function resultCount(result: QueryAgentToolResult): { resultCount?: number; evidenceCount?: number } {
|
||||
@@ -213,7 +382,7 @@ function messageRecordForModel(value: unknown): unknown {
|
||||
return record.messageType || !record.sourceKind ? record : { ...record, messageType: record.sourceKind }
|
||||
}
|
||||
|
||||
function toolResultForModel(name: string, result: QueryAgentToolResult, callsUsed: number, nextTools: AIChatToolDefinition[]): QueryAgentToolResult {
|
||||
function toolResultForModel(name: string, result: QueryAgentToolResult, callsUsed: number, nextTools: AIChatToolDefinition[], note?: string): QueryAgentToolResult {
|
||||
const visible: QueryAgentToolResult = { ...result }
|
||||
if (Array.isArray(result.messages)) visible.messages = result.messages.map(messageRecordForModel)
|
||||
if (Array.isArray(result.evidence)) visible.evidence = result.evidence.map(messageRecordForModel)
|
||||
@@ -225,29 +394,48 @@ function toolResultForModel(name: string, result: QueryAgentToolResult, callsUse
|
||||
toolCallsUsed: callsUsed,
|
||||
toolCallsRemaining: Math.max(0, MAX_TOOL_CALLS - callsUsed),
|
||||
availableNextTools: nextTools.map((tool) => tool.function.name),
|
||||
instruction: nextTools.length
|
||||
? '先判断当前 Evidence 是否足以回答；足够就立即回答，只在含义仍有明确歧义时使用当前可用 Tool。'
|
||||
: '工具阶段已经结束。必须直接给出有边界的最终回答，不得再调用 Tool。'
|
||||
...(note ? { note } : {}),
|
||||
instruction: [
|
||||
nextTools.length
|
||||
? '先判断当前 Evidence 是否足以回答；足够就立即回答，只在含义仍有明确歧义时使用当前可用 Tool。'
|
||||
: '工具阶段已经结束。必须直接给出有边界的最终回答，不得再调用 Tool。',
|
||||
note
|
||||
].filter(Boolean).join(' ')
|
||||
}
|
||||
return visible
|
||||
}
|
||||
|
||||
function nextToolDefinitions(name: string, result: QueryAgentToolResult): AIChatToolDefinition[] {
|
||||
function nextToolDefinitions(name: string, result: QueryAgentToolResult, state: ZeroResultRetryState): AIChatToolDefinition[] {
|
||||
// 重复重试已被拒绝，不再开放工具，避免用有限的 tool budget 反复试同一条件。
|
||||
if (result.constraint === 'duplicate_retry') return []
|
||||
if (result.status === 'invalid_tool_arguments') return toolDefinition(name)
|
||||
if (name === 'search_messages' && result.status === 'completed' && resultCount(result).evidenceCount) {
|
||||
return toolDefinition('message_context')
|
||||
if (result.status !== 'completed') return []
|
||||
const counts = resultCount(result)
|
||||
if (name === 'search_messages') {
|
||||
// 有 Evidence 时保持原有高效路径：只允许补一次上下文。
|
||||
if (counts.evidenceCount) return toolDefinition('message_context')
|
||||
// 首次检索完全没有 Evidence：允许一次实质不同的重试，之后关闭。
|
||||
return state.searchAttempts <= ZERO_RESULT_RETRY_LIMIT ? toolDefinition('search_messages') : []
|
||||
}
|
||||
if (name === 'query_messages') {
|
||||
// 只有 0 结果才开放一次重试；有结果时保持原有 stopping。
|
||||
return counts.resultCount === 0 && state.queryAttempts <= ZERO_RESULT_RETRY_LIMIT ? toolDefinition('query_messages') : []
|
||||
}
|
||||
return []
|
||||
}
|
||||
|
||||
export class QueryAgentPocService {
|
||||
constructor(private readonly provider: QueryAgentProvider, private readonly executeTool: QueryAgentToolExecutor) {}
|
||||
constructor(
|
||||
private readonly provider: QueryAgentProvider,
|
||||
private readonly executeTool: QueryAgentToolExecutor,
|
||||
private readonly nowProvider: () => Date = () => new Date()
|
||||
) {}
|
||||
|
||||
async run(question: string): Promise<QueryAgentPocResult> {
|
||||
const trimmed = question.trim()
|
||||
const startedAt = Date.now()
|
||||
const runtime = this.provider.getRuntimeConfig()
|
||||
const result: QueryAgentPocResult = { question: trimmed, provider: runtime.providerName, model: runtime.modelName || runtime.model, modelCallCount: 0, toolCallCount: 0, toolTotalMs: 0, totalMs: 0, traces: [] }
|
||||
const result: QueryAgentPocResult = { question: trimmed, provider: runtime.providerName, model: runtime.modelName || runtime.model, modelCallCount: 0, toolCallCount: 0, toolTotalMs: 0, totalMs: 0, traces: [], modelDurationsMs: [], modelDiagnostics: [] }
|
||||
if (!trimmed) return { ...result, error: '请输入查询问题', totalMs: Date.now() - startedAt }
|
||||
if (!runtime.configured) return { ...result, error: '当前 AI Provider 尚未配置', totalMs: Date.now() - startedAt }
|
||||
|
||||
@@ -256,6 +444,8 @@ export class QueryAgentPocService {
|
||||
{ role: 'user', content: trimmed }
|
||||
]
|
||||
let tools = toolDefinitions()
|
||||
const retry = newRetryState()
|
||||
const now = this.nowProvider()
|
||||
let firstModelAt: number | undefined
|
||||
let finalModelDuration: number | undefined
|
||||
while (result.toolCallCount < MAX_TOOL_CALLS) {
|
||||
@@ -263,6 +453,19 @@ export class QueryAgentPocService {
|
||||
const model = await this.provider.chatWithTools(messages, tools)
|
||||
result.modelCallCount += 1
|
||||
const modelDuration = Date.now() - modelStartedAt
|
||||
// 无论成功失败都记录本次调用耗时与请求级诊断，便于区分模型慢 / 工具慢 / 上游错误。
|
||||
result.modelDurationsMs.push(modelDuration)
|
||||
result.modelDiagnostics.push({
|
||||
index: result.modelCallCount,
|
||||
elapsedMs: typeof model.elapsedMs === 'number' ? model.elapsedMs : modelDuration,
|
||||
...(model.errorStatus !== undefined ? { status: model.errorStatus } : {}),
|
||||
...(model.errorContentType ? { contentType: model.errorContentType } : {}),
|
||||
...(model.timedOut ? { timedOut: true } : {}),
|
||||
...(model.htmlInsteadOfJson ? { htmlInsteadOfJson: true } : {}),
|
||||
...(model.errorCode ? { errorCode: model.errorCode } : {}),
|
||||
...(model.errorType ? { errorType: model.errorType } : {}),
|
||||
...(model.success ? {} : { error: model.error || '模型调用失败' })
|
||||
})
|
||||
if (firstModelAt === undefined) firstModelAt = Date.now()
|
||||
if (!model.success) return { ...result, error: model.error || '模型调用失败', firstModelMs: firstModelAt - startedAt, totalMs: Date.now() - startedAt }
|
||||
const calls = model.toolCalls || []
|
||||
@@ -292,12 +495,18 @@ export class QueryAgentPocService {
|
||||
toolResult = { status: 'invalid_tool_arguments', field: '$', constraint: 'json' }
|
||||
}
|
||||
if (!toolResult) {
|
||||
const validated = validateToolArguments(call.name, parsed)
|
||||
const validated = validateToolArguments(call.name, parsed, now)
|
||||
if (validated.error) {
|
||||
toolResult = validated.error
|
||||
} else {
|
||||
traceInput = validated.input || {}
|
||||
toolResult = await this.executeTool(call.name, traceInput)
|
||||
if (duplicateRetry(call.name, traceInput, retry)) {
|
||||
// 明确拒绝“换关键词重搜”里的 identical retry，让模型改用实质不同的条件。
|
||||
toolResult = { status: 'invalid_tool_arguments', field: '$', constraint: 'duplicate_retry', expected: '与上一次实质不同的条件', actual: '与上一次完全相同的条件' }
|
||||
} else {
|
||||
recordAttempt(call.name, traceInput, retry)
|
||||
toolResult = await this.executeTool(call.name, traceInput)
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -312,8 +521,9 @@ export class QueryAgentPocService {
|
||||
result.traces.push({ toolName: call.name, input: sanitizeInput(traceInput), durationMs, status: completedToolResult.status, ...counts })
|
||||
const nextTools = completedToolResult.constraint === 'tool_availability'
|
||||
? tools
|
||||
: nextToolDefinitions(call.name, completedToolResult)
|
||||
messages.push({ role: 'tool', tool_call_id: call.id, name: call.name, content: JSON.stringify(toolResultForModel(call.name, completedToolResult, result.toolCallCount, nextTools)) })
|
||||
: nextToolDefinitions(call.name, completedToolResult, retry)
|
||||
const note = retryNote(call.name, completedToolResult, retry)
|
||||
messages.push({ role: 'tool', tool_call_id: call.id, name: call.name, content: JSON.stringify(toolResultForModel(call.name, completedToolResult, result.toolCallCount, nextTools, note)) })
|
||||
tools = nextTools
|
||||
}
|
||||
}
|
||||
|
||||
@@ -18,10 +18,21 @@ export interface LocalQueryToolDefinition {
|
||||
parameters: Record<string, unknown>
|
||||
}
|
||||
const targetSchema = { type: 'object', properties: { query: { type: 'string', minLength: 1 } }, required: ['query'], additionalProperties: false }
|
||||
const timeRangeSchema = { type: 'object', properties: { kind: { enum: ['all', 'today', 'yesterday', 'this_week', 'last_7_days', 'this_month', 'previous_month', 'this_year', 'previous_year', 'absolute'] }, startTime: { type: 'number' }, endTime: { type: 'number' } }, required: ['kind'], additionalProperties: false }
|
||||
// LLM-facing 绝对时间契约：只接受带时区偏移的 ISO-8601 字符串，避免 epoch 单位歧义。
|
||||
// Local Query API 内部仍使用 epoch seconds（见 QueryTimeRange），由 Host Adapter 负责转换。
|
||||
const timeRangeSchema = {
|
||||
type: 'object',
|
||||
properties: {
|
||||
kind: { enum: ['all', 'today', 'yesterday', 'this_week', 'last_7_days', 'this_month', 'previous_month', 'this_year', 'previous_year', 'absolute'] },
|
||||
startTime: { type: 'string', minLength: 1, description: '仅 kind=absolute 时必填。带时区偏移的 ISO-8601 字符串，例如 2026-08-01T00:00:00+08:00 或 2026-07-31T16:00:00Z。不接受 epoch 数字，也不接受无时区的裸本地时间。' },
|
||||
endTime: { type: 'string', minLength: 1, description: '仅 kind=absolute 时必填。含义与格式同 startTime，且不得早于 startTime。' }
|
||||
},
|
||||
required: ['kind'],
|
||||
additionalProperties: false
|
||||
}
|
||||
export const LOCAL_QUERY_TOOL_DEFINITIONS: LocalQueryToolDefinition[] = [
|
||||
{ name: 'query_messages', description: '精确读取符合联系人、时间、方向、消息类型、顺序等结构条件的消息；适合具体事实和 earliest/latest 等时间边界查询，边界查询使用 order 与 limit。', parameters: { type: 'object', required: ['target', 'timeRange'], additionalProperties: false, properties: { target: targetSchema, timeRange: timeRangeSchema, direction: { enum: ['any', 'from_target', 'to_target'] }, messageTypes: { type: 'array', items: { enum: ['text', 'image', 'voice', 'video', 'file', 'link', 'sticker', 'system', 'other'] } }, order: { enum: ['asc', 'desc'] }, limit: { type: 'integer', minimum: 1, maximum: 200 }, excludeSystem: { type: 'boolean' } } } },
|
||||
{ name: 'search_messages', description: '在指定联系人和时间范围内按语义寻找相关 Evidence；首次调用应集中提供高质量 query 和 variants。已有相关 Evidence 时直接判断，不要重复搜索。', parameters: { type: 'object', required: ['target', 'timeRange', 'query'], additionalProperties: false, properties: { target: targetSchema, timeRange: timeRangeSchema, query: { type: 'string', minLength: 1 }, variants: { type: 'array', description: '可选的语义变体，数组元素总数不得超过 4；调用前必须自行计数。', maxItems: 4, items: { type: 'string' } }, limit: { type: 'integer', minimum: 1, maximum: 200 } } } },
|
||||
{ name: 'query_messages', description: '精确读取符合联系人、时间、方向、消息类型、顺序等结构条件的消息；适合具体事实和 earliest/latest 等时间边界查询，边界查询使用 order 与 limit。若结果为 0 且条件明显不合适，可再查一次并合理扩大或更换条件，但条件必须与上一次实质不同。', parameters: { type: 'object', required: ['target', 'timeRange'], additionalProperties: false, properties: { target: targetSchema, timeRange: timeRangeSchema, direction: { enum: ['any', 'from_target', 'to_target'] }, messageTypes: { type: 'array', items: { enum: ['text', 'image', 'voice', 'video', 'file', 'link', 'sticker', 'system', 'other'] } }, order: { enum: ['asc', 'desc'] }, limit: { type: 'integer', minimum: 1, maximum: 200 }, excludeSystem: { type: 'boolean' } } } },
|
||||
{ name: 'search_messages', description: '在指定联系人和时间范围内做关键词检索并返回相关 Evidence。queries 的每一项都是一次独立的字面检索：一项只放一个简短关键词，不要把多个近义词或整句话放进同一项，也不要指望一项内部被拆词理解。首次最多 4 项；只有在本次检索完全没有 Evidence 时，才允许再检索一次，且每一项都必须与上一次实质不同。', parameters: { type: 'object', required: ['target', 'timeRange', 'queries'], additionalProperties: false, properties: { target: targetSchema, timeRange: timeRangeSchema, queries: { type: 'array', description: '独立检索项列表，每项一个简短关键词，最多 4 项；每一项单独检索，不会组合成一句话理解。', minItems: 1, maxItems: 4, items: { type: 'string', minLength: 1 } }, limit: { type: 'integer', minimum: 1, maximum: 200 } } } },
|
||||
{ name: 'message_context', description: '补充已找到的单条有价值 Evidence 的前后消息；仅在该 Evidence 缺少语境、无法判断含义时使用，不是默认确认步骤。', parameters: { type: 'object', required: ['messageRef'], additionalProperties: false, properties: { messageRef: { type: 'string', minLength: 1 }, before: { type: 'integer', minimum: 0, maximum: 50 }, after: { type: 'integer', minimum: 0, maximum: 50 } } } },
|
||||
{ name: 'conversation_overview', description: '提取指定联系人和时间范围的整体聊天覆盖样本；只用于 broad summary，不是语义搜索 fallback，也不能确定 earliest/latest 等精确时间边界。', parameters: { type: 'object', required: ['target', 'timeRange'], additionalProperties: false, properties: { target: targetSchema, timeRange: timeRangeSchema } } }
|
||||
]
|
||||
|
||||
@@ -0,0 +1,143 @@
|
||||
import { mkdtempSync, rmSync } from 'fs'
|
||||
import { tmpdir } from 'os'
|
||||
import { join } from 'path'
|
||||
import { afterAll, describe, expect, it, vi } from 'vitest'
|
||||
import type { AIProviderConfig } from '../../src/shared/ai-provider'
|
||||
|
||||
const root = mkdtempSync(join(tmpdir(), 'wxe-ai-provider-diagnostics-'))
|
||||
|
||||
vi.mock('electron', () => ({
|
||||
app: { getPath: () => root },
|
||||
safeStorage: {
|
||||
isEncryptionAvailable: () => true,
|
||||
encryptString: (value: string) => Buffer.from(value),
|
||||
decryptString: (value: Buffer) => value.toString('utf8')
|
||||
}
|
||||
}))
|
||||
|
||||
import { AIProviderService } from '../../src/main/services/ai-provider-service'
|
||||
|
||||
function provider(baseUrl: string, timeoutMs = 120_000): AIProviderConfig {
|
||||
return {
|
||||
id: 'fixture-provider',
|
||||
name: 'Fixture Provider',
|
||||
type: 'openai-compatible' as const,
|
||||
baseUrl,
|
||||
auth: { type: 'none' as const },
|
||||
models: [
|
||||
{ id: 'fixture-model', name: 'Fixture Model', capabilities: { chat: true, vision: false, ocr: false, longContext: false } }
|
||||
],
|
||||
defaultModel: 'fixture-model',
|
||||
advanced: { timeoutMs, extraHeaders: {} }
|
||||
}
|
||||
}
|
||||
|
||||
const TOOLS = [{ type: 'function' as const, function: { name: 'query_messages', description: 'read', parameters: { type: 'object' } } }]
|
||||
|
||||
describe('AI Provider 请求级诊断', () => {
|
||||
afterAll(() => rmSync(root, { recursive: true, force: true }))
|
||||
|
||||
it('200 JSON：成功并带请求耗时，无错误诊断字段', async () => {
|
||||
const service = new AIProviderService()
|
||||
service.save(provider('https://diag.example.test/v1'))
|
||||
vi.stubGlobal('fetch', vi.fn().mockResolvedValue(
|
||||
new Response(JSON.stringify({ choices: [{ message: { content: 'ok', tool_calls: [] } }], usage: {} }), {
|
||||
status: 200,
|
||||
headers: { 'content-type': 'application/json' }
|
||||
})
|
||||
))
|
||||
try {
|
||||
const result = await service.chatWithTools([{ role: 'user', content: 'q' }], TOOLS)
|
||||
expect(result.success).toBe(true)
|
||||
expect(typeof result.elapsedMs).toBe('number')
|
||||
expect(result).not.toHaveProperty('errorStatus')
|
||||
expect(result).not.toHaveProperty('htmlInsteadOfJson')
|
||||
} finally {
|
||||
vi.unstubAllGlobals()
|
||||
}
|
||||
})
|
||||
|
||||
it('502 text/html：保留清晰错误，并记录 status / content-type / elapsed，且不泄漏 HTML 正文', async () => {
|
||||
const service = new AIProviderService()
|
||||
service.save(provider('https://diag.example.test/v1'))
|
||||
const html = '<!DOCTYPE html><html><head><title>502 Bad Gateway</title></head><body>relay error</body></html>'
|
||||
vi.stubGlobal('fetch', vi.fn().mockResolvedValue(
|
||||
new Response(html, { status: 502, statusText: 'Bad Gateway', headers: { 'content-type': 'text/html' } })
|
||||
))
|
||||
try {
|
||||
const result = await service.chatWithTools([{ role: 'user', content: 'q' }], TOOLS)
|
||||
expect(result.success).toBe(false)
|
||||
expect(result.errorStatus).toBe(502)
|
||||
expect(result.errorContentType).toBe('text/html')
|
||||
expect(result.htmlInsteadOfJson).toBe(true)
|
||||
expect(typeof result.elapsedMs).toBe('number')
|
||||
expect(result.error).toContain('网页而不是 JSON')
|
||||
expect(result.error).toContain('502')
|
||||
// HTML 正文不得出现在错误信息里
|
||||
expect(result.error).not.toContain('relay error')
|
||||
expect(result.error).not.toContain('<html')
|
||||
} finally {
|
||||
vi.unstubAllGlobals()
|
||||
}
|
||||
})
|
||||
|
||||
it('502 application/json：使用上游 error.message，并保留状态码', async () => {
|
||||
const service = new AIProviderService()
|
||||
service.save(provider('https://diag.example.test/v1'))
|
||||
vi.stubGlobal('fetch', vi.fn().mockResolvedValue(
|
||||
new Response(JSON.stringify({ error: { message: 'upstream overloaded', code: 'overloaded', type: 'server_error' } }), {
|
||||
status: 502,
|
||||
headers: { 'content-type': 'application/json' }
|
||||
})
|
||||
))
|
||||
try {
|
||||
const result = await service.chatWithTools([{ role: 'user', content: 'q' }], TOOLS)
|
||||
expect(result.success).toBe(false)
|
||||
expect(result.error).toBe('upstream overloaded')
|
||||
expect(result.errorStatus).toBe(502)
|
||||
expect(result.errorCode).toBe('overloaded')
|
||||
expect(result.errorType).toBe('server_error')
|
||||
expect(result.htmlInsteadOfJson).toBe(false)
|
||||
} finally {
|
||||
vi.unstubAllGlobals()
|
||||
}
|
||||
})
|
||||
|
||||
it('200 但 body 不是 JSON：明确报格式异常并带状态码', async () => {
|
||||
const service = new AIProviderService()
|
||||
service.save(provider('https://diag.example.test/v1'))
|
||||
vi.stubGlobal('fetch', vi.fn().mockResolvedValue(
|
||||
new Response('plain text body', { status: 200, headers: { 'content-type': 'text/plain' } })
|
||||
))
|
||||
try {
|
||||
const result = await service.chatWithTools([{ role: 'user', content: 'q' }], TOOLS)
|
||||
expect(result.success).toBe(false)
|
||||
expect(result.error).toContain('格式异常')
|
||||
expect(result.htmlInsteadOfJson).toBe(false)
|
||||
} finally {
|
||||
vi.unstubAllGlobals()
|
||||
}
|
||||
})
|
||||
|
||||
it('超时：标记 timedOut 且错误为 AI 请求超时', async () => {
|
||||
const service = new AIProviderService()
|
||||
service.save(provider('https://diag.example.test/v1', 1))
|
||||
vi.stubGlobal('fetch', vi.fn((_url: string, init?: RequestInit) => new Promise<Response>((_resolve, reject) => {
|
||||
init?.signal?.addEventListener('abort', () => reject(init.signal?.reason), { once: true })
|
||||
})))
|
||||
try {
|
||||
const result = await service.chatWithTools([{ role: 'user', content: 'q' }], TOOLS)
|
||||
expect(result.success).toBe(false)
|
||||
expect(result.timedOut).toBe(true)
|
||||
expect(result.error).toBe('AI 请求超时')
|
||||
} finally {
|
||||
vi.unstubAllGlobals()
|
||||
}
|
||||
}, 5_000)
|
||||
|
||||
it('只输出 host 作为 endpoint 诊断，不泄漏路径与 query', () => {
|
||||
const service = new AIProviderService()
|
||||
service.save(provider('https://relay.example.test/v1?token=secret-value'))
|
||||
expect(service.getRuntimeEndpointHost()).toBe('relay.example.test')
|
||||
})
|
||||
})
|
||||
@@ -0,0 +1,36 @@
|
||||
import { describe, expect, it } from 'vitest'
|
||||
import { parsePocQuestion } from '../../src/main/query-agent-poc-cli'
|
||||
|
||||
describe('Query Agent POC CLI 参数解析', () => {
|
||||
it('移除开头的分隔符 --', () => {
|
||||
expect(parsePocQuestion(['--', '我和BOBO第一次聊了什么'])).toBe('我和BOBO第一次聊了什么')
|
||||
})
|
||||
|
||||
it('没有分隔符时保持原样', () => {
|
||||
expect(parsePocQuestion(['我和BOBO第一次聊了什么'])).toBe('我和BOBO第一次聊了什么')
|
||||
})
|
||||
|
||||
it('不删除问题正文中间的合法 --', () => {
|
||||
expect(parsePocQuestion(['测试', '--', '内容'])).toBe('测试 -- 内容')
|
||||
})
|
||||
|
||||
it('开头分隔符只移除一个，后续 -- 仍是正文', () => {
|
||||
expect(parsePocQuestion(['--', '--', '内容'])).toBe('-- 内容')
|
||||
expect(parsePocQuestion(['--', 'a', '--', 'b'])).toBe('a -- b')
|
||||
})
|
||||
|
||||
it('多段参数按空格拼接，并去掉首尾空白', () => {
|
||||
expect(parsePocQuestion([' ', 'BOBO', '上个月', '有没有给我发过文件 '])).toBe('BOBO 上个月 有没有给我发过文件')
|
||||
})
|
||||
|
||||
it('只有分隔符或空参数时返回空字符串', () => {
|
||||
expect(parsePocQuestion(['--'])).toBe('')
|
||||
expect(parsePocQuestion([])).toBe('')
|
||||
})
|
||||
|
||||
it('不修改传入的数组', () => {
|
||||
const argv = ['--', '问题']
|
||||
parsePocQuestion(argv)
|
||||
expect(argv).toEqual(['--', '问题'])
|
||||
})
|
||||
})
|
||||
@@ -0,0 +1,120 @@
|
||||
import { describe, expect, it } from 'vitest'
|
||||
import { formatProviderDiagnostics, formatTiming } from '../../src/main/query-agent-poc-report'
|
||||
import type { QueryAgentPocResult } from '../../src/main/services/query-agent-poc-service'
|
||||
|
||||
function result(patch: Partial<QueryAgentPocResult> = {}): QueryAgentPocResult {
|
||||
return {
|
||||
question: 'q',
|
||||
provider: 'Fixture Provider',
|
||||
model: 'fixture-model',
|
||||
modelCallCount: 0,
|
||||
toolCallCount: 0,
|
||||
toolTotalMs: 0,
|
||||
totalMs: 0,
|
||||
traces: [],
|
||||
modelDurationsMs: [],
|
||||
modelDiagnostics: [],
|
||||
...patch
|
||||
}
|
||||
}
|
||||
|
||||
describe('formatTiming', () => {
|
||||
it('处理 0 次模型调用', () => {
|
||||
const text = formatTiming(result({ error: '当前 AI Provider 尚未配置', totalMs: 3 }))
|
||||
expect(text).toContain('（未发生模型调用）')
|
||||
expect(text).toContain('Total')
|
||||
expect(text).toContain('error: 当前 AI Provider 尚未配置')
|
||||
})
|
||||
|
||||
it('处理 1 次模型调用 + 0 次工具', () => {
|
||||
const text = formatTiming(result({ modelCallCount: 1, modelDurationsMs: [1234], totalMs: 1234 }))
|
||||
expect(text).toContain('Model #1')
|
||||
expect(text).toContain('1234 ms')
|
||||
expect(text).not.toContain('TM Tools')
|
||||
expect(text).toContain('Model total')
|
||||
expect(text).toContain('100.0%')
|
||||
})
|
||||
|
||||
it('处理 2 次模型调用 + 1 次工具，并还原交错顺序', () => {
|
||||
const text = formatTiming(
|
||||
result({ modelCallCount: 2, toolCallCount: 1, modelDurationsMs: [18784, 19690], toolTotalMs: 551, totalMs: 39025 })
|
||||
)
|
||||
const model1 = text.indexOf('Model #1')
|
||||
const tools = text.indexOf('TM Tools')
|
||||
const model2 = text.indexOf('Model #2')
|
||||
expect(model1).toBeGreaterThanOrEqual(0)
|
||||
expect(tools).toBeGreaterThan(model1)
|
||||
expect(model2).toBeGreaterThan(tools)
|
||||
expect(text).toContain('38474 ms (98.6%)')
|
||||
expect(text).toContain('551 ms (1.4%)')
|
||||
})
|
||||
|
||||
it('处理首次模型调用失败', () => {
|
||||
const text = formatTiming(
|
||||
result({ modelCallCount: 1, modelDurationsMs: [99419], totalMs: 99419, error: '模型服务返回了网页而不是 JSON（HTTP 502 Bad Gateway）' })
|
||||
)
|
||||
expect(text).toContain('Model #1')
|
||||
expect(text).toContain('99419 ms')
|
||||
expect(text).toContain('502')
|
||||
})
|
||||
|
||||
it('处理末尾模型调用失败（有首次耗时、无 finalModelMs）', () => {
|
||||
const text = formatTiming(
|
||||
result({ modelCallCount: 2, toolCallCount: 1, modelDurationsMs: [800, 1200], toolTotalMs: 300, totalMs: 2300, error: 'AI 请求超时' })
|
||||
)
|
||||
expect(text).toContain('Model #2')
|
||||
expect(text).toContain('error: AI 请求超时')
|
||||
})
|
||||
|
||||
it('3 次以上模型调用标注为聚合展示', () => {
|
||||
const text = formatTiming(result({ modelCallCount: 3, toolCallCount: 2, modelDurationsMs: [10, 20, 30], toolTotalMs: 5, totalMs: 65 }))
|
||||
expect(text).toContain('Model #3')
|
||||
expect(text).toContain('工具耗时按总计展示')
|
||||
})
|
||||
|
||||
it('totalMs 为 0 时不产生 NaN 百分比', () => {
|
||||
const text = formatTiming(result({ modelCallCount: 1, modelDurationsMs: [5], totalMs: 0 }))
|
||||
expect(text).not.toContain('NaN')
|
||||
expect(text).toContain('n/a')
|
||||
})
|
||||
})
|
||||
|
||||
describe('formatProviderDiagnostics', () => {
|
||||
it('输出 provider / model / host 与每次尝试的请求级诊断', () => {
|
||||
const text = formatProviderDiagnostics(
|
||||
result({
|
||||
provider: 'OpenAI',
|
||||
model: 'gpt-5.6-sol',
|
||||
modelCallCount: 1,
|
||||
modelDiagnostics: [{ index: 1, elapsedMs: 99419, status: 502, contentType: 'text/html', htmlInsteadOfJson: true, error: '模型服务返回了网页而不是 JSON（HTTP 502 Bad Gateway）' }]
|
||||
}),
|
||||
'relay.example.test'
|
||||
)
|
||||
expect(text).toContain('provider OpenAI')
|
||||
expect(text).toContain('model gpt-5.6-sol')
|
||||
expect(text).toContain('host relay.example.test')
|
||||
expect(text).toContain('status=502')
|
||||
expect(text).toContain('contentType=text/html')
|
||||
expect(text).toContain('htmlInsteadOfJson=true')
|
||||
expect(text).toContain('upstreamGateway=true')
|
||||
expect(text).toContain('elapsedMs=99419')
|
||||
})
|
||||
|
||||
it('标记超时', () => {
|
||||
const text = formatProviderDiagnostics(
|
||||
result({ modelCallCount: 1, modelDiagnostics: [{ index: 1, elapsedMs: 240000, timedOut: true, error: 'AI 请求超时' }] })
|
||||
)
|
||||
expect(text).toContain('timedOut=true')
|
||||
expect(text).toContain('(unknown)')
|
||||
})
|
||||
|
||||
it('不泄漏凭据：输出中不出现 token / Authorization', () => {
|
||||
const text = formatProviderDiagnostics(
|
||||
result({ modelDiagnostics: [{ index: 1, elapsedMs: 1, status: 200, contentType: 'application/json' }] }),
|
||||
'relay.example.test'
|
||||
)
|
||||
expect(text).not.toMatch(/authorization/i)
|
||||
expect(text).not.toMatch(/bearer/i)
|
||||
expect(text).not.toMatch(/sk-/)
|
||||
})
|
||||
})
|
||||
@@ -32,7 +32,7 @@ describe('QueryAgentPocService', () => {
|
||||
evidence: [{ messageRef: 'opaque', timestamp: 1, sender: 'BOBO', sourceKind: 'image', text: 'caption' }]
|
||||
}))
|
||||
const configuredProvider = provider([
|
||||
{ success: true, toolCalls: [{ id: 'call-1', name: 'search_messages', arguments: JSON.stringify({ target: { query: 'BOBO' }, timeRange: { kind: 'all' }, query: 'topic' }) }] },
|
||||
{ success: true, toolCalls: [{ id: 'call-1', name: 'search_messages', arguments: JSON.stringify({ target: { query: 'BOBO' }, timeRange: { kind: 'all' }, queries: ['topic'] }) }] },
|
||||
{ success: true, data: '根据这条证据可以回答。' }
|
||||
])
|
||||
await new QueryAgentPocService(configuredProvider, execute).run('查找相关记录')
|
||||
@@ -88,18 +88,21 @@ describe('QueryAgentPocService', () => {
|
||||
expect(configuredProvider.chatWithTools).not.toHaveBeenCalled()
|
||||
})
|
||||
|
||||
it('accepts schema-valid variants and rejects invalid arguments before executing a tool', async () => {
|
||||
it('maps LLM queries[] to the Local Query API query + variants contract', async () => {
|
||||
const execute = vi.fn(async () => ({ status: 'completed', returnedCount: 0 }))
|
||||
const valid = await new QueryAgentPocService(provider([
|
||||
{ success: true, toolCalls: [{ id: 'valid', name: 'search_messages', arguments: JSON.stringify({ target: { query: 'BOBO' }, timeRange: { kind: 'all' }, query: '答应', variants: ['承诺', '保证', '说好', '一定'] }) }] },
|
||||
{ success: true, toolCalls: [{ id: 'valid', name: 'search_messages', arguments: JSON.stringify({ target: { query: 'BOBO' }, timeRange: { kind: 'all' }, queries: ['答应', '承诺', '保证', '说好'] }) }] },
|
||||
{ success: true, data: 'done' }
|
||||
]), execute).run('test')
|
||||
expect(valid.traces[0]).toMatchObject({ status: 'completed', toolName: 'search_messages' })
|
||||
expect(execute).toHaveBeenCalledTimes(1)
|
||||
// Local Query API 仍然只认 query + variants，Host 负责映射。
|
||||
expect(execute.mock.calls[0][1]).toMatchObject({ query: '答应', variants: ['承诺', '保证', '说好'] })
|
||||
expect(execute.mock.calls[0][1]).not.toHaveProperty('queries')
|
||||
|
||||
execute.mockClear()
|
||||
const invalid = await new QueryAgentPocService(provider([
|
||||
{ success: true, toolCalls: [{ id: 'invalid', name: 'search_messages', arguments: JSON.stringify({ target: { query: 'BOBO' }, timeRange: { kind: 'all' }, query: '答应', variants: ['1', '2', '3', '4', '5'] }) }] },
|
||||
{ success: true, toolCalls: [{ id: 'invalid', name: 'search_messages', arguments: JSON.stringify({ target: { query: 'BOBO' }, timeRange: { kind: 'all' }, queries: ['1', '2', '3', '4', '5'] }) }] },
|
||||
{ success: true, data: '修正后完成' }
|
||||
]), execute).run('test')
|
||||
expect(invalid.traces[0]).toMatchObject({ status: 'invalid_tool_arguments' })
|
||||
@@ -108,13 +111,249 @@ describe('QueryAgentPocService', () => {
|
||||
})
|
||||
|
||||
it('enforces generic nested tool schema constraints', () => {
|
||||
const base = { target: { query: 'BOBO' }, timeRange: { kind: 'all' }, query: 'x' }
|
||||
expect(validateToolArguments('search_messages', { ...base, variants: ['1', '2', '3', '4'] }).error).toBeUndefined()
|
||||
expect(validateToolArguments('search_messages', { ...base, variants: ['1', '2', '3', '4', '5'] }).error).toMatchObject({ field: 'variants', constraint: 'maxItems', expected: 4, actual: 5 })
|
||||
const base = { target: { query: 'BOBO' }, timeRange: { kind: 'all' }, queries: ['x'] }
|
||||
expect(validateToolArguments('search_messages', { ...base, queries: ['1', '2', '3', '4'] }).error).toBeUndefined()
|
||||
expect(validateToolArguments('search_messages', { ...base, queries: ['1', '2', '3', '4', '5'] }).error).toMatchObject({ field: 'queries', constraint: 'maxItems', expected: 4, actual: 5 })
|
||||
expect(validateToolArguments('search_messages', { ...base, queries: [] }).error).toMatchObject({ field: 'queries', constraint: 'minItems', expected: 1, actual: 0 })
|
||||
expect(validateToolArguments('search_messages', { ...base, extra: true }).error).toMatchObject({ field: 'extra', constraint: 'additionalProperties' })
|
||||
// 旧的 query 字段已不再是合法属性，必须被拒绝而不是静默忽略。
|
||||
expect(validateToolArguments('search_messages', { ...base, query: 'x' }).error).toMatchObject({ field: 'query', constraint: 'additionalProperties' })
|
||||
// 漏掉 queries 时按 required 拒绝。
|
||||
expect(validateToolArguments('search_messages', { target: { query: 'BOBO' }, timeRange: { kind: 'all' } }).error).toMatchObject({ field: 'queries', constraint: 'required' })
|
||||
expect(validateToolArguments('query_messages', { target: { query: 'BOBO' }, timeRange: { kind: 'all' }, direction: 'sideways' }).error).toMatchObject({ field: 'direction', constraint: 'enum' })
|
||||
expect(validateToolArguments('query_messages', { target: { query: '' }, timeRange: { kind: 'all' } }).error).toMatchObject({ field: 'target.query', constraint: 'minLength' })
|
||||
expect(validateToolArguments('query_messages', { target: { query: 'BOBO' }, timeRange: { kind: 'all' }, limit: 0 }).error).toMatchObject({ field: 'limit', constraint: 'minimum' })
|
||||
expect(validateToolArguments('message_context', { messageRef: 'opaque', before: 51 }).error).toMatchObject({ field: 'before', constraint: 'maximum' })
|
||||
})
|
||||
|
||||
it('rejects a whitespace-only search probe instead of sending an empty query', () => {
|
||||
const error = validateToolArguments('search_messages', { target: { query: 'BOBO' }, timeRange: { kind: 'all' }, queries: [' '] }).error
|
||||
expect(error).toMatchObject({ field: 'queries', constraint: 'required' })
|
||||
})
|
||||
})
|
||||
|
||||
describe('QueryAgent absolute time contract', () => {
|
||||
const now = new Date('2026-09-10T00:00:00Z')
|
||||
const target = { query: 'BOBO' }
|
||||
|
||||
function canonical(timeRange: Record<string, unknown>) {
|
||||
return validateToolArguments('query_messages', { target, timeRange }, now)
|
||||
}
|
||||
|
||||
it('canonicalizes an ISO-8601 absolute range with offset into Local Query API epoch seconds', () => {
|
||||
const result = canonical({ kind: 'absolute', startTime: '2025-08-01T00:00:00+08:00', endTime: '2025-09-01T00:00:00+08:00' })
|
||||
expect(result.error).toBeUndefined()
|
||||
expect(result.input?.timeRange).toEqual({
|
||||
kind: 'absolute',
|
||||
startTime: Math.floor(Date.parse('2025-07-31T16:00:00Z') / 1000),
|
||||
endTime: Math.floor(Date.parse('2025-08-31T16:00:00Z') / 1000)
|
||||
})
|
||||
})
|
||||
|
||||
it('round-trips seconds and ISO to the same instant', () => {
|
||||
const result = canonical({ kind: 'absolute', startTime: '2025-08-13T03:27:34Z', endTime: '2025-08-13T04:00:00Z' })
|
||||
const startSeconds = (result.input?.timeRange as { startTime: number }).startTime
|
||||
expect(startSeconds).toBe(Math.floor(1755055654000 / 1000))
|
||||
expect(new Date(startSeconds * 1000).toISOString()).toBe('2025-08-13T03:27:34.000Z')
|
||||
})
|
||||
|
||||
it('treats equivalent offsets as the same instant', () => {
|
||||
const beijing = canonical({ kind: 'absolute', startTime: '2025-08-01T00:00:00+08:00', endTime: '2025-08-02T00:00:00+08:00' })
|
||||
const utc = canonical({ kind: 'absolute', startTime: '2025-07-31T16:00:00Z', endTime: '2025-08-01T16:00:00Z' })
|
||||
expect(beijing.input?.timeRange).toEqual(utc.input?.timeRange)
|
||||
})
|
||||
|
||||
it('supports cross-year ranges', () => {
|
||||
const result = canonical({ kind: 'absolute', startTime: '2025-12-31T20:00:00+08:00', endTime: '2026-01-01T04:00:00+08:00' })
|
||||
const range = result.input?.timeRange as { startTime: number; endTime: number }
|
||||
expect(result.error).toBeUndefined()
|
||||
expect(range.endTime).toBeGreaterThan(range.startTime)
|
||||
expect(new Date(range.startTime * 1000).toISOString()).toBe('2025-12-31T12:00:00.000Z')
|
||||
expect(new Date(range.endTime * 1000).toISOString()).toBe('2025-12-31T20:00:00.000Z')
|
||||
})
|
||||
|
||||
it('never lets millisecond epoch numbers reach the executor silently', () => {
|
||||
const error = canonical({ kind: 'absolute', startTime: 1753977600000, endTime: 1756656000000 }).error
|
||||
expect(error).toMatchObject({ field: 'timeRange.startTime', constraint: 'type', expected: 'string', actual: 'number' })
|
||||
})
|
||||
|
||||
it('rejects naive local datetimes without a timezone', () => {
|
||||
expect(canonical({ kind: 'absolute', startTime: '2025-08-01T00:00:00', endTime: '2025-09-01T00:00:00' }).error).toMatchObject({ field: 'timeRange.startTime', constraint: 'format' })
|
||||
})
|
||||
|
||||
it('rejects invalid calendar days and out-of-range clock values', () => {
|
||||
expect(canonical({ kind: 'absolute', startTime: '2026-02-31T00:00:00Z', endTime: '2026-03-01T00:00:00Z' }).error).toMatchObject({ field: 'timeRange.startTime', constraint: 'format' })
|
||||
expect(canonical({ kind: 'absolute', startTime: '2026-03-01T25:00:00Z', endTime: '2026-03-02T00:00:00Z' }).error).toMatchObject({ field: 'timeRange.startTime', constraint: 'format' })
|
||||
})
|
||||
|
||||
it('rejects start after end', () => {
|
||||
expect(canonical({ kind: 'absolute', startTime: '2026-03-02T00:00:00Z', endTime: '2026-03-01T00:00:00Z' }).error).toMatchObject({ field: 'timeRange.endTime', constraint: 'range_order' })
|
||||
})
|
||||
|
||||
it('rejects absurd future and past ranges instead of returning complete-zero', () => {
|
||||
expect(canonical({ kind: 'absolute', startTime: '2999-01-01T00:00:00Z', endTime: '2999-02-01T00:00:00Z' }).error).toMatchObject({ field: 'timeRange.startTime', constraint: 'range_sanity' })
|
||||
expect(canonical({ kind: 'absolute', startTime: '1999-01-01T00:00:00Z', endTime: '1999-02-01T00:00:00Z' }).error).toMatchObject({ field: 'timeRange.startTime', constraint: 'range_sanity' })
|
||||
expect(canonical({ kind: 'absolute', startTime: '57526-01-01T00:00:00Z', endTime: '57526-02-01T00:00:00Z' }).error).toMatchObject({ field: 'timeRange.startTime', constraint: 'format' })
|
||||
})
|
||||
|
||||
it('drops stray startTime/endTime for non-absolute kinds', () => {
|
||||
expect(canonical({ kind: 'all', startTime: '2026-01-01T00:00:00Z', endTime: '2026-02-01T00:00:00Z' }).input?.timeRange).toEqual({ kind: 'all' })
|
||||
expect(canonical({ kind: 'previous_month' }).input?.timeRange).toEqual({ kind: 'previous_month' })
|
||||
})
|
||||
|
||||
it('requires both bounds for absolute ranges', () => {
|
||||
expect(canonical({ kind: 'absolute', startTime: '2026-01-01T00:00:00Z' }).error).toMatchObject({ field: 'timeRange.startTime', constraint: 'required' })
|
||||
})
|
||||
})
|
||||
|
||||
describe('QueryAgent zero-result limited retry', () => {
|
||||
function searchCall(id: string, queries: string[], timeRange: Record<string, unknown> = { kind: 'all' }) {
|
||||
return { success: true as const, toolCalls: [{ id, name: 'search_messages', arguments: JSON.stringify({ target: { query: 'BOBO' }, timeRange, queries }) }] }
|
||||
}
|
||||
function queryCall(id: string, timeRange: Record<string, unknown> = { kind: 'all' }) {
|
||||
return { success: true as const, toolCalls: [{ id, name: 'query_messages', arguments: JSON.stringify({ target: { query: 'BOBO' }, timeRange }) }] }
|
||||
}
|
||||
const emptySearch = async () => ({ status: 'completed', evidenceCount: 0, evidence: [] })
|
||||
const emptyQuery = async () => ({ status: 'completed', returnedCount: 0, messages: [] })
|
||||
|
||||
it('keeps search_messages available once after a zero-evidence search', async () => {
|
||||
const execute = vi.fn(emptySearch)
|
||||
const configured = provider([searchCall('c1', ['答应']), { success: true, data: '没有找到相关证据。' }])
|
||||
await new QueryAgentPocService(configured, execute).run('找承诺')
|
||||
expect(vi.mocked(configured.chatWithTools).mock.calls[1]?.[1].map((tool) => tool.function.name)).toEqual(['search_messages'])
|
||||
const toolMessage = vi.mocked(configured.chatWithTools).mock.calls[1]?.[0].find((message) => message.role === 'tool')
|
||||
const presented = JSON.parse(String(toolMessage?.content)) as Record<string, any>
|
||||
expect(presented._agent).toMatchObject({ toolCallsUsed: 1, availableNextTools: ['search_messages'] })
|
||||
expect(String(presented._agent.note)).toContain('实质不同')
|
||||
})
|
||||
|
||||
it('allows exactly one substantively different second search and then closes tools', async () => {
|
||||
const execute = vi.fn(emptySearch)
|
||||
const configured = provider([searchCall('c1', ['答应']), searchCall('c2', ['公积金']), { success: true, data: '仍然没有。' }])
|
||||
const result = await new QueryAgentPocService(configured, execute).run('找承诺')
|
||||
expect(execute).toHaveBeenCalledTimes(2)
|
||||
expect(result.traces.map((trace) => trace.status)).toEqual(['completed', 'completed'])
|
||||
expect(vi.mocked(configured.chatWithTools).mock.calls[2]?.[1]).toEqual([])
|
||||
expect(result.error).toBeUndefined()
|
||||
})
|
||||
|
||||
it('rejects an identical search retry instead of spending the budget again', async () => {
|
||||
const execute = vi.fn(emptySearch)
|
||||
const configured = provider([searchCall('c1', ['答应']), searchCall('c2', ['答应']), { success: true, data: '没有找到。' }])
|
||||
const result = await new QueryAgentPocService(configured, execute).run('找承诺')
|
||||
expect(execute).toHaveBeenCalledTimes(1)
|
||||
expect(result.traces[1]).toMatchObject({ status: 'invalid_tool_arguments' })
|
||||
expect(vi.mocked(configured.chatWithTools).mock.calls[2]?.[1]).toEqual([])
|
||||
const toolMessage = vi.mocked(configured.chatWithTools).mock.calls[2]?.[0].findLast((message) => message.role === 'tool')
|
||||
expect(String(toolMessage?.content)).toContain('duplicate_retry')
|
||||
})
|
||||
|
||||
it('treats a reordered identical probe set as an identical retry', async () => {
|
||||
const execute = vi.fn(emptySearch)
|
||||
const configured = provider([searchCall('c1', ['答应', '承诺']), searchCall('c2', ['承诺', '答应']), { success: true, data: 'x' }])
|
||||
const result = await new QueryAgentPocService(configured, execute).run('找承诺')
|
||||
expect(execute).toHaveBeenCalledTimes(1)
|
||||
expect(result.traces[1].status).toBe('invalid_tool_arguments')
|
||||
})
|
||||
|
||||
it('keeps the efficient path when the first search already returned evidence', async () => {
|
||||
const execute = vi.fn(async () => ({ status: 'completed', evidenceCount: 2, evidence: [{ messageRef: 'ref', text: 't' }] }))
|
||||
const configured = provider([searchCall('c1', ['答应']), { success: true, data: 'ok' }])
|
||||
await new QueryAgentPocService(configured, execute).run('找承诺')
|
||||
expect(vi.mocked(configured.chatWithTools).mock.calls[1]?.[1].map((tool) => tool.function.name)).toEqual(['message_context'])
|
||||
})
|
||||
|
||||
it('allows one query_messages retry after a zero-result exact query', async () => {
|
||||
const execute = vi.fn()
|
||||
.mockResolvedValueOnce({ status: 'completed', returnedCount: 0, messages: [] })
|
||||
.mockResolvedValueOnce({ status: 'completed', returnedCount: 1, messages: [{ messageRef: 'ref' }] })
|
||||
const configured = provider([queryCall('c1', { kind: 'last_7_days' }), queryCall('c2', { kind: 'all' }), { success: true, data: '找到。' }])
|
||||
const result = await new QueryAgentPocService(configured, execute).run('找文件')
|
||||
expect(execute).toHaveBeenCalledTimes(2)
|
||||
expect(execute.mock.calls[1][1]).toMatchObject({ timeRange: { kind: 'all' } })
|
||||
expect(result.traces[1].status).toBe('completed')
|
||||
})
|
||||
|
||||
it('rejects an identical query_messages retry', async () => {
|
||||
const execute = vi.fn(emptyQuery)
|
||||
const configured = provider([queryCall('c1', { kind: 'previous_month' }), queryCall('c2', { kind: 'previous_month' }), { success: true, data: 'x' }])
|
||||
const result = await new QueryAgentPocService(configured, execute).run('找文件')
|
||||
expect(execute).toHaveBeenCalledTimes(1)
|
||||
expect(result.traces[1].status).toBe('invalid_tool_arguments')
|
||||
})
|
||||
|
||||
it('keeps the efficient stop when the exact query already returned messages', async () => {
|
||||
const execute = vi.fn(async () => ({ status: 'completed', returnedCount: 3, messages: [] }))
|
||||
const configured = provider([queryCall('c1'), { success: true, data: 'ok' }])
|
||||
await new QueryAgentPocService(configured, execute).run('找')
|
||||
expect(vi.mocked(configured.chatWithTools).mock.calls[1]?.[1]).toEqual([])
|
||||
})
|
||||
|
||||
it('still caps total tool calls at five', async () => {
|
||||
const execute = vi.fn(emptyQuery)
|
||||
const responses = [
|
||||
queryCall('c1', { kind: 'last_7_days' }),
|
||||
queryCall('c2', { kind: 'all' }),
|
||||
queryCall('c3', { kind: 'today' }),
|
||||
queryCall('c4', { kind: 'yesterday' }),
|
||||
queryCall('c5', { kind: 'this_year' }),
|
||||
queryCall('c6', { kind: 'this_month' })
|
||||
]
|
||||
const result = await new QueryAgentPocService(provider(responses), execute).run('找')
|
||||
expect(result.toolCallCount).toBe(5)
|
||||
expect(result.error).toContain('最大工具调用次数')
|
||||
})
|
||||
})
|
||||
|
||||
describe('QueryAgent 耗时与请求级诊断记录', () => {
|
||||
function queryCall(id: string) {
|
||||
return { success: true as const, toolCalls: [{ id, name: 'query_messages', arguments: JSON.stringify({ target: { query: 'BOBO' }, timeRange: { kind: 'all' }, limit: 1 }) }] }
|
||||
}
|
||||
|
||||
it('记录每次模型调用耗时，包括失败的那次（首次调用失败）', async () => {
|
||||
const configured = provider([
|
||||
{ success: false, error: '模型服务返回了网页而不是 JSON（HTTP 502 Bad Gateway）', errorStatus: 502, errorContentType: 'text/html', htmlInsteadOfJson: true, elapsedMs: 99419 }
|
||||
])
|
||||
const result = await new QueryAgentPocService(configured, vi.fn()).run('测试')
|
||||
expect(result.modelCallCount).toBe(1)
|
||||
expect(result.modelDurationsMs).toHaveLength(1)
|
||||
expect(result.modelDiagnostics).toEqual([
|
||||
expect.objectContaining({ index: 1, status: 502, contentType: 'text/html', htmlInsteadOfJson: true, elapsedMs: 99419 })
|
||||
])
|
||||
expect(result.error).toContain('502')
|
||||
// 首次模型调用就失败，不应产生任何工具调用
|
||||
expect(result.toolCallCount).toBe(0)
|
||||
expect(result.finalModelMs).toBeUndefined()
|
||||
})
|
||||
|
||||
it('2 次模型调用 + 1 次工具：耗时数组与调用次数一致', async () => {
|
||||
const configured = provider([queryCall('c1'), { success: true, data: '完成' }])
|
||||
const result = await new QueryAgentPocService(configured, vi.fn(async () => ({ status: 'completed', returnedCount: 1 }))).run('测试')
|
||||
expect(result.modelCallCount).toBe(2)
|
||||
expect(result.modelDurationsMs).toHaveLength(2)
|
||||
expect(result.modelDurationsMs.every((value) => typeof value === 'number' && value >= 0)).toBe(true)
|
||||
expect(result.modelDiagnostics.map((entry) => entry.index)).toEqual([1, 2])
|
||||
expect(result.toolCallCount).toBe(1)
|
||||
})
|
||||
|
||||
it('1 次模型调用 + 0 次工具：无工具诊断噪声', async () => {
|
||||
const configured = provider([{ success: true, data: '直接回答' }])
|
||||
const result = await new QueryAgentPocService(configured, vi.fn()).run('测试')
|
||||
expect(result.modelCallCount).toBe(1)
|
||||
expect(result.toolCallCount).toBe(0)
|
||||
expect(result.modelDurationsMs).toHaveLength(1)
|
||||
expect(result.modelDiagnostics[0]).not.toHaveProperty('error')
|
||||
})
|
||||
|
||||
it('末尾模型调用失败时两条记录都在，且第二条带错误', async () => {
|
||||
const configured = provider([
|
||||
queryCall('c1'),
|
||||
{ success: false, error: 'AI 请求超时', timedOut: true }
|
||||
])
|
||||
const result = await new QueryAgentPocService(configured, vi.fn(async () => ({ status: 'completed', returnedCount: 1 }))).run('测试')
|
||||
expect(result.modelDurationsMs).toHaveLength(2)
|
||||
expect(result.modelDiagnostics[1]).toMatchObject({ index: 2, timedOut: true })
|
||||
expect(result.error).toBe('AI 请求超时')
|
||||
expect(result.answer).toBeUndefined()
|
||||
})
|
||||
})
|
||||
|
||||
@@ -0,0 +1,63 @@
|
||||
import { copyFileSync, mkdirSync, mkdtempSync, rmSync, writeFileSync } from 'fs'
|
||||
import { tmpdir } from 'os'
|
||||
import { join } from 'path'
|
||||
import { spawnSync } from 'child_process'
|
||||
import { afterAll, describe, expect, it } from 'vitest'
|
||||
|
||||
const repoRoot = process.cwd()
|
||||
const sandboxes: string[] = []
|
||||
|
||||
/** 搭一个最小"仓库"：只放 runner 脚本，可选放假的构建产物。 */
|
||||
function sandbox(withBuild: boolean): string {
|
||||
const dir = mkdtempSync(join(tmpdir(), 'wxe-poc-run-'))
|
||||
sandboxes.push(dir)
|
||||
mkdirSync(join(dir, 'scripts'), { recursive: true })
|
||||
copyFileSync(join(repoRoot, 'scripts', 'run-query-agent-poc.cjs'), join(dir, 'scripts', 'run-query-agent-poc.cjs'))
|
||||
if (withBuild) {
|
||||
mkdirSync(join(dir, 'out', 'main'), { recursive: true })
|
||||
writeFileSync(
|
||||
join(dir, 'out', 'main', 'queryAgentPoc.js'),
|
||||
'process.stdout.write(JSON.stringify({ ranEntry: true, argv: process.argv.slice(2) }) + "\\n")\n'
|
||||
)
|
||||
}
|
||||
return dir
|
||||
}
|
||||
|
||||
function run(dir: string, args: string[], env: Record<string, string> = {}) {
|
||||
return spawnSync(process.execPath, [join(dir, 'scripts', 'run-query-agent-poc.cjs'), ...args], {
|
||||
encoding: 'utf8',
|
||||
cwd: dir,
|
||||
env: { ...process.env, ...env }
|
||||
})
|
||||
}
|
||||
|
||||
describe('poc:query-agent:run 快速入口', () => {
|
||||
afterAll(() => {
|
||||
for (const dir of sandboxes) rmSync(dir, { recursive: true, force: true })
|
||||
})
|
||||
|
||||
it('构建缺失时给出明确提示并退出码 1，不触发构建', () => {
|
||||
const result = run(sandbox(false), ['问题'])
|
||||
expect(result.status).toBe(1)
|
||||
expect(result.stderr).toContain('POC build 不存在')
|
||||
expect(result.stderr).toContain('pnpm poc:query-agent')
|
||||
expect(result.stderr).toContain('不自动构建')
|
||||
expect(result.stdout).toBe('')
|
||||
})
|
||||
|
||||
it('构建存在时直接执行，且原样转发参数（分隔符交给 entry 处理）', () => {
|
||||
const dir = sandbox(true)
|
||||
const result = run(dir, ['--', '我和BOBO第一次聊了什么'], { TRACEMEMO_POC_ELECTRON: process.execPath })
|
||||
expect(result.status).toBe(0)
|
||||
const payload = JSON.parse(result.stdout.trim()) as { ranEntry: boolean; argv: string[] }
|
||||
expect(payload.ranEntry).toBe(true)
|
||||
// runner 不修改参数：leading `--` 保留，由 POC entry 决定是否移除。
|
||||
expect(payload.argv).toEqual(['--', '我和BOBO第一次聊了什么'])
|
||||
})
|
||||
|
||||
it('构建存在但启动器失败时返回非 0 退出码并给出提示', () => {
|
||||
const result = run(sandbox(true), ['问题'], { TRACEMEMO_POC_ELECTRON: '/nonexistent/electron-binary' })
|
||||
expect(result.status).toBe(1)
|
||||
expect(result.stderr).toContain('启动 Electron 失败')
|
||||
})
|
||||
})
|
||||
Reference in New Issue
Block a user