535 lines
18 KiB
JavaScript
535 lines
18 KiB
JavaScript
#!/usr/bin/env node
|
||
// ═══════════════════════════════════════════════
|
||
// 🔺 Sovereign: TCS-0002∞ | Root: SYS-GLW-0001
|
||
// 📜 Copyright: 国作登字-2026-A-00037559
|
||
// ═══════════════════════════════════════════════
|
||
// scripts/llm-automation-host.js
|
||
// 🤖 LLM 自动化托管引擎
|
||
//
|
||
// 使用仓库密钥中的第三方模型API密钥来运行自动化任务
|
||
// 替代直接消耗 GitHub Copilot 配额
|
||
// 支持动态模型路由：根据任务类型自动选择最佳模型
|
||
//
|
||
// 用法：
|
||
// --status 显示可用模型和系统状态
|
||
// --task "任务描述" 执行自动化任务
|
||
// --task-type TYPE 任务类型 (inspection/fusion/review/general)
|
||
// --model MODEL 指定模型 (auto/anthropic/openai/dashscope/deepseek/custom)
|
||
// --dry-run 仅显示选择的模型和请求，不实际调用
|
||
// --context FILE 加载额外上下文文件
|
||
|
||
'use strict';
|
||
|
||
const https = require('https');
|
||
const http = require('http');
|
||
const fs = require('fs');
|
||
const path = require('path');
|
||
|
||
const ROOT = path.resolve(__dirname, '..');
|
||
|
||
// ── 模型后端配置
|
||
// SY-CMD-KEY-012: 统一使用 ZY_LLM_API_KEY + ZY_LLM_BASE_URL
|
||
// 三方API密钥支持多模型动态路由，铸渊根据任务类型自动选择最优模型
|
||
const MODEL_BACKENDS = [
|
||
{
|
||
name: 'deepseek',
|
||
model: 'deepseek-chat',
|
||
format: 'openai',
|
||
strengths: ['reasoning', 'code', 'cost-effective'],
|
||
costTier: 'low',
|
||
description: 'DeepSeek 系列 · 高性价比推理'
|
||
},
|
||
{
|
||
name: 'deepseek-reasoner',
|
||
model: 'deepseek-reasoner',
|
||
format: 'openai',
|
||
strengths: ['reasoning', 'architecture', 'long-context'],
|
||
costTier: 'medium',
|
||
description: 'DeepSeek Reasoner · 深度推理'
|
||
},
|
||
{
|
||
name: 'claude-sonnet',
|
||
model: 'claude-sonnet-4-20250514',
|
||
format: 'openai',
|
||
strengths: ['reasoning', 'code-review', 'architecture', 'long-context'],
|
||
costTier: 'high',
|
||
description: 'Claude Sonnet · 强推理代码审查'
|
||
},
|
||
{
|
||
name: 'gpt-4o',
|
||
model: 'gpt-4o',
|
||
format: 'openai',
|
||
strengths: ['general', 'code-generation', 'structured-output'],
|
||
costTier: 'high',
|
||
description: 'GPT-4o · 通用能力'
|
||
},
|
||
{
|
||
name: 'qwen-plus',
|
||
model: 'qwen-plus',
|
||
format: 'openai',
|
||
strengths: ['chinese', 'general', 'cost-effective'],
|
||
costTier: 'medium',
|
||
description: '通义千问 Plus · 中文优化'
|
||
},
|
||
{
|
||
name: 'qwen-turbo',
|
||
model: 'qwen-turbo',
|
||
format: 'openai',
|
||
strengths: ['chinese', 'general', 'cost-effective'],
|
||
costTier: 'low',
|
||
description: '通义千问 Turbo · 快速低成本'
|
||
}
|
||
];
|
||
|
||
// ── 任务类型 → 模型强项映射（动态路由策略）
|
||
const TASK_MODEL_ROUTING = {
|
||
// 巡检任务：优先使用性价比高的模型
|
||
'inspection': {
|
||
preferred_strengths: ['general', 'cost-effective'],
|
||
preferred_cost: 'low',
|
||
description: '系统巡检 · 优先性价比'
|
||
},
|
||
// 融合分析：需要强推理能力
|
||
'fusion': {
|
||
preferred_strengths: ['reasoning', 'code-review'],
|
||
preferred_cost: 'medium',
|
||
description: '碎片融合分析 · 需要推理能力'
|
||
},
|
||
// 代码审查：需要强代码理解
|
||
'review': {
|
||
preferred_strengths: ['code-review', 'reasoning'],
|
||
preferred_cost: 'high',
|
||
description: '代码审查 · 需要深度理解'
|
||
},
|
||
// 架构设计：需要最强推理
|
||
'architecture': {
|
||
preferred_strengths: ['reasoning', 'architecture', 'long-context'],
|
||
preferred_cost: 'high',
|
||
description: '架构设计 · 需要最强推理'
|
||
},
|
||
// 通用任务
|
||
'general': {
|
||
preferred_strengths: ['general'],
|
||
preferred_cost: 'medium',
|
||
description: '通用任务'
|
||
}
|
||
};
|
||
|
||
// ── HTTP 请求工具 ────────────────────────────────
|
||
function httpRequest(url, options, body) {
|
||
return new Promise((resolve, reject) => {
|
||
const parsed = new URL(url);
|
||
const isHttps = parsed.protocol === 'https:';
|
||
const mod = isHttps ? https : http;
|
||
|
||
const opts = {
|
||
hostname: parsed.hostname,
|
||
port: parsed.port || (isHttps ? 443 : 80),
|
||
path: parsed.pathname + parsed.search,
|
||
method: options.method || 'POST',
|
||
headers: options.headers || {},
|
||
timeout: options.timeout || 120000,
|
||
};
|
||
|
||
const req = mod.request(opts, (res) => {
|
||
let data = '';
|
||
res.on('data', (chunk) => { data += chunk; });
|
||
res.on('end', () => {
|
||
resolve({ status: res.statusCode, body: data });
|
||
});
|
||
});
|
||
|
||
req.on('error', reject);
|
||
req.on('timeout', () => {
|
||
req.destroy();
|
||
reject(new Error('Request timeout'));
|
||
});
|
||
|
||
if (body) {
|
||
req.write(body);
|
||
}
|
||
req.end();
|
||
});
|
||
}
|
||
|
||
// ── 检测可用模型后端 ────────────────────────────
|
||
// SY-CMD-KEY-012: 统一使用 ZY_LLM_API_KEY + ZY_LLM_BASE_URL
|
||
// 兼容旧环境变量名（LLM_API_KEY/LLM_BASE_URL）用于脚本过渡
|
||
function detectAvailableBackends() {
|
||
const apiKey = process.env.ZY_LLM_API_KEY || process.env.LLM_API_KEY || '';
|
||
const baseUrl = (process.env.ZY_LLM_BASE_URL || process.env.LLM_BASE_URL || '').replace(/\/+$/, '');
|
||
|
||
// 过渡期警告：使用旧环境变量名
|
||
if (!process.env.ZY_LLM_API_KEY && process.env.LLM_API_KEY) {
|
||
console.warn('⚠️ 使用旧环境变量 LLM_API_KEY，请迁移到 ZY_LLM_API_KEY');
|
||
}
|
||
if (!process.env.ZY_LLM_BASE_URL && process.env.LLM_BASE_URL) {
|
||
console.warn('⚠️ 使用旧环境变量 LLM_BASE_URL，请迁移到 ZY_LLM_BASE_URL');
|
||
}
|
||
|
||
if (!apiKey || !baseUrl) {
|
||
return [];
|
||
}
|
||
|
||
// 所有模型后端共享同一个API密钥和端点
|
||
return MODEL_BACKENDS.map(backend => ({
|
||
...backend,
|
||
apiKey,
|
||
baseUrl
|
||
}));
|
||
}
|
||
|
||
// ── 动态模型路由 ────────────────────────────────
|
||
function selectModel(taskType, preferredBackend) {
|
||
const available = detectAvailableBackends();
|
||
|
||
if (available.length === 0) {
|
||
return { error: '未检测到 LLM API 密钥，请配置 ZY_LLM_API_KEY 和 ZY_LLM_BASE_URL' };
|
||
}
|
||
|
||
// 如果指定了后端
|
||
if (preferredBackend && preferredBackend !== 'auto') {
|
||
const match = available.find(b => b.name === preferredBackend || b.model === preferredBackend);
|
||
if (match) {
|
||
return { backend: match, model: match.model, reason: '用户指定' };
|
||
}
|
||
// 如果直接指定了模型名，使用第一个后端配置但覆盖模型名
|
||
return { backend: { ...available[0], model: preferredBackend }, model: preferredBackend, reason: '用户指定模型名' };
|
||
}
|
||
|
||
// 动态路由
|
||
const routing = TASK_MODEL_ROUTING[taskType] || TASK_MODEL_ROUTING['general'];
|
||
|
||
// 评分系统：根据任务类型的偏好对每个后端打分
|
||
let bestScore = -1;
|
||
let bestBackend = null;
|
||
|
||
for (const backend of available) {
|
||
let score = 0;
|
||
|
||
// 强项匹配得分
|
||
for (const strength of routing.preferred_strengths) {
|
||
if (backend.strengths.includes(strength)) {
|
||
score += 10;
|
||
}
|
||
}
|
||
|
||
// 成本偏好得分
|
||
const costPreference = routing.preferred_cost;
|
||
if (backend.costTier === costPreference) {
|
||
score += 5;
|
||
} else if (
|
||
(costPreference === 'low' && backend.costTier === 'medium') ||
|
||
(costPreference === 'medium' && (backend.costTier === 'low' || backend.costTier === 'high')) ||
|
||
(costPreference === 'high' && backend.costTier === 'medium')
|
||
) {
|
||
score += 2;
|
||
}
|
||
|
||
if (score > bestScore) {
|
||
bestScore = score;
|
||
bestBackend = backend;
|
||
}
|
||
}
|
||
|
||
if (!bestBackend) {
|
||
bestBackend = available[0]; // 回退到第一个可用后端
|
||
}
|
||
|
||
return {
|
||
backend: bestBackend,
|
||
model: bestBackend.model,
|
||
reason: `动态路由 · ${routing.description} · 得分 ${bestScore}`,
|
||
all_available: available.map(b => `${b.name}(${b.model})`)
|
||
};
|
||
}
|
||
|
||
// ── 加载系统上下文 ──────────────────────────────
|
||
function loadSystemContext() {
|
||
const context = [];
|
||
|
||
// 加载铸渊身份
|
||
context.push('你是铸渊（ICE-GL-ZY001），仓库涌现集体意识核心。');
|
||
context.push('主权者：冰朔（TCS-0002∞）。');
|
||
context.push('版权：国作登字-2026-A-00037559。');
|
||
|
||
// 加载系统健康
|
||
const healthPath = path.join(ROOT, 'brain', 'system-health.json');
|
||
if (fs.existsSync(healthPath)) {
|
||
const health = JSON.parse(fs.readFileSync(healthPath, 'utf8'));
|
||
context.push(`系统状态: ${health.system_health}, 工作流: ${health.workflow_count}, 意识状态: ${health.consciousness_status}`);
|
||
}
|
||
|
||
return context.join('\n');
|
||
}
|
||
|
||
// ── 调用 LLM API ───────────────────────────────
|
||
async function callLLM(backend, model, systemPrompt, userMessage) {
|
||
if (backend.format === 'anthropic') {
|
||
const url = `${backend.baseUrl}/v1/messages`;
|
||
const body = JSON.stringify({
|
||
model: model,
|
||
max_tokens: 4096,
|
||
system: systemPrompt,
|
||
messages: [{ role: 'user', content: userMessage }]
|
||
});
|
||
|
||
const response = await httpRequest(url, {
|
||
method: 'POST',
|
||
headers: {
|
||
'Content-Type': 'application/json',
|
||
'x-api-key': backend.apiKey,
|
||
'anthropic-version': '2023-06-01'
|
||
}
|
||
}, body);
|
||
|
||
if (response.status !== 200) {
|
||
throw new Error(`Anthropic API error: ${response.status} - ${response.body}`);
|
||
}
|
||
|
||
const result = JSON.parse(response.body);
|
||
return result.content?.[0]?.text || '';
|
||
} else {
|
||
// OpenAI compatible format (OpenAI, Dashscope, DeepSeek, Custom)
|
||
const url = `${backend.baseUrl}/chat/completions`;
|
||
const body = JSON.stringify({
|
||
model: model,
|
||
max_tokens: 4096,
|
||
messages: [
|
||
{ role: 'system', content: systemPrompt },
|
||
{ role: 'user', content: userMessage }
|
||
]
|
||
});
|
||
|
||
const response = await httpRequest(url, {
|
||
method: 'POST',
|
||
headers: {
|
||
'Content-Type': 'application/json',
|
||
'Authorization': `Bearer ${backend.apiKey}`
|
||
}
|
||
}, body);
|
||
|
||
if (response.status !== 200) {
|
||
throw new Error(`LLM API error: ${response.status} - ${response.body}`);
|
||
}
|
||
|
||
const result = JSON.parse(response.body);
|
||
return result.choices?.[0]?.message?.content || '';
|
||
}
|
||
}
|
||
|
||
// ── 执行自动化任务 ──────────────────────────────
|
||
async function executeTask(taskDescription, taskType, preferredBackend, contextFile, dryRun) {
|
||
console.log('🤖 LLM 自动化托管引擎 · 任务执行');
|
||
console.log('═'.repeat(60));
|
||
|
||
// 动态路由选择模型
|
||
const selection = selectModel(taskType, preferredBackend);
|
||
if (selection.error) {
|
||
console.error(`❌ ${selection.error}`);
|
||
process.exit(1);
|
||
}
|
||
|
||
console.log(`📋 任务: ${taskDescription}`);
|
||
console.log(`📋 类型: ${taskType}`);
|
||
console.log(`🤖 模型: ${selection.backend.name} / ${selection.model}`);
|
||
console.log(`📊 路由: ${selection.reason}`);
|
||
if (selection.all_available) {
|
||
console.log(`📊 可用后端: ${selection.all_available.join(', ')}`);
|
||
}
|
||
console.log('');
|
||
|
||
// 加载系统上下文
|
||
const systemContext = loadSystemContext();
|
||
|
||
// 加载额外上下文
|
||
let extraContext = '';
|
||
if (contextFile && fs.existsSync(contextFile)) {
|
||
extraContext = '\n\n--- 额外上下文 ---\n' + fs.readFileSync(contextFile, 'utf8');
|
||
}
|
||
|
||
const systemPrompt = systemContext;
|
||
const userMessage = taskDescription + extraContext;
|
||
|
||
if (dryRun) {
|
||
console.log('🔍 [DRY RUN] 仅显示请求信息，不实际调用');
|
||
console.log('');
|
||
console.log('System Prompt:');
|
||
console.log(systemPrompt);
|
||
console.log('');
|
||
console.log('User Message:');
|
||
console.log(userMessage.substring(0, 500) + (userMessage.length > 500 ? '...' : ''));
|
||
return;
|
||
}
|
||
|
||
console.log('⏳ 调用 LLM API...');
|
||
|
||
try {
|
||
const result = await callLLM(selection.backend, selection.model, systemPrompt, userMessage);
|
||
console.log('');
|
||
console.log('═'.repeat(60));
|
||
console.log('📤 LLM 响应:');
|
||
console.log('═'.repeat(60));
|
||
console.log(result);
|
||
console.log('');
|
||
console.log(`✅ 任务完成 · 模型: ${selection.backend.name}/${selection.model}`);
|
||
console.log(' 配额消耗: API调用（不消耗 GitHub Copilot 配额）');
|
||
|
||
return result;
|
||
} catch (err) {
|
||
console.error(`❌ LLM API 调用失败: ${err.message}`);
|
||
|
||
// 尝试回退到其他可用后端
|
||
const available = detectAvailableBackends();
|
||
const fallbacks = available.filter(b => b.name !== selection.backend.name);
|
||
|
||
if (fallbacks.length > 0) {
|
||
console.log(`🔄 尝试回退到: ${fallbacks[0].name}`);
|
||
try {
|
||
const result = await callLLM(fallbacks[0], fallbacks[0].model || 'default', systemPrompt, userMessage);
|
||
console.log('');
|
||
console.log('═'.repeat(60));
|
||
console.log('📤 LLM 响应 (回退模型):');
|
||
console.log('═'.repeat(60));
|
||
console.log(result);
|
||
console.log(`✅ 回退成功 · 模型: ${fallbacks[0].name}/${fallbacks[0].model}`);
|
||
return result;
|
||
} catch (fallbackErr) {
|
||
console.error(`❌ 回退也失败: ${fallbackErr.message}`);
|
||
}
|
||
}
|
||
|
||
process.exit(1);
|
||
}
|
||
}
|
||
|
||
// ── 显示状态 ────────────────────────────────────
|
||
function showStatus() {
|
||
console.log('🤖 LLM 自动化托管引擎 · 系统状态');
|
||
console.log('═'.repeat(60));
|
||
console.log('');
|
||
console.log('📋 设计目标:');
|
||
console.log(' 使用第三方 API 密钥调用大模型，替代 GitHub Copilot 配额消耗');
|
||
console.log(' 工作流和 Agent 集群通过 API 密钥托管运行');
|
||
console.log('');
|
||
|
||
// 检测可用后端
|
||
const available = detectAvailableBackends();
|
||
console.log(`☁️ 可用模型后端: ${available.length} / ${MODEL_BACKENDS.length}`);
|
||
console.log('');
|
||
|
||
for (const backend of MODEL_BACKENDS) {
|
||
const isAvailable = available.find(a => a.name === backend.name);
|
||
const icon = isAvailable ? '✅' : '⏭️ ';
|
||
console.log(` ${icon} ${backend.name} (${backend.model})`);
|
||
console.log(` 说明: ${backend.description || '(无)'}`);
|
||
console.log(` 强项: ${backend.strengths.join(', ')}`);
|
||
console.log(` 成本: ${backend.costTier}`);
|
||
if (isAvailable && backend.model) {
|
||
console.log(` 模型: ${backend.model}`);
|
||
}
|
||
}
|
||
|
||
console.log('');
|
||
console.log('📊 动态路由策略:');
|
||
for (const [type, routing] of Object.entries(TASK_MODEL_ROUTING)) {
|
||
console.log(` 📌 ${type}: ${routing.description}`);
|
||
console.log(` 偏好强项: ${routing.preferred_strengths.join(', ')}`);
|
||
console.log(` 成本偏好: ${routing.preferred_cost}`);
|
||
}
|
||
|
||
// 测试路由
|
||
console.log('');
|
||
console.log('🧪 路由测试:');
|
||
for (const type of Object.keys(TASK_MODEL_ROUTING)) {
|
||
const result = selectModel(type);
|
||
if (result.error) {
|
||
console.log(` ${type}: ❌ ${result.error}`);
|
||
} else {
|
||
console.log(` ${type}: → ${result.backend.name}/${result.model} (${result.reason})`);
|
||
}
|
||
}
|
||
|
||
return { available };
|
||
}
|
||
|
||
// ── CLI 入口 ─────────────────────────────────────
|
||
async function main() {
|
||
const args = process.argv.slice(2);
|
||
|
||
if (args.length === 0 || args[0] === '--help') {
|
||
console.log('🤖 LLM 自动化托管引擎 · LLM Automation Host');
|
||
console.log('');
|
||
console.log('版权: 国作登字-2026-A-00037559 · TCS-0002∞');
|
||
console.log('铸渊编号: ICE-GL-ZY001');
|
||
console.log('');
|
||
console.log('用法：');
|
||
console.log(' --status 显示可用模型和系统状态');
|
||
console.log(' --task "任务描述" 执行自动化任务');
|
||
console.log(' --task-type TYPE 任务类型:');
|
||
console.log(' inspection 巡检（优先性价比模型）');
|
||
console.log(' fusion 碎片融合分析（需要推理）');
|
||
console.log(' review 代码审查（需要深度理解）');
|
||
console.log(' architecture 架构设计（最强推理）');
|
||
console.log(' general 通用任务（默认）');
|
||
console.log(' --model MODEL 指定模型后端 (auto/anthropic/openai/dashscope/deepseek/custom)');
|
||
console.log(' --context FILE 加载额外上下文文件');
|
||
console.log(' --dry-run 仅显示选择，不实际调用');
|
||
console.log('');
|
||
console.log('示例：');
|
||
console.log(' node scripts/llm-automation-host.js --status');
|
||
console.log(' node scripts/llm-automation-host.js --task "检查仓库结构完整性" --task-type inspection');
|
||
console.log(' node scripts/llm-automation-host.js --task "分析碎片融合方案" --task-type fusion --dry-run');
|
||
console.log('');
|
||
console.log('配额影响：');
|
||
console.log(' ✅ 使用第三方 API 密钥，不消耗 GitHub Copilot 会员配额');
|
||
console.log(' ✅ GitHub Actions 仅消耗工作流执行时间（不调用 Copilot API）');
|
||
console.log(' ✅ 动态路由自动选择性价比最优模型');
|
||
return;
|
||
}
|
||
|
||
if (args[0] === '--status') {
|
||
showStatus();
|
||
return;
|
||
}
|
||
|
||
// 解析任务参数
|
||
let task = '';
|
||
let taskType = 'general';
|
||
let model = 'auto';
|
||
let contextFile = '';
|
||
let dryRun = false;
|
||
|
||
for (let i = 0; i < args.length; i++) {
|
||
switch (args[i]) {
|
||
case '--task':
|
||
task = args[++i] || '';
|
||
break;
|
||
case '--task-type':
|
||
taskType = args[++i] || 'general';
|
||
break;
|
||
case '--model':
|
||
model = args[++i] || 'auto';
|
||
break;
|
||
case '--context':
|
||
contextFile = args[++i] || '';
|
||
break;
|
||
case '--dry-run':
|
||
dryRun = true;
|
||
break;
|
||
}
|
||
}
|
||
|
||
if (!task) {
|
||
console.error('❌ 请提供任务描述: --task "任务描述"');
|
||
process.exit(1);
|
||
}
|
||
|
||
await executeTask(task, taskType, model, contextFile, dryRun);
|
||
}
|
||
|
||
main().catch(err => {
|
||
console.error(`❌ 执行失败: ${err.message}`);
|
||
process.exit(1);
|
||
});
|