feat: v0.7.3 成本收口 · 状态一致 · 死账清理 — Prompt Cache 根治 + SSRF DNS Pinning + 87 用例扩充全量回归
CI / 类型检查 + Lint + 单元测试 (push) Failing after 5m47s
CI / 全量测试 (Electron ABI) (push) Failing after 5m19s
CI / 产物编译验证 (push) Successful in 9m55s

P1 修复面收口: Prompt Cache 根治(日期/记忆/附件三类易变内容出 system 入用户消息
  前置块 user-context.ts, system 跨 run 字节级稳定; Anthropic system 块数组化 +
  cache_control ephemeral 断言, DeepSeek 自动缓存前缀命中 — 多轮对话输入 token
  成本降数量级); 编辑重发/重新生成幽灵 Trace 双侧根治(DB truncateMessagesAfter
  同步过滤 metadata.traceSteps + 前端 trimTraceStepsByAnchor 镜像, 严格小于锚点
  时间戳, 同毫秒等值判废); sessions:deleteMessage 死通道全链路删除(渲染层零调用
  + message_count 漂移面); Ollama vision 能力门控全链路(MetonaModelInfo
  .supportsVision 贯穿 adapter/IPC/store/UI, model-capabilities.ts 三道判定纯函数,
  未知保守放行); 记忆固化节流(consolidation-policy 纯函数: 总开关 + 内容门控
  [回答>=200字符或存在成功工具调用] + 会话级 10 分钟频率窗口, 三 memory.* 配置键)

P2 安全纵深: SSRF DNS Pinning 关闭 rebinding 窗口(ssrf-guard 重构
  resolvePublicAddresses 单源; ssrf-dispatcher 以 undici Agent.connect.lookup
  钉死校验 IP, TLS SNI 保持原域名, 一次性 dispatcher 用后即毁; 代理激活显式
  退化为仅入口校验); web_fetch 重写手动逐跳重定向循环(每跳先校验后连接,
  替代 redirect:follow 内核跟跳的中间跳裸奔, 上限 5 跳); http_request 换用
  pinned fetch; web_search 可达性预检加固(私有 URL 零请求 + 不跟跳, 3xx 视为
  可达); Agent 浏览器 CORS 通配收紧为 Origin 回显 + Vary: Origin;
  ConfirmationHook.forgetSession 会话终态清理(会话删除/abort 联动/SubAgent
  终结三处接线, 根治 rememberedDecisions 泄漏)

P3 架构还债: agent.enableReflection 死配置全链路接线(main→shared→引擎→
  Orchestrator→设置开关, REFLECTING 状态真实可达); AgentLoopConfig.timeoutMs
  死字段删除; MemoryManager.cleanupExpired 挂入健康检查周期(expires_at 回收
  管道真实化); buildSafeEnv 收敛 utils/safe-env.ts 单源(run_command 与 MCP
  stdio 共用, 终结双实现漂移); Trace 生命周期治理(metadata 只保留最近 20 个
  run — keepRecentRuns 纯函数; JSONL 录制启动自动清理保留 200 个 + 设置页
  手动清理); SLO/健康快照可视化(app:healthSnapshot IPC + 设置页只读卡片 +
  审计链一键校验)

P4 能力演进: 会话标题 LLM 自动生成(TitleGenerator — 每会话幂等/并发重入复用
  同一 Promise/自定义标题不覆盖/失败静默回退, Sidebar 经 config:changed 实时
  刷新); MCP 自动重连(5s/15s/60s 退避最多 3 次, reconnecting 状态机,
  teardownConnection 内部拆除保留簿记 — 用户断开/开关关闭即时取消, 设置页
  显示第 N/3 次); 死循环检测 ABAB 乒乓模式(最近4轮 A→B→A→B 交替判定, 补齐
  docs 第五章"两状态反复切换"检测契约); i18n 第三阶段(ChatInput/LLMSettings/
  OnboardingWizard/MemoryViewer 主链路文案出层, zh-CN + en-US 双字典补齐)

测试: 737 → 824 用例(+87, 新增 8 个测试文件 + 扩展 3 个)。新覆盖: user-context
  分组/空值收缩/拼接契约、context-builder 字节级稳定性、Anthropic cache_control
  四态、consolidation-policy 九路判定矩阵、ssrf-dispatcher(pinned lookup/重定向
  解析/IP 校验)、forget-session 会话隔离、trace-lifecycle run 淘汰、
  trace-trim 严格小于边界、safe-env 净化矩阵、mcp-reconnect 退避状态机
  (fake timers)、title-generator 并发重入、SQLite 侧 truncate×TRACE 联动
  (Electron ABI)。测试驱动修复: GIT_*/ 注释终止块注释、重连计数被自身重试
  前置断开重置(拆 teardownConnection 保留簿记)、TitleGenerator 幂等占位与
  并发去重的检查顺序竞态(去重先于幂等)

版本: 0.7.3; README 同步(配置表新增 agent.enableReflection/memory.*/mcp.autoReconnect)

回归: typecheck 双端 0 错误; ESLint 0/0; 系统 Node 771 通过 53 跳过
  (better-sqlite3 ABI); Electron ABI 全量 824/824 零跳过
This commit is contained in:
2026-08-30 09:44:43 +08:00
parent 26169b7be4
commit ebe45482b0
68 changed files with 4568 additions and 664 deletions
+110 -64
View File
@@ -19,8 +19,24 @@ import { MetonaErrorCode, MetonaStreamEventType } from '../harness/types';
import { estimateMessagesTokens } from '../harness/utils/token-estimator';
import { DeepSeekAdapter } from '../harness/adapters/deepseek.adapter';
import { OllamaAdapter } from '../harness/adapters/ollama.adapter';
// v0.7.3 P1-1: 用户上下文前置块(动态内容出 system,保 prompt cache 前缀稳定)
import { buildUserContextPrefix, withUserContextPrefix } from '../harness/prompts/user-context';
// v0.7.3 P1-5: 记忆固化触发决策(纯函数)
import { shouldConsolidate } from '../harness/memory/consolidation-policy';
import log from 'electron-log';
/** 构建 "时区名 (UTC±N)" 标签(注入用户上下文前置块;失败回退 UTC) */
function buildTimezoneLabel(): string {
try {
const tz = Intl.DateTimeFormat().resolvedOptions().timeZone ?? 'UTC';
const offset = -new Date().getTimezoneOffset() / 60;
const offsetStr = offset >= 0 ? `UTC+${offset}` : `UTC${offset}`;
return `${tz} (${offsetStr})`;
} catch {
return 'UTC';
}
}
/** 单会话的 text_delta 节流状态 */
interface ThrottleState {
buffer: string;
@@ -56,12 +72,16 @@ export function registerAgentHandlers(ctx: IPCContext): void {
sessionSummaryService,
orchestrator,
confirmationHook,
titleGenerator,
} = ctx;
// ===== 常驻事件管道:text_delta 按会话节流(F8 =====
const throttleStates = new Map<string, ThrottleState>();
const iterationTraces = new Map<string, IterationTrace>();
// v0.7.3 P1-5: 会话级记忆固化时间戳(consolidation-policy 频率门控的状态持有方)
const lastConsolidationBySession = new Map<string, number>();
const flushThrottle = (sessionId: string): void => {
const st = throttleStates.get(sessionId);
if (!st) return;
@@ -411,59 +431,46 @@ export function registerAgentHandlers(ctx: IPCContext): void {
});
}
// 检索与用户消息相关的记忆,注入到 System Prompt 动态区
// 检索与用户消息相关的记忆 + 附件元信息 → 构建用户消息上下文前置块。
// v0.7.3 P1-1 根治: 记忆注入与附件提示此前追加进 systemPrompt.dynamicReminders
// 每条消息都改变 system 字节 → 跨 run 缓存全 miss。现随首条 user 消息注入
// LLM 语义等价),system prompt 保持跨 run 字节级稳定。
let userContextPrefix = '';
try {
const memories = memoryManager.search(userMessage.content, { topK: 5, minImportance: 0.3 });
const attachments = (
userMessage as MetonaMessage & {
attachments?: Array<{ name: string; type: string; truncated?: boolean }>;
}
).attachments;
userContextPrefix = buildUserContextPrefix({
now: Date.now(),
memories,
attachments: Array.isArray(attachments) ? attachments : [],
timezoneLabel: buildTimezoneLabel(),
});
if (memories.length > 0) {
const memorySection = memories
.map(
(m, i) =>
`[${i + 1}] (${m.type}, 重要度: ${m.importance.toFixed(1)}) ${m.content.slice(0, 200)}`,
)
.join('\n');
const memoryBlock = `## Relevant Memories (Retrieved)\n${memorySection}`;
systemPrompt.dynamicReminders = systemPrompt.dynamicReminders
? `${systemPrompt.dynamicReminders}\n\n---\n\n${memoryBlock}`
: memoryBlock;
log.debug(`[AGENT] Injected ${memories.length} memories into system prompt`);
log.debug(`[AGENT] Injected ${memories.length} memories into user context prefix`);
}
} catch (err) {
log.warn('[AGENT] Memory retrieval failed, proceeding without memories:', err);
// 记忆检索失败时前置块退化为仅含日期时间(附件提示随之丢失可接受——
// 主进程附件提示是辅助语义,附件内容本体仍在消息中)
userContextPrefix = buildUserContextPrefix({
now: Date.now(),
memories: [],
attachments: [],
timezoneLabel: buildTimezoneLabel(),
});
}
// 附件提示注入:用户直接上传的文件/图片,避免 LLM 误以为需要在工作空间查找
const attachments = (
userMessage as MetonaMessage & { attachments?: Array<{ name: string; type: string }> }
).attachments;
if (Array.isArray(attachments) && attachments.length > 0) {
const attachmentList = attachments
.map((att, i) => {
const typeLabel =
att.type === 'image' ? 'image' : att.type === 'text' ? 'text file' : 'file';
// v0.7.2 A5: 文本附件被上传入口截断(512KB 上限)时,明确告知 LLM 内容不完整,
// 防止模型把残缺内容当作完整文件事实
const truncatedNote =
(att as { truncated?: boolean }).truncated === true
? ' (TRUNCATED — only the first 512KB is included; the full content is NOT available)'
: '';
const note =
att.type === 'image'
? 'already provided to you via vision capability — you can SEE it directly, do NOT call view_image or any tool to read it again'
: att.type === 'text'
? `content${truncatedNote} already inlined in the user message, do NOT search in workspace or read it again`
: 'uploaded directly by user, do NOT search in workspace';
return `${i + 1}. [${typeLabel}] ${att.name}${note}`;
})
.join('\n');
const attachmentBlock = `## User Attachments (Direct Upload)\nThe following files were uploaded directly by the user to this conversation. They are inline attachments, NOT workspace files:\n${attachmentList}\n\n**IMPORTANT**: Images listed above are already visible to you in this conversation. Do NOT call \`view_image\`, \`read_file\`, or any file tool to read them — doing so wastes a tool call and may fail (they are not workspace files).`;
systemPrompt.dynamicReminders = systemPrompt.dynamicReminders
? `${systemPrompt.dynamicReminders}\n\n---\n\n${attachmentBlock}`
: attachmentBlock;
log.debug(`[AGENT] Injected ${attachments.length} attachment hints into system prompt`);
}
// 构建发送给引擎的用户消息副本 —— 前置块只存在于该副本:
// DB 持久化(上方 saveMessage 已用原始内容)、前端展示、记忆固化、
// 注入检测均使用原始干净内容,互不污染。
const engineUserMessage: MetonaMessage = {
...userMessage,
content: withUserContextPrefix(userContextPrefix, userMessage.content),
};
try {
// 提示注入检测(安全模块)
@@ -503,8 +510,9 @@ export function registerAgentHandlers(ctx: IPCContext): void {
});
// 启动 Agent LoopP2-10: 每会话独立引擎)
// v0.7.3 P1-1: 传入带上下文前置块的消息副本 —— 原始 userMessage 保持干净
const engine = agentEngineManager.getEngine(sessionId);
const output = await engine.runStream(userMessage, sessionId, history, systemPrompt);
const output = await engine.runStream(engineUserMessage, sessionId, history, systemPrompt);
// 输出验证(不阻塞响应,仅记录警告)
// v0.3.0 修复: 传入 toolResults 和 context,启用事实一致性检查和幻觉检测
@@ -624,23 +632,58 @@ export function registerAgentHandlers(ctx: IPCContext): void {
workspaceService.updateMemoryTimestamp();
// 会话结束:AI 判断本次对话有哪些重要内容需要持久化到 MEMORY.md
// 异步执行,不阻塞主流程返回;失败仅记录日志
memoryConsolidator
.consolidate(userMessage.content, output.finalAnswer, output.iterations)
.then((result) => {
if (result.appended > 0) {
log.info(
`[AGENT] Memory consolidated: ${result.appended} entries appended to MEMORY.md`,
);
broadcast('toast:show', {
type: 'info',
message: `AI 已将 ${result.appended} 条重要记忆写入 MEMORY.md`,
});
}
})
.catch((err) => {
log.warn('[AGENT] Memory consolidation failed:', err);
});
// v0.7.3 P1-5 节流: 此前每次 run 无条件发起固化 LLM 请求,短寒暄同样触发。
// 现按 consolidation-policy 决策(总开关 + 内容门控 + 会话级频率窗口)触发;
// 异步执行不阻塞主流程返回;失败仅记录日志。
const lastConsolidationAt = lastConsolidationBySession.get(sessionId) ?? 0;
const hadSuccessfulToolCall = output.iterations.some((step) =>
(step.toolResults ?? []).some((r) => r.success),
);
const decision = shouldConsolidate({
enabled: configService.get<boolean>('memory.consolidationEnabled') !== false,
answerChars: output.finalAnswer?.length ?? 0,
minChars: configService.get<number>('memory.consolidationMinChars') ?? 200,
hadSuccessfulToolCall,
lastConsolidationAt,
now: Date.now(),
intervalMs: configService.get<number>('memory.consolidationIntervalMs') ?? 600_000,
});
if (decision.consolidate) {
memoryConsolidator
.consolidate(userMessage.content, output.finalAnswer, output.iterations)
.then((result) => {
if (result.appended > 0) {
lastConsolidationBySession.set(sessionId, Date.now());
log.info(
`[AGENT] Memory consolidated: ${result.appended} entries appended to MEMORY.md`,
);
broadcast('toast:show', {
type: 'info',
message: `AI 已将 ${result.appended} 条重要记忆写入 MEMORY.md`,
});
}
})
.catch((err) => {
log.warn('[AGENT] Memory consolidation failed:', err);
});
} else {
log.debug(`[AGENT] Memory consolidation skipped (${decision.reason})`);
}
// v0.7.3 P4-1: 首个完成的 run 之后生成精炼会话标题(每会话幂等,失败静默)
if (output.terminationReason === 'completed') {
titleGenerator
.maybeGenerateTitle(sessionId, userMessage.content, output.finalAnswer)
.then((title) => {
if (title) {
// 广播重命名结果,前端 Sidebar 实时刷新标题
broadcast('config:changed', { key: `session.title.${sessionId}`, value: title });
}
})
.catch(() => {
/* 静默 —— 标题失败已有 debug 日志 */
});
}
// TOOL 层:记录会话结束 / TRACE 层:停止录制
auditService.logSessionEnd({
@@ -832,8 +875,9 @@ export function registerAgentHandlers(ctx: IPCContext): void {
// v0.5.0: 按会话清理 — 只拒绝被中断会话的 pending,不影响其他并发会话等待中的确认
confirmationHook.clearPending(sessionId);
// v0.5.1: 被中止 SubAgent 的 pending 确认一并拒绝(含 SubAgent 递归派生的孙任务)
// v0.7.3 P2-3: 被中止的 SubAgent 是会话终态 —— 用 forgetSession 连决策记忆一并清理
for (const taskId of abortedTaskIds) {
confirmationHook.clearPending(taskId);
confirmationHook.forgetSession(taskId);
}
// TOOL 层:记录中断
@@ -880,6 +924,8 @@ export function registerAgentHandlers(ctx: IPCContext): void {
});
subTraces.delete(taskId);
subMeta.delete(taskId);
// v0.7.3 P2-3: SubAgent 终态 —— 决策记忆随任务终结清理(防长期运行泄漏)
confirmationHook.forgetSession(taskId);
};
orchestrator.on(