全栈安全加固、性能提速与架构演进：连接池与沙箱止血、双路混合检索、多版本归档及前端分包治理

- 第一批（安全加固与生命周期止血）：隔离测试宿主机密钥与 .env 泄漏；修复 SQLite 取消查询挂死、:memory: 连接池报错与 EngineCache 世代竞态；补齐运行与评测删除时的 Checkpointer 线程、Artifact 磁盘文件与无主附件回收；加固三套沙箱路径越界、进程组清理与 Seatbelt 数据目录隔离，封堵 SSRF IPv4-mapped IPv6 元数据绕过与响应体 OOM；修复推理模型多轮 reasoning_content、结构化输出 Token 漏计与缓存折扣计费
- 第二批（数据库/集群/问数提速与首屏分包）：补齐核心表索引与 SQLite WAL 调优，列表接口按需 defer 大列，收敛启动扫描并打通跨进程 slot 槽位唤醒；优化跨源全局选表、哈希桶数字绑定、流式 CSV 导入及多方言分析算子；剥离首屏静态引入的 @xyflow/react 与全局样式（首屏主包降至 461.59 kB），补充流式 rAF 攒批、Store 精细订阅与表单无障碍绑定
- 第三批（架构演进与工程自动化闭环）：记录库超挂接阈值历史版本自动增量归档至单库多表（解除 SQLite 10 库 ATTACH 硬上限）；重构知识库为词面 BM25 与全集合归一化向量矩阵真·双路独立召回并支持游标分批重建；升级多方言感知 SQL 模板词法分析（E'...' / $$...$$ / q'[...]' / # / 嵌套注释）与 Decimal / bool / NaN 参数校验，消除 ledger 层对 api 路由层的反向依赖；拆分 3101 行 CaliberDrafts 页面为 7 个领域子模块，消除检查脚本本机路径硬编码并新增 GitHub Actions CI 门禁与三批回归测试套件

scripts/check-all.mjs

@@ -52,13 +52,13 @@ import {
   readFileSync, readdirSync, rmSync, writeFileSync,
 } from 'node:fs'
 import { createServer } from 'node:net'
-import { homedir } from 'node:os'
 import { fileURLToPath } from 'node:url'
 import { dirname, join, resolve } from 'node:path'
+import { resolveChrome, resolvePython } from './lib/env.mjs'
 
 const HERE = dirname(fileURLToPath(import.meta.url))
 const ROOT = dirname(HERE)
-const SYSTEM_CHROME = '/Applications/Google Chrome.app/Contents/MacOS/Google Chrome'
+const SYSTEM_CHROME = resolveChrome()
 
 // 先跑快的、不开浏览器的，再按页面从外壳往里跑；最后是连真数据的两项
 const ORDER = [
@@ -189,8 +189,8 @@ if (unknown.length) {
   process.exit(2)
 }
 const names = picked.length ? ORDER.filter((n) => picked.includes(n)) : ORDER
-// 纯静态的几项不用起服务
-const STATIC = new Set(['tokens', 'copy'])
+// 纯静态或内存转译单测项不用起前后端服务
+const STATIC = new Set(['tokens', 'copy', 'trace', 'decode'])
 const needsServers = names.some((n) => !STATIC.has(n))
 
 // ---------------------------------------------------------------- 收尾
@@ -421,11 +421,11 @@ async function warmUp(stacks) {
 let stacks = []
 if (stackCount > 0 && needsServers) {
   const dataSrc = process.env.CHECK_DATA_SRC && resolve(process.env.CHECK_DATA_SRC)
-  const python = process.env.CHECK_PYTHON ?? join(homedir(), 'miniforge3', 'envs', 'agentlab', 'bin', 'python')
+  const python = resolvePython({ requireBackend: true })
   const problems = []
   if (!dataSrc) problems.push('--stacks 要知道拿哪份数据：用 CHECK_DATA_SRC=<数据目录> 指定（里面有 agentlab.db 的那一层）')
   else if (!existsSync(join(dataSrc, 'agentlab.db'))) problems.push(`CHECK_DATA_SRC=${dataSrc} 里没有 agentlab.db`)
-  if (!existsSync(python)) problems.push(`找不到后端解释器 ${python}，用 CHECK_PYTHON 指定`)
+  if (!python) problems.push('找不到可用的后端 Python 解释器，请用 CHECK_PYTHON 指定')
   if (problems.length) {
     for (const p of problems) console.error(`✗ ${p}`)
     process.exit(2)

scripts/check-bundle.mjs

@@ -54,13 +54,13 @@ import { homedir, tmpdir } from 'node:os'
 import { dirname, join } from 'node:path'
 import { fileURLToPath, pathToFileURL } from 'node:url'
 import { chromium } from '../frontend/node_modules/playwright-core/index.mjs'
+import { resolveChrome, resolvePython } from './lib/env.mjs'
 
 const WEB = process.env.AGENTLAB_WEB ?? 'http://localhost:5273'
 const API = process.env.AGENTLAB_API ?? 'http://localhost:8000/api'
-const CHROME = process.env.CHROME_PATH ?? '/Applications/Google Chrome.app/Contents/MacOS/Google Chrome'
+const CHROME = resolveChrome()
 const BACKEND = join(dirname(dirname(fileURLToPath(import.meta.url))), 'backend')
-const PYTHON = [process.env.CHECK_PYTHON, join(homedir(), 'miniforge3/envs/agentlab/bin/python')]
-  .find((p) => p && existsSync(p))
+const PYTHON = resolvePython({ requireBackend: true })
 
 let failed = 0
 const check = (name, cond, detail = '') => {

scripts/check-copy.mjs

@@ -23,11 +23,11 @@
 // 确实该保留的写在 scripts/copy-allow.json：每条写明文件、原文片段、规则和理由。白名单里有一条
 // 已经对不上任何命中（原文改掉了、挪走了），也算不通过：白名单只收现在还需要的例外。
 import { spawnSync } from 'node:child_process'
-import { existsSync, readdirSync, readFileSync, realpathSync, statSync } from 'node:fs'
+import { readdirSync, readFileSync, realpathSync, statSync } from 'node:fs'
 import { createRequire } from 'node:module'
-import { homedir } from 'node:os'
 import { dirname, join, relative } from 'node:path'
 import { fileURLToPath, pathToFileURL } from 'node:url'
+import { resolvePython } from './lib/env.mjs'
 
 const HERE = dirname(fileURLToPath(import.meta.url))
 const ROOT = dirname(HERE)
@@ -332,8 +332,7 @@ for p, rel in files:
 print(json.dumps(out, ensure_ascii=False))
 `
 
-const PYTHON = [process.env.CHECK_PYTHON, join(homedir(), 'miniforge3/envs/agentlab/bin/python'), 'python3']
-  .find((p) => p && (p === 'python3' || existsSync(p)))
+const PYTHON = resolvePython()
 
 function backStrings(root, only = []) {
   const r = spawnSync(PYTHON, ['-c', PY_EXTRACT, root, ...only], { encoding: 'utf8', maxBuffer: 256 << 20 })

scripts/check-decode.mjs

@@ -9,38 +9,22 @@
 // 跑之前前端得起着（./scripts/dev.sh），默认连 5273。对别的实例（比如一份沙箱拷贝）跑时
 // 带上地址：AGENTLAB_WEB=http://localhost:<前端端口> node scripts/check-decode.mjs
 import { readFileSync } from 'node:fs'
+import { join } from 'node:path'
+import { createTsLoader, ROOT } from './lib/env.mjs'
 
 const WEB = process.env.AGENTLAB_WEB ?? 'http://localhost:5273'
-const root = new URL('..', import.meta.url).pathname
+const root = `${ROOT}/`
 const fixtures = JSON.parse(
-  readFileSync(`${root}frontend/src/run/__tests__/fixtures.json`, 'utf8'))
+  readFileSync(join(ROOT, 'frontend/src/run/__tests__/fixtures.json'), 'utf8'))
 
-// decode.ts 在运行时 import 了 lib/format、lib/terms。data: URL 里的模块没法解析
-// "/src/..." 这种路径，所以把依赖也各自转成 data: URL 再替换进去（同 check-trace）
-const loaded = new Map()
-async function load(path) {
-  if (loaded.has(path)) return loaded.get(path)
-  const res = await fetch(`${WEB}${path}?t=${Date.now()}`)
-  if (!res.ok) throw new Error(`${path} → HTTP ${res.status}`)
-  let code = await res.text()
-  const deps = new Set([...code.matchAll(/from\s+["'](\/src\/[^"'?]+)(?:\?[^"']*)?["']/g)]
-    .map((m) => m[1]))
-  for (const dep of deps) {
-    const url = await load(dep)
-    const pattern = new RegExp(`(["'])${dep.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')}(\\?[^"']*)?\\1`, 'g')
-    code = code.replace(pattern, JSON.stringify(url))
-  }
-  const url = 'data:text/javascript;base64,' + Buffer.from(code).toString('base64')
-  loaded.set(path, url)
-  return url
-}
+const load = await createTsLoader(WEB)
 
 let mod, synthetic
 try {
   mod = await import(await load('/src/run/decode.ts'))
   synthetic = await import(await load('/src/run/__tests__/synthetic.ts'))
 } catch (e) {
-  console.error(`✗ 拿不到转译结果（${WEB}）——前端没起？先跑 ./scripts/dev.sh\n  ${e.message}`)
+  console.error(`✗ 加载前端解码器模块失败（${WEB}）：\n  ${e.message}`)
   process.exit(1)
 }
 

scripts/check-trace.mjs

@@ -9,31 +9,15 @@
 // 跑之前前端得起着（./scripts/dev.sh），默认连 5273。对别的实例（比如一份沙箱拷贝）跑时
 // 带上地址：AGENTLAB_WEB=http://localhost:<前端端口> node scripts/check-trace.mjs
 import { readFileSync } from 'node:fs'
+import { join } from 'node:path'
+import { createTsLoader, ROOT } from './lib/env.mjs'
 
 const WEB = process.env.AGENTLAB_WEB ?? 'http://localhost:5273'
-const root = new URL('..', import.meta.url).pathname
+const root = `${ROOT}/`
 const fixtures = JSON.parse(
-  readFileSync(`${root}frontend/src/run/__tests__/fixtures.json`, 'utf8'))
+  readFileSync(join(ROOT, 'frontend/src/run/__tests__/fixtures.json'), 'utf8'))
 
-// trace.ts 在运行时 import 了 decode.ts 和 derive.ts。data: URL 里的模块没法
-// 解析 "/src/..." 这种路径，所以把依赖也各自转成 data: URL 再替换进去
-const loaded = new Map()
-async function load(path) {
-  if (loaded.has(path)) return loaded.get(path)
-  const res = await fetch(`${WEB}${path}?t=${Date.now()}`)
-  if (!res.ok) throw new Error(`${path} → HTTP ${res.status}`)
-  let code = await res.text()
-  const deps = new Set([...code.matchAll(/from\s+["'](\/src\/[^"'?]+)(?:\?[^"']*)?["']/g)]
-    .map((m) => m[1]))
-  for (const dep of deps) {
-    const url = await load(dep)
-    const pattern = new RegExp(`(["'])${dep.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')}(\\?[^"']*)?\\1`, 'g')
-    code = code.replace(pattern, JSON.stringify(url))
-  }
-  const url = 'data:text/javascript;base64,' + Buffer.from(code).toString('base64')
-  loaded.set(path, url)
-  return url
-}
+const load = await createTsLoader(WEB)
 
 let trace, derive, decode
 try {
@@ -41,7 +25,7 @@ try {
   derive = await import(await load('/src/run/derive.ts'))
   decode = await import(await load('/src/run/decode.ts'))
 } catch (e) {
-  console.error(`✗ 拿不到转译结果（${WEB}）——前端没起？先跑 ./scripts/dev.sh\n  ${e.message}`)
+  console.error(`✗ 加载前端运行态模块失败（${WEB}）：\n  ${e.message}`)
   process.exit(1)
 }
 const { emptyTrace, foldEvent, finalizeTrace, project, isSettled } = trace

scripts/lib/env.mjs

@@ -0,0 +1,142 @@
+import { spawnSync } from 'node:child_process'
+import { existsSync, readFileSync } from 'node:fs'
+import { homedir } from 'node:os'
+import { dirname, extname, join, resolve } from 'node:path'
+import { fileURLToPath, pathToFileURL } from 'node:url'
+
+export const ROOT = dirname(dirname( dirname(fileURLToPath(import.meta.url)) ))
+const FRONTEND = join(ROOT, 'frontend')
+
+/**
+ * 跨平台解析可用的 Python 解释器：
+ * 优先尊重 CHECK_PYTHON、当前激活的虚拟环境 / Conda 环境、仓库内 .venv，
+ * 再回退到常见开发路径与系统 PATH 中的 python3 / python，并实际试跑确认可执行。
+ */
+export function resolvePython({ requireBackend = false } = {}) {
+  const isWin = process.platform === 'win32'
+  const venvBin = (dir) => (isWin ? join(dir, 'Scripts', 'python.exe') : join(dir, 'bin', 'python'))
+  const candidates = [
+    process.env.CHECK_PYTHON,
+    process.env.VIRTUAL_ENV && venvBin(process.env.VIRTUAL_ENV),
+    process.env.CONDA_PREFIX && (isWin ? join(process.env.CONDA_PREFIX, 'python.exe') : venvBin(process.env.CONDA_PREFIX)),
+    venvBin(join(ROOT, '.venv')),
+    venvBin(join(ROOT, 'backend', '.venv')),
+    join(homedir(), 'miniforge3', 'envs', 'agentlab', 'bin', 'python'),
+    join(homedir(), 'miniconda3', 'envs', 'agentlab', 'bin', 'python'),
+    '/opt/homebrew/bin/python3',
+    '/usr/local/bin/python3',
+    'python3',
+    'python',
+  ].filter(Boolean)
+
+  const probe = requireBackend
+    ? 'import sys, app; assert sys.version_info >= (3, 12)'
+    : 'import sys; assert sys.version_info >= (3, 12)'
+  const cwd = requireBackend ? join(ROOT, 'backend') : ROOT
+
+  const seen = new Set()
+  for (const bin of candidates) {
+    if (seen.has(bin)) continue
+    seen.add(bin)
+    if ((bin.includes('/') || bin.includes('\\')) && !existsSync(bin)) continue
+    const res = spawnSync(bin, ['-c', probe], { cwd, stdio: 'ignore' })
+    if (res.status === 0) return bin
+  }
+  return process.env.CHECK_PYTHON ?? 'python3'
+}
+
+/** 跨平台解析 Chrome / Chromium 可执行文件路径 */
+export function resolveChrome() {
+  if (process.env.CHROME_PATH) return process.env.CHROME_PATH
+  const candidates = [
+    '/Applications/Google Chrome.app/Contents/MacOS/Google Chrome',
+    '/usr/bin/google-chrome',
+    '/usr/bin/google-chrome-stable',
+    '/usr/bin/chromium',
+    '/usr/bin/chromium-browser',
+    '/snap/bin/chromium',
+  ]
+  for (const p of candidates) {
+    if (existsSync(p)) return p
+  }
+  return candidates[0]
+}
+
+function resolveTsFile(fromDir, spec) {
+  const base = spec.startsWith('/src/')
+    ? join(FRONTEND, spec.slice(1))
+    : resolve(fromDir, spec)
+  const tries = extname(base)
+    ? [base]
+    : [base, `${base}.ts`, `${base}.tsx`, join(base, 'index.ts'), join(base, 'index.tsx')]
+  for (const candidate of tries) {
+    if (existsSync(candidate)) return candidate
+  }
+  throw new Error(`找不到模块 ${spec}（自 ${fromDir}）`)
+}
+
+/**
+ * 为 check-trace / check-decode 提供前端 TS 模块加载器：
+ * - 若本地或指定的 AGENTLAB_WEB 开发服务器可达，直接拉取其转译结果；
+ * - 若开发服务器未启动（如 CI、离线环境或单跑脚本），自动回退到 Vite 内置的 Oxc 内存转译，无需启服务或绑端口。
+ */
+export async function createTsLoader(webUrl = process.env.AGENTLAB_WEB ?? 'http://localhost:5273') {
+  let serverOnline = false
+  if (process.env.CHECK_OFFLINE !== '1') {
+    try {
+      const r = await fetch(`${webUrl}/`, { signal: AbortSignal.timeout(400) })
+      serverOnline = r.ok
+    } catch {
+      serverOnline = false
+    }
+  }
+
+  if (serverOnline) {
+    const loaded = new Map()
+    const loadFromWeb = async (modPath) => {
+      if (loaded.has(modPath)) return loaded.get(modPath)
+      const res = await fetch(`${webUrl}${modPath}?t=${Date.now()}`)
+      if (!res.ok) throw new Error(`${modPath} → HTTP ${res.status}`)
+      let code = await res.text()
+      const deps = new Set([...code.matchAll(/from\s+["'](\/src\/[^"'?]+)(?:\?[^"']*)?["']/g)].map((m) => m[1]))
+      for (const dep of deps) {
+        const url = await loadFromWeb(dep)
+        const pattern = new RegExp(`(["'])${dep.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')}(\\?[^"']*)?\\1`, 'g')
+        code = code.replace(pattern, JSON.stringify(url))
+      }
+      const url = 'data:text/javascript;base64,' + Buffer.from(code).toString('base64')
+      loaded.set(modPath, url)
+      return url
+    }
+    return loadFromWeb
+  }
+
+  const viteEntry = pathToFileURL(join(FRONTEND, 'node_modules', 'vite', 'dist', 'node', 'index.js')).href
+  const { transformWithOxc } = await import(viteEntry)
+  const loaded = new Map()
+
+  const loadLocalFile = async (absPath) => {
+    if (loaded.has(absPath)) return loaded.get(absPath)
+    const raw = readFileSync(absPath, 'utf8')
+    const lang = absPath.endsWith('.tsx') ? 'tsx' : 'ts'
+    const transformed = await transformWithOxc(raw, absPath, { lang, sourcemap: false })
+    let code = transformed.code
+    const dir = dirname(absPath)
+    const specs = new Set(
+      [...code.matchAll(/(?:import|export)\s+(?:[^"']*?\s+from\s+)?["']((?:\.\.?\/|\/src\/)[^"'?]+)(?:\?[^"']*)?["']/g)]
+        .map((m) => m[1]),
+    )
+    for (const spec of specs) {
+      const depAbs = resolveTsFile(dir, spec)
+      const depUrl = await loadLocalFile(depAbs)
+      const escaped = spec.replace(/[.*+?^${}()|[\]\\]/g, '\\$&')
+      const pattern = new RegExp(`(["'])${escaped}(\\?[^"']*)?\\1`, 'g')
+      code = code.replace(pattern, JSON.stringify(depUrl))
+    }
+    const url = 'data:text/javascript;base64,' + Buffer.from(code).toString('base64')
+    loaded.set(absPath, url)
+    return url
+  }
+
+  return async (modPath) => loadLocalFile(resolveTsFile(FRONTEND, modPath))
+}

scripts/package-docker.sh

@@ -6,7 +6,7 @@ set -euo pipefail
 usage() {
   cat <<'USAGE'
 用法：./scripts/package-docker.sh [--platform linux/arm64|linux/amd64] [--version <版本>] [--out <目录>] [--gzip] [--no-microvm]
-                                  [--no-postgres]
+                                  [--no-postgres] [--keep <数量>]
 
   --platform <平台>  目标机的平台。默认按本机架构取 linux/<arch>。64 位系统的 ARM 单板机是 linux/arm64，
                      普通 PC 和服务器是 linux/amd64。和本机架构不同时 docker 用模拟构建，会慢很多。
@@ -18,6 +18,7 @@ usage() {
   --no-postgres      不带内置 PostgreSQL 的镜像（postgres-image.tar，约 160 MB）。只用 SQLite 或外接 PostgreSQL 时用不上；
                      不带的包在目标机上 install --postgres 要先 docker pull 到同一个标签。默认带上，标签取 deploy.sh 里的
                      POSTGRES_IMAGE（只在那一处定）。
+  --keep <数量>      在产物目录里最多保留最近几份同架构的离线包（默认取 AGENTLAB_KEEP_RELEASES，未设时为 3；传 0 不清理旧包）。
   -h, --help         显示这段说明。
 
 产物：<out>/agentlab-<版本>-linux-<arch>.tar，里面有 image.tar、postgres-image.tar、deploy.sh、agentlab.env、README.md、
@@ -34,6 +35,7 @@ OUT="$ROOT/release"
 GZIP=""
 WITH_MICROVM=1
 WITH_POSTGRES=1
+KEEP_RELEASES="${AGENTLAB_KEEP_RELEASES:-3}"
 need_value() { [ $# -ge 2 ] && [ -n "$2" ] || { echo "$1 后面要跟一个值"; echo; usage; exit 2; }; }
 while [ $# -gt 0 ]; do
   case "$1" in
@@ -46,11 +48,16 @@ while [ $# -gt 0 ]; do
     --gzip) GZIP=1 ;;
     --no-microvm) WITH_MICROVM=0 ;;
     --no-postgres) WITH_POSTGRES=0 ;;
+    --keep) need_value "$@"; KEEP_RELEASES="$2"; shift ;;
+    --keep=*) KEEP_RELEASES="${1#--keep=}" ;;
     -h|--help) usage; exit 0 ;;
     *) echo "认不出的参数：$1"; echo; usage; exit 2 ;;
   esac
   shift
 done
+case "$KEEP_RELEASES" in
+  ''|*[!0-9]*) echo "--keep 需要是非负整数：$KEEP_RELEASES"; exit 2 ;;
+esac
 
 # ---- 平台 ----
 if [ -z "$PLATFORM" ]; then
@@ -183,6 +190,29 @@ if [ -n "$GZIP" ]; then
   if command -v pigz >/dev/null 2>&1; then pigz -c "$TAR" > "$TAR.gz"; else gzip -c "$TAR" > "$TAR.gz"; fi
 fi
 
+# ---- 4. 清理旧离线包 ----
+if [ "$KEEP_RELEASES" -gt 0 ]; then
+  SEEN_STEMS=" $NAME "
+  KEPT_COUNT=1
+  for OLD_FILE in $(ls -1t "$OUT"/agentlab-*-linux-"$ARCH".tar "$OUT"/agentlab-*-linux-"$ARCH".tar.gz 2>/dev/null || true); do
+    [ -f "$OLD_FILE" ] || continue
+    BASE_NAME="$(basename "$OLD_FILE")"
+    STEM="${BASE_NAME%.gz}"
+    STEM="${STEM%.tar}"
+    case "$SEEN_STEMS" in
+      *" $STEM "*) ;;
+      *)
+        SEEN_STEMS="$SEEN_STEMS$STEM "
+        KEPT_COUNT=$((KEPT_COUNT + 1))
+        ;;
+    esac
+    if [ "$KEPT_COUNT" -gt "$KEEP_RELEASES" ] && [ "$STEM" != "$NAME" ]; then
+      echo "==> 清理旧离线包 $OLD_FILE（保留最近 $KEEP_RELEASES 份）"
+      rm -f "$OLD_FILE"
+    fi
+  done
+fi
+
 size() { du -h "$1" | awk '{print $1}'; }
 echo
 echo "完成："