fix: harden hwlab cd cache

2026-05-24 05:59:29 +00:00
parent b517a08b95
commit bfdfd10923
3 changed files with 932 additions and 45 deletions
@@ -51,9 +51,11 @@ bun scripts/cli.ts hwlab cd preflight --env dev
 bun scripts/cli.ts hwlab cd apply --env dev --dry-run
 ```

-wrapper 的职责是把 host commander 常用的 HWLAB DEV rollout 查看/准备动作收敛到单一入口。默认路径通过 UniDesk main-server frontend/backend-core 的 provider `host.ssh` 能力进入 D601，再在 D601 上执行有界检查脚本；它只调用 HWLAB repo-owned 受控脚本，不在 UniDesk 内重写发布流程或拼接 ad hoc `kubectl apply`：
+wrapper 的职责是把 host commander 常用的 HWLAB DEV rollout 查看/准备动作收敛到单一入口。默认路径通过 UniDesk main-server frontend/backend-core 的 provider `host.ssh` 能力进入 D601；frontend transport 先用多个短 `host.ssh` 命令把远端 runner 分块上传到 `/tmp/unidesk-hwlab-cd/`，再在 D601 上执行有界检查脚本，避免 provider-gateway 的短命令长度限制成为审计入口单点故障。经 frontend/backend-core 返回的 stdout 只允许承载短 JSON 摘要；完整结果必须写入 D601 `~/.state/unidesk-hwlab-cd/<run-id>/result.full.json`，防止 backend JSON 安全摘要把 stdout 截断后导致 CLI 失明。它只调用 HWLAB repo-owned 受控脚本，不在 UniDesk 内重写发布流程或拼接 ad hoc `kubectl apply`：

- 默认 HWLAB CD repo 是 D601 固定干净 mirror `/home/ubuntu/hwlab_cd`，也可用 `--hwlab-repo` 显式指定同等干净 clone。wrapper 必须检查 `git status --short --branch`、origin remote、当前 branch `main`、本地 `origin/main`、`FETCH_HEAD` 和 worktree 权限；任何 dirty worktree、错误 remote、非 main、HEAD 未跟上本地 `origin/main` 或权限异常都返回结构化 blocker。`/home/ubuntu/hwlab` 是 runner 历史目录，不得作为发布真相。
+- 默认 HWLAB CD repo 是每次运行在 D601 `~/.state/unidesk-hwlab-cd/<run-id>/ephemeral-repo/HWLAB` 新建的一次性 clone。一次性 repo 只从专用 full bare cache `~/.cache/unidesk/hwlab-cd/git-cache/HWLAB.git` 拷贝生成，必须使用 `--no-hardlinks` 形成独立 `.git` 对象库，不能通过 `--shared` 或 alternates 在运行时依赖 cache 对象库。cache 本身只允许作为 HWLAB CD 专用 Git 加速源，不承载 runner workspace、部署副本、手工改代码、报告留痕或任何其他用途。
+- 专用 cache 由当前 `host.ssh` 执行用户拥有，`~/.cache/unidesk/hwlab-cd` 必须保持 owner-only 权限；owner 不匹配、不可写或权限收紧失败时返回结构化 blocker。cache 初始化可以从本机已有 HWLAB clone seed 一次性拷贝，但 seed 只能用于降低首次出网成本，不能作为 release truth。cache remote 固定优先 `git@github.com:pikasTech/HWLAB.git`，GitHub 出网通过 D601 provider egress proxy `127.0.0.1:18789` 和 SSH `ProxyCommand`。cache 必须保存 full bare repo 历史，而不是 depth=1 浅缓存；wrapper 需要能解析 `refs/heads/main:deploy/deploy.json` 指向的 promotion commit，浅缓存即使 main HEAD 一致也不能证明 CD desired-state 完整。若 GitHub refresh 失败但本地 full cache 同时满足 main HEAD 和 deploy promotion commit，可作为 stale-cache 诊断继续读数，但必须暴露 `hwlab-cache-refresh` blocker，不能伪装成 release-truth PASS。
+- 一次性 repo 是当前 `host.ssh` 执行用户创建的临时发布读数，避免长期 `/home/ubuntu/hwlab_cd` 被 root 或其他 runner owner 污染后成为 CD 单点故障。`--hwlab-repo` 或 `UNIDESK_HWLAB_REPO` 仍可显式指定同等干净 clone 供人工诊断；被指定的 clone 必须由执行用户拥有并可读写。root-owned 或其他用户拥有的 clone 会触发 Git dubious ownership、`FETCH_HEAD` 不可写和 `.git/worktrees` 权限 blocker。修复 owner 污染应重建或显式修正 mirror 权限，不能用 `git config --global safe.directory` 掩盖发布真相污染。wrapper 必须检查 `git status --short --branch`、origin remote、当前 branch `main`、本地 `origin/main`、`FETCH_HEAD` 和 worktree 权限；任何 dirty worktree、错误 remote、非 main、HEAD 未跟上本地 `origin/main` 或权限异常都返回结构化 blocker。`/home/ubuntu/hwlab` 是 runner 历史目录，不得作为发布真相。
 - `deploy/deploy.json` 是唯一 desired-state。wrapper 只把 `deploy/artifact-catalog.dev.json`、`deploy/k8s/base/workloads.yaml` 和 `reports/dev-gate/dev-artifacts.json` 当作派生/证据读数；`status`/`preflight` 必须显示 target commit/ref、deploy.json、artifact catalog、workloads 和 live workload image 是否同源/收敛，不引入第二套 desired state。
 - `status` 只读汇总 HWLAB repo path、Git clean/main/origin-main、desired-state 收敛、D601 native k3s guard 和 `Lease/hwlab-dev/hwlab-dev-cd-lock`；同时调用 HWLAB `scripts/dev-cd-apply.mjs --status --skip-live-verify` 取得 repo-owned target/promotion/deploy.json/artifact 摘要。16666/16667 live verification 不由本 runner 执行。
 - `audit` 是 DEV CD 恢复后的只读健康审计，不是验收 gate 或报告生成器。它在 `status` 受控路径上补充只读 `kubectl get`/HTTP health probes，输出有界 JSON summary，分类 `control-plane-unavailable`、`docker-desktop-context-risk`、`second-control-plane-risk`、`workspace-unavailable`、`dirty-worktree`、`secret-missing`、`registry-unavailable`、`lease-held`、`lease-stale-candidate`、`artifact-missing`、`artifact-mismatch`、`runtime-job-blocked`、`rollout-unhealthy`、`public-tunnel-unhealthy` 和 `db-runtime-durability-risk`。audit 只显示 Secret 对象/key 是否存在，不显示值；只读判断 Lease 是否 stale，不释放或 break；只读拉取 16666/16667 `/health/live` 的 commit/readiness 摘要，不把它当作 M3 DEV-LIVE 验收。
@@ -8,8 +8,14 @@ const path = require("node:path");
 const namespace = "hwlab-dev";
 const lockName = "hwlab-dev-cd-lock";
 const nativeKubeconfig = "/etc/rancher/k3s/k3s.yaml";
-const defaultRepo = "/home/ubuntu/hwlab_cd";
+const legacyDefaultRepo = "/home/ubuntu/hwlab_cd";
 const rejectedRepo = "/home/ubuntu/hwlab";
+const defaultRemoteCandidates = ["git@github.com:pikasTech/HWLAB.git"];
+const dedicatedCacheRoot = path.join(os.homedir(), ".cache", "unidesk", "hwlab-cd");
+const dedicatedCacheRepo = path.join(os.homedir(), ".cache", "unidesk", "hwlab-cd", "git-cache", "HWLAB.git");
+const defaultEgressProxy = "http://127.0.0.1:18789";
+const defaultNoProxy = "localhost,127.0.0.1,::1,host.docker.internal,d601-provider-egress-proxy,d601-provider-egress-proxy.unidesk,d601-provider-egress-proxy.unidesk.svc,d601-provider-egress-proxy.unidesk.svc.cluster.local";
+const seedCacheRepos = ["/home/ubuntu/hwlab", "/home/ubuntu/workspace/hwlab", "/tmp/hwlab-dev-cd-35bbbee-host"];
 const requiredNodeName = "d601";
 const tailChars = 1400;
 const parseCaptureLimitBytes = 1024 * 1024;
@@ -169,25 +175,279 @@ function accessCheck(targetPath, mode) {
  }
 }

-function resolveRepo(provided) {
-  const rawPath = provided || process.env.UNIDESK_HWLAB_REPO || defaultRepo;
+function repoRecord(rawPath, source, extra = {}) {
  const absolutePath = path.resolve(rawPath);
  const rejected = absolutePath === rejectedRepo;
  const devCdApply = path.join(absolutePath, "scripts/dev-cd-apply.mjs");
  const deployJson = path.join(absolutePath, "deploy/deploy.json");
  return {
    status: !rejected && fs.existsSync(absolutePath) && fs.existsSync(devCdApply) && fs.existsSync(deployJson) ? "selected" : "blocked",
-    source: provided ? "option" : process.env.UNIDESK_HWLAB_REPO ? "env:UNIDESK_HWLAB_REPO" : "default:d601-clean-mirror",
+    source,
    path: absolutePath,
-    defaultPath: defaultRepo,
+    defaultPath: legacyDefaultRepo,
+    defaultMode: "ephemeral-clone",
    rejected,
    rejectionReason: rejected ? "runner-history-directory-is-not-hwlab-cd-release-truth" : null,
    exists: fs.existsSync(absolutePath),
    hasDevCdApply: fs.existsSync(devCdApply),
    hasDeployJson: fs.existsSync(deployJson),
+    ...extra,
  };
 }

+function remoteCandidates() {
+  const fromEnv = process.env.UNIDESK_HWLAB_REMOTE_URL;
+  if (typeof fromEnv === "string" && fromEnv.trim().length > 0) return [fromEnv.trim()];
+  return defaultRemoteCandidates;
+}
+
+function seedRepoCandidates() {
+  const fromEnv = process.env.UNIDESK_HWLAB_CACHE_SEED_REPOS;
+  const configured = typeof fromEnv === "string" && fromEnv.trim().length > 0
+    ? fromEnv.split(":").map((item) => item.trim()).filter(Boolean)
+    : seedCacheRepos;
+  return [...new Set(configured.map((item) => path.resolve(item)))];
+}
+
+function egressEnv() {
+  const proxy = process.env.UNIDESK_HWLAB_EGRESS_PROXY || defaultEgressProxy;
+  const noProxy = process.env.NO_PROXY || process.env.no_proxy || defaultNoProxy;
+  const proxyHostPort = proxy.replace(/^https?:\/\//u, "");
+  const gitSshCommand = process.env.GIT_SSH_COMMAND || [
+    "ssh",
+    "-o", "BatchMode=yes",
+    "-o", "StrictHostKeyChecking=accept-new",
+    "-o", `"ProxyCommand=nc -X connect -x ${proxyHostPort} %h %p"`,
+  ].join(" ");
+  return {
+    ...process.env,
+    HTTP_PROXY: proxy,
+    HTTPS_PROXY: proxy,
+    ALL_PROXY: proxy,
+    http_proxy: proxy,
+    https_proxy: proxy,
+    all_proxy: proxy,
+    NO_PROXY: noProxy,
+    no_proxy: noProxy,
+    GIT_SSH_COMMAND: gitSshCommand,
+  };
+}
+
+async function egressProxyHealth(dumpDir, timeoutMs) {
+  const proxy = process.env.UNIDESK_HWLAB_EGRESS_PROXY || defaultEgressProxy;
+  const healthUrl = `${proxy.replace(/\/$/u, "")}/__unidesk/egress-proxy/health`;
+  const command = await runCaptured(["curl", "-fsS", "--max-time", "3", healthUrl], os.homedir(), dumpDir, "egress-proxy-health", { env: egressEnv(), timeoutMs: Math.min(timeoutMs, 5000) });
+  const body = parseJson(command.stdoutText);
+  return {
+    status: command.ok && asRecord(body)?.connected === true ? "pass" : "blocked",
+    proxy,
+    healthUrl,
+    connected: asRecord(body)?.connected ?? null,
+    command: commandView(command),
+  };
+}
+
+function pathOwner(targetPath) {
+  try {
+    const stat = fs.statSync(targetPath);
+    return { uid: stat.uid, gid: stat.gid, mode: (stat.mode & 0o777).toString(8), isDirectory: stat.isDirectory() };
+  } catch (error) {
+    return { uid: null, gid: null, mode: null, isDirectory: false, error: error instanceof Error ? error.message : String(error) };
+  }
+}
+
+function cacheRepoRecord(cachePath) {
+  const absolutePath = path.resolve(cachePath);
+  const gitPath = path.join(absolutePath, ".git");
+  const bareHead = path.join(absolutePath, "HEAD");
+  const bareObjects = path.join(absolutePath, "objects");
+  const shallowPath = path.join(absolutePath, "shallow");
+  const exists = fs.existsSync(gitPath) || (fs.existsSync(bareHead) && fs.existsSync(bareObjects));
+  const readable = accessCheck(absolutePath, fs.constants.R_OK | fs.constants.X_OK);
+  const writable = accessCheck(absolutePath, fs.constants.R_OK | fs.constants.W_OK | fs.constants.X_OK);
+  const owner = pathOwner(absolutePath);
+  return { path: absolutePath, exists, readable: readable.ok, writable: writable.ok, owner, shallow: fs.existsSync(shallowPath), error: readable.error || writable.error || null };
+}
+
+async function deployJsonCommitFromCache(dumpDir, timeoutMs) {
+  const command = await runCaptured(["git", "--git-dir", dedicatedCacheRepo, "show", "refs/heads/main:deploy/deploy.json"], os.homedir(), dumpDir, "git-dedicated-cache-deploy-json", { timeoutMs: Math.min(timeoutMs, 5000) });
+  if (!command.ok) return { commitId: null, command: commandView(command) };
+  const parsed = parseJson(command.stdoutText);
+  return { commitId: stringValue(asRecord(parsed)?.commitId), command: commandView(command) };
+}
+
+async function cacheCommitExists(commitId, dumpDir, timeoutMs, idPrefix = "git-dedicated-cache-commit") {
+  if (!commitId) return { exists: false, command: null };
+  const command = await runCaptured(["git", "--git-dir", dedicatedCacheRepo, "rev-parse", "--verify", `${commitId}^{commit}`], os.homedir(), dumpDir, idPrefix, { timeoutMs: Math.min(timeoutMs, 5000) });
+  return { exists: command.ok, resolvedCommit: command.stdoutText.trim() || null, command: commandView(command) };
+}
+
+async function refreshDedicatedCache(remoteUrl, dumpDir, timeoutMs) {
+  const env = egressEnv();
+  const parent = path.dirname(dedicatedCacheRepo);
+  await fsp.mkdir(parent, { recursive: true, mode: 0o700 });
+  await fsp.chmod(dedicatedCacheRoot, 0o700).catch(() => undefined);
+  await fsp.chmod(path.dirname(dedicatedCacheRoot), 0o700).catch(() => undefined);
+  const attempts = [];
+  const existing = cacheRepoRecord(dedicatedCacheRepo);
+  if (existing.exists && (!existing.writable || existing.owner.uid !== process.getuid?.())) {
+    attempts.push({ stage: "dedicated-cache-owner", ok: false, cache: existing });
+    return { ok: false, cache: existing, attempts, stage: "dedicated-cache-owner" };
+  }
+  if (existing.exists) {
+    const permissions = await runCaptured(["chmod", "-R", "go-rwx", dedicatedCacheRoot], os.homedir(), dumpDir, "git-dedicated-cache-permissions", { timeoutMs: Math.min(timeoutMs, 5000) });
+    attempts.push({ stage: "cache-permissions", ok: permissions.ok, command: commandView(permissions) });
+    if (!permissions.ok) return { ok: false, cache: cacheRepoRecord(dedicatedCacheRepo), attempts, stage: "cache-permissions" };
+  }
+
+  if (!existing.exists) {
+    for (const seed of seedRepoCandidates().map(cacheRepoRecord)) {
+      if (!seed.exists || !seed.readable) {
+        attempts.push({ stage: "seed-unavailable", ok: false, seed });
+        continue;
+      }
+      await fsp.rm(dedicatedCacheRepo, { recursive: true, force: true });
+      const cloneSeed = await runCaptured(["git", "clone", "--bare", seed.path, dedicatedCacheRepo], parent, dumpDir, "git-dedicated-cache-seed-clone", { timeoutMs: Math.min(timeoutMs, 15000) });
+      attempts.push({ stage: "seed-clone", ok: cloneSeed.ok, seed, command: commandView(cloneSeed) });
+      if (cloneSeed.ok) break;
+    }
+    if (!fs.existsSync(dedicatedCacheRepo)) {
+      const init = await runCaptured(["git", "init", "--bare", dedicatedCacheRepo], parent, dumpDir, "git-dedicated-cache-init", { timeoutMs: Math.min(timeoutMs, 5000) });
+      attempts.push({ stage: "bare-init", ok: init.ok, command: commandView(init) });
+      if (!init.ok) return { ok: false, cache: cacheRepoRecord(dedicatedCacheRepo), attempts, stage: "bare-init" };
+    }
+    const permissions = await runCaptured(["chmod", "-R", "go-rwx", dedicatedCacheRoot], os.homedir(), dumpDir, "git-dedicated-cache-permissions", { timeoutMs: Math.min(timeoutMs, 5000) });
+    attempts.push({ stage: "cache-permissions", ok: permissions.ok, command: commandView(permissions) });
+    if (!permissions.ok) return { ok: false, cache: cacheRepoRecord(dedicatedCacheRepo), attempts, stage: "cache-permissions" };
+  }
+
+  const remoteSet = await runCaptured(["git", "--git-dir", dedicatedCacheRepo, "remote", "set-url", "origin", remoteUrl], os.homedir(), dumpDir, "git-dedicated-cache-remote-set-url", { timeoutMs: Math.min(timeoutMs, 5000) });
+  const remoteAdd = remoteSet.ok ? null : await runCaptured(["git", "--git-dir", dedicatedCacheRepo, "remote", "add", "origin", remoteUrl], os.homedir(), dumpDir, "git-dedicated-cache-remote-add", { timeoutMs: Math.min(timeoutMs, 5000) });
+  attempts.push({ stage: "remote-set-url", ok: remoteSet.ok || remoteAdd?.ok === true, command: commandView(remoteSet), fallback: remoteAdd ? commandView(remoteAdd) : null });
+  if (!remoteSet.ok && remoteAdd?.ok !== true) return { ok: false, cache: cacheRepoRecord(dedicatedCacheRepo), attempts, stage: "remote-set-url" };
+
+  const remoteHead = await runCaptured(["git", "--git-dir", dedicatedCacheRepo, "ls-remote", "--heads", "origin", "main"], os.homedir(), dumpDir, "git-dedicated-cache-ls-remote-main", { env, timeoutMs: Math.min(timeoutMs, 15000) });
+  const remoteMainCommit = remoteHead.stdoutText.trim().split(/\s+/u)[0] || null;
+  attempts.push({ stage: "ls-remote-main", ok: remoteHead.ok && Boolean(remoteMainCommit), remoteMainCommit, command: commandView(remoteHead) });
+  const localHead = await runCaptured(["git", "--git-dir", dedicatedCacheRepo, "rev-parse", "refs/heads/main"], os.homedir(), dumpDir, "git-dedicated-cache-local-main", { timeoutMs: Math.min(timeoutMs, 5000) });
+  const localMainCommit = localHead.stdoutText.trim() || null;
+  attempts.push({ stage: "local-main", ok: localHead.ok && Boolean(localMainCommit), localMainCommit, command: commandView(localHead) });
+  const deployJson = localHead.ok ? await deployJsonCommitFromCache(dumpDir, timeoutMs) : { commitId: null, command: null };
+  attempts.push({ stage: "deploy-json-commit", ok: Boolean(deployJson.commitId), commitId: deployJson.commitId, command: deployJson.command });
+  const deployCommit = await cacheCommitExists(deployJson.commitId, dumpDir, timeoutMs, "git-dedicated-cache-deploy-commit");
+  attempts.push({ stage: "deploy-commit-present", ok: deployCommit.exists, commitId: deployJson.commitId, resolvedCommit: deployCommit.resolvedCommit, command: deployCommit.command });
+  const cacheRecordBeforeFetch = cacheRepoRecord(dedicatedCacheRepo);
+  if (remoteHead.ok && remoteMainCommit && localHead.ok && localMainCommit === remoteMainCommit && deployCommit.exists && cacheRecordBeforeFetch.shallow !== true) {
+    return { ok: true, cache: cacheRecordBeforeFetch, attempts, stage: "ready", refreshStatus: "current-full", remoteMainCommit, localMainCommit, deployCommitId: deployJson.commitId };
+  }
+
+  const fetchArgs = ["git", "--git-dir", dedicatedCacheRepo, "fetch", "--prune", "--tags", "origin", "+refs/heads/*:refs/heads/*"];
+  if (cacheRecordBeforeFetch.shallow === true) fetchArgs.splice(4, 0, "--unshallow");
+  const fetch = await runCaptured(fetchArgs, os.homedir(), dumpDir, "git-dedicated-cache-fetch-full", { env, timeoutMs: Math.min(timeoutMs, 55000) });
+  attempts.push({ stage: "fetch-main", ok: fetch.ok, command: commandView(fetch) });
+  if (!fetch.ok) {
+    if (localHead.ok && localMainCommit && deployCommit.exists) {
+      return { ok: true, cache: cacheRepoRecord(dedicatedCacheRepo), attempts, stage: "stale-cache", refreshStatus: "blocked", remoteMainCommit, localMainCommit, deployCommitId: deployJson.commitId };
+    }
+    return { ok: false, cache: cacheRepoRecord(dedicatedCacheRepo), attempts, stage: "fetch-full", refreshStatus: "blocked", remoteMainCommit, localMainCommit, deployCommitId: deployJson.commitId };
+  }
+  const fetchedLocalHead = await runCaptured(["git", "--git-dir", dedicatedCacheRepo, "rev-parse", "refs/heads/main"], os.homedir(), dumpDir, "git-dedicated-cache-local-main-after-fetch", { timeoutMs: Math.min(timeoutMs, 5000) });
+  const fetchedLocalMainCommit = fetchedLocalHead.stdoutText.trim() || localMainCommit;
+  attempts.push({ stage: "local-main-after-fetch", ok: fetchedLocalHead.ok && Boolean(fetchedLocalMainCommit), localMainCommit: fetchedLocalMainCommit, command: commandView(fetchedLocalHead) });
+  const fetchedDeployJson = await deployJsonCommitFromCache(dumpDir, timeoutMs);
+  attempts.push({ stage: "deploy-json-commit-after-fetch", ok: Boolean(fetchedDeployJson.commitId), commitId: fetchedDeployJson.commitId, command: fetchedDeployJson.command });
+  const fetchedDeployCommit = await cacheCommitExists(fetchedDeployJson.commitId, dumpDir, timeoutMs, "git-dedicated-cache-deploy-commit-after-fetch");
+  attempts.push({ stage: "deploy-commit-present-after-fetch", ok: fetchedDeployCommit.exists, commitId: fetchedDeployJson.commitId, resolvedCommit: fetchedDeployCommit.resolvedCommit, command: fetchedDeployCommit.command });
+  const permissions = await runCaptured(["chmod", "-R", "go-rwx", dedicatedCacheRoot], os.homedir(), dumpDir, "git-dedicated-cache-permissions-after-fetch", { timeoutMs: Math.min(timeoutMs, 5000) });
+  attempts.push({ stage: "cache-permissions-after-fetch", ok: permissions.ok, command: commandView(permissions) });
+  if (!permissions.ok) return { ok: false, cache: cacheRepoRecord(dedicatedCacheRepo), attempts, stage: "cache-permissions-after-fetch", refreshStatus: "blocked", remoteMainCommit, localMainCommit: fetchedLocalMainCommit, deployCommitId: fetchedDeployJson.commitId };
+  if (!fetchedDeployCommit.exists) return { ok: false, cache: cacheRepoRecord(dedicatedCacheRepo), attempts, stage: "deploy-commit-unresolved", refreshStatus: "blocked", remoteMainCommit, localMainCommit: fetchedLocalMainCommit, deployCommitId: fetchedDeployJson.commitId };
+  return { ok: true, cache: cacheRepoRecord(dedicatedCacheRepo), attempts, stage: "ready", refreshStatus: "refreshed-full", remoteMainCommit, localMainCommit: fetchedLocalMainCommit, deployCommitId: fetchedDeployJson.commitId };
+}
+
+async function materializeFromCache(cachePath, repoPath, remoteUrl, dumpDir, timeoutMs) {
+  await fsp.rm(repoPath, { recursive: true, force: true });
+  const clone = await runCaptured(["git", "clone", "--no-hardlinks", "--no-checkout", cachePath, repoPath], path.dirname(repoPath), dumpDir, "git-cache-clone", { timeoutMs: Math.min(timeoutMs, 15000) });
+  if (!clone.ok) return { ok: false, stage: "cache-clone", commands: [commandView(clone)] };
+  const remote = await runCaptured(["git", "remote", "set-url", "origin", remoteUrl], repoPath, dumpDir, "git-cache-remote-set-url", { timeoutMs: Math.min(timeoutMs, 5000) });
+  if (!remote.ok) return { ok: false, stage: "remote-set-url", commands: [commandView(clone), commandView(remote)] };
+  const checkout = await runCaptured(["git", "checkout", "-B", "main", "origin/main"], repoPath, dumpDir, "git-cache-checkout-main", { timeoutMs: Math.min(timeoutMs, 5000) });
+  const reset = checkout.ok
+    ? await runCaptured(["git", "reset", "--hard", "origin/main"], repoPath, dumpDir, "git-cache-reset-hard", { timeoutMs: Math.min(timeoutMs, 5000) })
+    : null;
+  const clean = checkout.ok && reset?.ok
+    ? await runCaptured(["git", "clean", "-ffdqx"], repoPath, dumpDir, "git-cache-clean", { timeoutMs: Math.min(timeoutMs, 5000) })
+    : null;
+  const upstream = checkout.ok && reset?.ok
+    ? await runCaptured(["git", "branch", "--set-upstream-to=origin/main", "main"], repoPath, dumpDir, "git-cache-upstream", { timeoutMs: Math.min(timeoutMs, 5000) })
+    : null;
+  const fetchHead = checkout.ok && reset?.ok && clean?.ok
+    ? await runCaptured(["bash", "-lc", "p=\"$(git rev-parse --path-format=absolute --git-common-dir)/FETCH_HEAD\"; : > \"$p\"; chmod 600 \"$p\""], repoPath, dumpDir, "git-cache-fetch-head", { timeoutMs: Math.min(timeoutMs, 5000) })
+    : null;
+  const commands = [clone, remote, checkout, reset, clean, upstream, fetchHead].filter(Boolean).map(commandView);
+  return { ok: checkout.ok && reset?.ok === true && clean?.ok === true && fetchHead?.ok === true, stage: "ready", commands };
+}
+
+async function cloneEphemeralRepo(dumpDir, timeoutMs) {
+  const root = path.join(dumpDir, "ephemeral-repo");
+  const attempts = [];
+  await fsp.mkdir(root, { recursive: true, mode: 0o700 });
+  const egress = await egressProxyHealth(dumpDir, Math.min(timeoutMs, 5000));
+  if (egress.status !== "pass") {
+    return repoRecord(path.join(root, "HWLAB"), "ephemeral:git-cache", {
+      ephemeral: true,
+      remoteUrl: null,
+      egressProxy: egress,
+      dedicatedCache: cacheRepoRecord(dedicatedCacheRepo),
+      seedRepos: seedRepoCandidates().map(cacheRepoRecord),
+      cloneAttempts: [],
+      cloneFailed: true,
+      cloneFailureStage: "egress-proxy",
+      retainedForDiagnostics: true,
+    });
+  }
+  for (const remoteUrl of remoteCandidates()) {
+    const repoPath = path.join(root, "HWLAB");
+    const refreshed = await refreshDedicatedCache(remoteUrl, dumpDir, Math.min(timeoutMs, 45000));
+    attempts.push({ remoteUrl, dedicatedCache: refreshed.cache, ok: refreshed.ok, stage: refreshed.stage, refreshAttempts: refreshed.attempts });
+    if (!refreshed.ok) continue;
+    const materialized = await materializeFromCache(dedicatedCacheRepo, repoPath, remoteUrl, dumpDir, Math.min(timeoutMs, 9000));
+    attempts.push({ remoteUrl, dedicatedCache: refreshed.cache, ok: materialized.ok, stage: materialized.stage, commands: materialized.commands });
+    if (materialized.ok) {
+      return repoRecord(repoPath, "ephemeral:dedicated-git-cache", {
+        ephemeral: true,
+        remoteUrl,
+        egressProxy: egress,
+        dedicatedCache: refreshed.cache,
+        cacheRefreshStatus: refreshed.refreshStatus || "current",
+        remoteMainCommit: refreshed.remoteMainCommit || null,
+        cacheMainCommit: refreshed.localMainCommit || null,
+        deployCommitId: refreshed.deployCommitId || null,
+        seedRepos: seedRepoCandidates().map(cacheRepoRecord),
+        cloneAttempts: attempts,
+        retainedForDiagnostics: true,
+      });
+    }
+  }
+  return repoRecord(path.join(root, "HWLAB"), "ephemeral:dedicated-git-cache", {
+    ephemeral: true,
+    remoteUrl: null,
+    egressProxy: egress,
+    dedicatedCache: cacheRepoRecord(dedicatedCacheRepo),
+    seedRepos: seedRepoCandidates().map(cacheRepoRecord),
+    cloneAttempts: attempts,
+    cloneFailed: true,
+    cloneFailureStage: attempts.findLast((attempt) => attempt.stage)?.stage || "cache-unavailable",
+    retainedForDiagnostics: true,
+  });
+}
+
+async function resolveRepo(provided, dumpDir, timeoutMs) {
+  if (provided) return repoRecord(provided, "option", { ephemeral: false });
+  if (process.env.UNIDESK_HWLAB_REPO) return repoRecord(process.env.UNIDESK_HWLAB_REPO, "env:UNIDESK_HWLAB_REPO", { ephemeral: false });
+  return cloneEphemeralRepo(dumpDir, Math.min(timeoutMs, 55000));
+}
+
 async function gitSummary(repoPath, dumpDir, timeoutMs) {
  const [branch, head, originMain, remote, gitDir, gitCommonDir, statusShort, statusPorcelain] = await Promise.all([
    runCaptured(["git", "rev-parse", "--abbrev-ref", "HEAD"], repoPath, dumpDir, "git-branch", { timeoutMs }),
@@ -1113,6 +1373,14 @@ function collectBlockers({ repo, git, guard, lock, secretRefs, controlled, actio
  if (repo.status !== "selected") {
    blockers.push({ scope: "hwlab-repo", summary: repo.rejected ? "Rejected runner history HWLAB directory." : "No eligible HWLAB CD repo with scripts/dev-cd-apply.mjs and deploy/deploy.json was found." });
  }
+  if (repo.cacheRefreshStatus === "blocked") {
+    blockers.push({
+      scope: "hwlab-cache-refresh",
+      summary: "Dedicated HWLAB git cache could not refresh from GitHub through the provider egress proxy; using the last local cache state is not a release-truth PASS.",
+      remoteMainCommit: repo.remoteMainCommit || null,
+      cacheMainCommit: repo.cacheMainCommit || null,
+    });
+  }
  if (git?.blockers) blockers.push(...git.blockers);
  if (guard) {
    if (guard.refusal) blockers.push({ scope: "d601-native-k3s-guard", summary: guard.summary, refusal: true });
@@ -1139,6 +1407,341 @@ function nextSafeCommand(action, blockers) {
  return "host commander or the unique CD runner may decide whether to run node scripts/dev-cd-apply.mjs --apply --confirm-dev --confirmed-non-production --write-report on D601; this wrapper did not apply or rollout";
 }

+function compactBlockers(blockers, limit = 12) {
+  return (Array.isArray(blockers) ? blockers : []).slice(0, limit).map((item) => ({
+    scope: item?.scope ?? item?.blockerScope ?? null,
+    type: item?.type ?? null,
+    status: item?.status ?? null,
+    summary: item?.summary ?? item?.reason ?? null,
+  }));
+}
+
+function compactGit(git) {
+  if (!git) return null;
+  return {
+    status: git.status,
+    clean: git.clean,
+    branch: git.branch,
+    onMain: git.onMain,
+    remoteMatches: git.remoteMatches,
+    headMatchesOriginMain: git.headMatchesOriginMain,
+    dirtyCount: git.dirtyCount,
+    blockers: compactBlockers(git.blockers, 8),
+  };
+}
+
+function compactNodeGuard(guard) {
+  if (!guard) return null;
+  return {
+    status: guard.status,
+    refusal: guard.refusal,
+    kubeconfig: guard.kubeconfig,
+    currentContext: guard.currentContext,
+    apiServer: guard.apiServer,
+    nodeNames: guard.nodeNames,
+    requiredNodePresent: guard.requiredNodePresent,
+    refusalSignals: guard.refusalSignals,
+    secondControlPlaneRisk: guard.secondControlPlaneRisk,
+    defaultKubectlStatus: guard.defaultKubectlDiagnostic?.status ?? null,
+    summary: guard.summary,
+  };
+}
+
+function compactSecretPreflight(secretRefs) {
+  if (!secretRefs) return null;
+  return {
+    status: secretRefs.status,
+    valuesRead: secretRefs.safety?.secretValuesRead === true,
+    valuesPrinted: secretRefs.safety?.secretValuesPrinted === true,
+    secretRefs: (secretRefs.secretRefs || []).map((ref) => ({
+      secretName: ref.secretName,
+      secretKey: ref.secretKey,
+      exists: ref.exists === true,
+      keyPresent: ref.keyPresent === true,
+      status: ref.status,
+    })).slice(0, 8),
+    blockers: compactBlockers(secretRefs.blockers, 8),
+  };
+}
+
+function compactDesiredState(desired) {
+  if (!desired) return null;
+  return {
+    status: desired.status,
+    targetCommit: desired.targetCommit,
+    deployJson: desired.deployJson ? {
+      hash: desired.deployJson.hash,
+      commitId: desired.deployJson.commitId,
+      environment: desired.deployJson.environment,
+      namespace: desired.deployJson.namespace,
+      serviceCount: desired.deployJson.serviceCount,
+    } : null,
+    artifactCatalog: desired.artifactCatalog ? {
+      hash: desired.artifactCatalog.hash,
+      commitId: desired.artifactCatalog.commitId,
+      artifactState: desired.artifactCatalog.artifactState,
+      ciPublished: desired.artifactCatalog.ciPublished,
+      registryVerified: desired.artifactCatalog.registryVerified,
+      digestCounts: desired.artifactCatalog.digestCounts,
+    } : null,
+    imageConvergence: desired.imageConvergence ? {
+      status: desired.imageConvergence.status,
+      missing: (desired.imageConvergence.missing || []).slice(0, 5),
+      mismatches: (desired.imageConvergence.mismatches || []).slice(0, 5),
+    } : null,
+  };
+}
+
+function compactControlled(controlled) {
+  if (!controlled) return null;
+  return {
+    status: controlled.status,
+    commandOk: controlled.commandOk,
+    mode: controlled.mode,
+    mutationAttempted: controlled.mutationAttempted,
+    prodTouched: controlled.prodTouched,
+    target: controlled.target ? {
+      ref: controlled.target.ref,
+      promotionCommit: controlled.target.promotionCommit,
+      shortCommitId: controlled.target.shortCommitId,
+      promotionSource: controlled.target.promotionSource,
+      publishRequired: controlled.target.publishRequired,
+      headMatchesTarget: controlled.target.headMatchesTarget,
+      desiredStateStatus: controlled.target.desiredStateCheck?.status ?? null,
+      artifactBoundaryStatus: controlled.target.artifactBoundary?.status ?? null,
+      namespace: controlled.target.namespace,
+    } : null,
+    blockers: compactBlockers(controlled.blockers, 8),
+  };
+}
+
+function compactAudit(audit) {
+  if (!audit) return null;
+  return {
+    ok: audit.ok,
+    status: audit.status,
+    env: audit.env,
+    namespace: audit.namespace,
+    nodeGuard: audit.nodeGuard,
+    secrets: audit.secrets ? {
+      status: audit.secrets.status,
+      valuesRead: audit.secrets.valuesRead,
+      valuesPrinted: audit.secrets.valuesPrinted,
+    } : null,
+    registry: audit.registry ? {
+      status: audit.registry.status,
+      processHttpAccess: audit.registry.processHttpAccess,
+      k3sPullAccess: audit.registry.k3sPullAccess,
+    } : null,
+    lease: audit.lease,
+    desiredState: audit.desiredState ? {
+      status: audit.desiredState.status,
+      targetCommit: audit.desiredState.targetCommit,
+      imageConvergence: audit.desiredState.imageConvergence,
+    } : null,
+    workload: audit.workload ? {
+      status: audit.workload.status,
+      deploymentCount: audit.workload.deploymentCount,
+      currentImageConvergence: audit.workload.currentImageConvergence,
+      podWaiting: (audit.workload.podWaiting || []).slice(0, 5),
+      runtimeJobs: (audit.workload.runtimeJobs || []).slice(0, 5),
+    } : null,
+    publicHealth: audit.publicHealth ? {
+      status: audit.publicHealth.status,
+      expectedCommit: audit.publicHealth.expectedCommit,
+      summary: audit.publicHealth.summary,
+    } : null,
+    durability: audit.durability ? {
+      status: audit.durability.status,
+      unavailableReason: audit.durability.unavailableReason,
+    } : null,
+    controlledDevCd: compactControlled(audit.controlledDevCd),
+    blockerTypes: audit.blockerTypes || [],
+    blockers: compactBlockers(audit.blockers, 12),
+    safety: audit.safety,
+  };
+}
+
+function compactForTransport(payload, fullResultPath) {
+  return {
+    ok: payload.ok,
+    env: payload.env,
+    environment: payload.environment,
+    status: payload.status,
+    error: payload.error ?? null,
+    action: payload.action ?? null,
+    dryRun: payload.dryRun,
+    mutation: payload.mutation,
+    remoteHost: payload.remoteHost,
+    workspace: payload.workspace,
+    kubeconfig: payload.kubeconfig,
+    dumpPath: payload.dumpPath,
+    remoteFullResultPath: fullResultPath,
+    repo: payload.repo ? {
+      status: payload.repo.status,
+      path: payload.repo.path,
+      source: payload.repo.source,
+      rejected: payload.repo.rejected,
+      ephemeral: payload.repo.ephemeral === true,
+      cacheRefreshStatus: payload.repo.cacheRefreshStatus || null,
+    } : null,
+    worktreeGuard: compactGit(payload.worktreeGuard),
+    nodeGuard: compactNodeGuard(payload.nodeGuard),
+    secretPreflight: compactSecretPreflight(payload.secretPreflight),
+    lockState: payload.lockState ? {
+      status: payload.lockState.status,
+      phase: payload.lockState.phase,
+      holder: payload.lockState.ownerTaskId || payload.lockState.holderIdentity || null,
+      held: payload.lockState.held,
+      stale: payload.lockState.stale,
+      retryAfterSeconds: payload.lockState.retryAfterSeconds,
+      lockName: payload.lockState.lockName,
+    } : null,
+    desiredState: payload.desiredState ? {
+      deployJson: payload.desiredState.deployJson,
+      artifactCatalog: payload.desiredState.artifactCatalog,
+      artifactReport: payload.desiredState.artifactReport,
+    } : null,
+    target: payload.target,
+    promotion: payload.promotion ? {
+      source: payload.promotion.source,
+      deployJson: payload.promotion.deployJson,
+      artifactCatalog: payload.promotion.artifactCatalog,
+      artifactReport: payload.promotion.artifactReport,
+    } : null,
+    controlledDevCd: compactControlled(payload.controlledDevCd),
+    audit: compactAudit(payload.audit),
+    blockers: compactBlockers(payload.blockers, 12),
+    blockerTypes: payload.blockerTypes || [...new Set(compactBlockers(payload.blockers, 12).map((item) => item.scope).filter(Boolean))],
+    nextSafeCommand: payload.nextSafeCommand,
+    safety: payload.safety,
+  };
+}
+
+function emitResult(payload, dumpDir, compactStdout = false) {
+  const fullResultPath = path.join(dumpDir, "result.full.json");
+  const compactResultPath = path.join(dumpDir, "result.compact.json");
+  fs.writeFileSync(fullResultPath, `${JSON.stringify(payload, null, 2)}\n`, { mode: 0o600 });
+  if (!compactStdout) {
+    fs.writeFileSync(compactResultPath, `${JSON.stringify(compactForTransport(payload, fullResultPath), null, 2)}\n`, { mode: 0o600 });
+    console.log(JSON.stringify(payload, null, 2));
+    return;
+  }
+  let compact = compactForTransport(payload, fullResultPath);
+  let text = JSON.stringify(compact, null, 2);
+  if (Buffer.byteLength(text, "utf8") > 3600) {
+    compact = {
+      ok: compact.ok,
+      env: compact.env,
+      status: compact.status,
+      error: compact.error,
+      action: compact.action,
+      dryRun: compact.dryRun,
+      mutation: compact.mutation,
+      workspace: compact.workspace,
+      dumpPath: compact.dumpPath,
+      remoteFullResultPath: compact.remoteFullResultPath,
+      repo: compact.repo,
+      worktreeGuard: compact.worktreeGuard,
+      nodeGuard: compact.nodeGuard,
+      lockState: compact.lockState,
+      target: compact.target ? {
+        ref: compact.target.ref,
+        promotionCommit: compact.target.promotionCommit,
+        shortCommitId: compact.target.shortCommitId,
+        promotionSource: compact.target.promotionSource,
+        publishRequired: compact.target.publishRequired,
+      } : null,
+      controlledDevCd: compact.controlledDevCd,
+      audit: compact.audit ? {
+        status: compact.audit.status,
+        blockerTypes: compact.audit.blockerTypes,
+        blockers: compact.audit.blockers,
+        publicHealth: compact.audit.publicHealth,
+        workload: compact.audit.workload,
+        durability: compact.audit.durability,
+      } : null,
+      blockers: compact.blockers,
+      blockerTypes: compact.blockerTypes,
+      nextSafeCommand: compact.nextSafeCommand,
+      safety: compact.safety,
+      outputCompacted: "provider-host-ssh-stdout-limit",
+    };
+    text = JSON.stringify(compact, null, 2);
+  }
+  if (Buffer.byteLength(text, "utf8") > 2400) {
+    compact = {
+      ok: payload.ok,
+      env: payload.env,
+      status: payload.status,
+      error: payload.error ?? null,
+      action: payload.action ?? null,
+      dryRun: payload.dryRun,
+      mutation: payload.mutation,
+      workspace: payload.workspace,
+      dumpPath: payload.dumpPath,
+      remoteFullResultPath: fullResultPath,
+      repo: payload.repo ? {
+        status: payload.repo.status,
+        path: payload.repo.path,
+        source: payload.repo.source,
+        rejected: payload.repo.rejected,
+        ephemeral: payload.repo.ephemeral === true,
+        cacheRefreshStatus: payload.repo.cacheRefreshStatus || null,
+      } : null,
+      worktreeStatus: payload.worktreeGuard?.status ?? null,
+      nodeGuardStatus: payload.nodeGuard?.status ?? payload.audit?.nodeGuard?.status ?? null,
+      lockState: payload.lockState ? {
+        status: payload.lockState.status,
+        phase: payload.lockState.phase,
+        held: payload.lockState.held,
+        holder: payload.lockState.ownerTaskId || payload.lockState.holderIdentity || null,
+        stale: payload.lockState.stale,
+      } : null,
+      target: payload.target ? {
+        ref: payload.target.ref,
+        promotionCommit: payload.target.promotionCommit,
+        shortCommitId: payload.target.shortCommitId,
+        promotionSource: payload.target.promotionSource,
+      } : payload.audit?.desiredState ? {
+        targetCommit: payload.audit.desiredState.targetCommit,
+      } : null,
+      controlledDevCdStatus: payload.controlledDevCd?.status ?? payload.audit?.controlledDevCd?.status ?? null,
+      auditStatus: payload.audit?.status ?? null,
+      blockerTypes: payload.blockerTypes || payload.audit?.blockerTypes || [...new Set(compactBlockers(payload.blockers, 8).map((item) => item.scope).filter(Boolean))],
+      blockers: compactBlockers(payload.blockers || payload.audit?.blockers, 6),
+      nextSafeCommand: payload.nextSafeCommand,
+      outputCompacted: "provider-host-ssh-stdout-limit",
+    };
+    text = JSON.stringify(compact);
+  }
+  if (Buffer.byteLength(text, "utf8") > 520) {
+    const blockerTypes = payload.blockerTypes
+      || payload.audit?.blockerTypes
+      || [...new Set(compactBlockers(payload.blockers || payload.audit?.blockers, 8).map((item) => item.scope).filter(Boolean))];
+    compact = {
+      ok: payload.ok,
+      env: payload.env,
+      status: payload.status,
+      error: payload.error ?? null,
+      action: payload.action ?? null,
+      remoteFullResultPath: fullResultPath,
+      repoPath: payload.repo?.path ?? payload.workspace?.path ?? null,
+      repoEphemeral: payload.repo?.ephemeral === true,
+      cacheRefreshStatus: payload.repo?.cacheRefreshStatus ?? null,
+      worktreeStatus: payload.worktreeGuard?.status ?? null,
+      nodeGuardStatus: payload.nodeGuard?.status ?? payload.audit?.nodeGuard?.status ?? null,
+      lockStatus: payload.lockState ? `${payload.lockState.status || "unknown"}/${payload.lockState.phase || "unknown"}/${payload.lockState.held === true ? "held" : "free"}` : null,
+      targetCommit: payload.target?.promotionCommit ?? payload.audit?.desiredState?.targetCommit ?? null,
+      blockerTypes: Array.isArray(blockerTypes) ? blockerTypes.slice(0, 4) : [],
+      outputCompacted: true,
+    };
+    text = JSON.stringify(compact);
+  }
+  fs.writeFileSync(compactResultPath, `${text}\n`, { mode: 0o600 });
+  console.log(text);
+}
+
 async function main() {
  const options = decodeOptions();
  const timeoutMs = Math.min(Number(options.timeoutMs || 45000), 60000);
@@ -1146,8 +1749,8 @@ async function main() {
  const dumpDir = path.join(os.homedir(), ".state", "unidesk-hwlab-cd", runId);
  await fsp.mkdir(dumpDir, { recursive: true, mode: 0o700 });

-  const repo = resolveRepo(options.repoPath || null);
  const action = options.action;
+  const repo = await resolveRepo(options.repoPath || null, dumpDir, Math.min(timeoutMs, 55000));
  const base = {
    ok: false,
    env: "dev",
@@ -1176,7 +1779,7 @@ async function main() {

  if (repo.status !== "selected") {
    const blockers = collectBlockers({ repo, action });
-    console.log(JSON.stringify({ ...base, status: "blocked", error: "hwlab-repo-not-found", repo, blockers, nextSafeCommand: nextSafeCommand(action, blockers) }, null, 2));
+    emitResult({ ...base, status: "blocked", error: "hwlab-repo-not-found", repo, blockers, nextSafeCommand: nextSafeCommand(action, blockers) }, dumpDir, options.compactStdout === true);
    process.exitCode = 1;
    return;
  }
@@ -1218,7 +1821,7 @@ async function main() {
    ]);
    const durability = runtimeDurabilityAudit(auditDesiredState, publicHealth, secretRefs, workload);
    const summary = auditSummary({ repo, git, guard, secretRefs, registry, lock, desired: auditDesiredState, workload, publicHealth, durability, controlled, dumpDir });
-    console.log(JSON.stringify({
+    emitResult({
      ...base,
      ok: summary.ok,
      status: summary.status,
@@ -1235,13 +1838,13 @@ async function main() {
      blockers: summary.blockers,
      blockerTypes: summary.blockerTypes,
      nextSafeCommand: nextSafeCommand(action, summary.blockers),
-    }, null, 2));
+    }, dumpDir, options.compactStdout === true);
    process.exitCode = summary.ok ? 0 : 1;
    return;
  }
  const blockers = collectBlockers({ repo, git, guard, lock, secretRefs, controlled, action });
  const ok = blockers.length === 0;
-  console.log(JSON.stringify({
+  emitResult({
    ...base,
    ok,
    status: ok ? action === "status" ? "ready" : "prepared" : guard.refusal ? "refused" : "blocked",
@@ -1263,7 +1866,7 @@ async function main() {
    reportDumpPath: controlled.command?.dump?.stdout || dumpDir,
    blockers,
    nextSafeCommand: nextSafeCommand(action, blockers),
-  }, null, 2));
+  }, dumpDir, options.compactStdout === true);
  process.exitCode = ok ? 0 : 1;
 }

@@ -3,6 +3,7 @@ import { randomBytes } from "node:crypto";
 import { createWriteStream, mkdirSync, readFileSync } from "node:fs";
 import { mkdir } from "node:fs/promises";
 import { join } from "node:path";
+import { gzipSync } from "node:zlib";
 import { readConfig, repoRoot, rootPath, type UniDeskConfig } from "./config";
 import { d601NativeKubeconfig } from "./d601-k3s-guard";

@@ -46,6 +47,24 @@ interface FrontendSession {
  cookie: string;
 }

+interface FrontendHostSshResult {
+  ok: boolean;
+  taskId: string | null;
+  taskStatus: unknown;
+  stdout: string;
+  stderr: string;
+  exitCode: number | null;
+  error: string;
+  raw: unknown;
+}
+
+interface RemoteRunnerJob {
+  pid: number | null;
+  remoteRunDir: string | null;
+  remoteCompactResultPath: string | null;
+  remoteFullResultPath: string | null;
+}
+
 interface FetchJsonResult {
  ok: boolean;
  status?: number;
@@ -54,7 +73,8 @@ interface FetchJsonResult {
 }

 const defaultProviderId = "D601";
-const defaultHwlabCdRepoPath = "/home/ubuntu/hwlab_cd";
+const defaultHwlabCdRepoPath = "~/.state/unidesk-hwlab-cd/<run-id>/ephemeral-repo/HWLAB";
+const defaultHwlabCdCachePath = "~/.cache/unidesk/hwlab-cd/git-cache/HWLAB.git";
 const rejectedRunnerHistoryRepoPath = "/home/ubuntu/hwlab";
 const namespace = "hwlab-dev";
 const lockName = "hwlab-dev-cd-lock";
@@ -160,8 +180,8 @@ function b64Json(value: unknown): string {
  return Buffer.from(JSON.stringify(value), "utf8").toString("base64");
 }

-function buildRemoteCommand(options: HwlabCdOptions, runId: string): string {
-  const remoteOptions = {
+function buildRemoteOptions(options: HwlabCdOptions, runId: string): Record<string, unknown> {
+  return {
    action: options.action,
    environment: options.environment,
    dryRun: options.dryRun,
@@ -169,14 +189,113 @@ function buildRemoteCommand(options: HwlabCdOptions, runId: string): string {
    kubeconfig: options.kubeconfig,
    timeoutMs: options.timeoutMs,
    runId,
+    compactStdout: options.transport !== "local",
  };
+}
+
+function buildRemoteInlineCommand(options: HwlabCdOptions, runId: string): string {
  return [
-    `UNIDESK_HWLAB_CD_OPTIONS_B64=${shellQuote(b64Json(remoteOptions))} node <<'UNIDESK_HWLAB_CD_JS'`,
+    `UNIDESK_HWLAB_CD_OPTIONS_B64=${shellQuote(b64Json(buildRemoteOptions(options, runId)))} node <<'UNIDESK_HWLAB_CD_JS'`,
    remoteScriptSource,
    "UNIDESK_HWLAB_CD_JS",
  ].join("\n");
 }

+function buildRemoteUploadedCommand(options: HwlabCdOptions, runId: string, remoteScriptPath: string): string {
+  return `UNIDESK_HWLAB_CD_OPTIONS_B64=${shellQuote(b64Json(buildRemoteOptions(options, runId)))} node ${shellQuote(remoteScriptPath)}`;
+}
+
+function remoteRunnerDir(runId: string): string {
+  return `$HOME/.state/unidesk-hwlab-cd/${safeRemoteId(runId)}`;
+}
+
+function buildRemoteStartCommand(options: HwlabCdOptions, runId: string, remoteScriptPath: string): string {
+  const runDir = remoteRunnerDir(runId);
+  const encodedOptions = shellQuote(b64Json(buildRemoteOptions(options, runId)));
+  return [
+    "umask 077",
+    `run_dir="${runDir}"`,
+    `mkdir -p "$run_dir"`,
+    `rm -f "$run_dir/result.full.json" "$run_dir/result.compact.json" "$run_dir/runner.stdout.txt" "$run_dir/runner.stderr.txt" "$run_dir/runner.exit"`,
+    `nohup sh -c 'UNIDESK_HWLAB_CD_OPTIONS_B64="$1" node "$2" > "$3" 2> "$4"; printf "%s\\n" "$?" > "$5"' sh ${encodedOptions} ${shellQuote(remoteScriptPath)} "$run_dir/runner.stdout.txt" "$run_dir/runner.stderr.txt" "$run_dir/runner.exit" >/dev/null 2>&1 & pid=$!`,
+    `printf '{"pid":%s,"remoteRunDir":"%s","remoteCompactResultPath":"%s/result.compact.json","remoteFullResultPath":"%s/result.full.json"}\\n' "$pid" "$run_dir" "$run_dir" "$run_dir"`,
+  ].join("; ");
+}
+
+function buildRemotePollCommand(runId: string): string {
+  const runDir = remoteRunnerDir(runId);
+  const pollJs = [
+    "const fs=require(\"fs\")",
+    "const dir=process.argv[1]",
+    "const compact=`${dir}/result.compact.json`",
+    "const exitPath=`${dir}/runner.exit`",
+    "const tail=(p)=>{try{const b=fs.readFileSync(p);return b.slice(Math.max(0,b.length-500)).toString(\"utf8\")}catch{return \"\"}}",
+    "if(fs.existsSync(compact)&&fs.statSync(compact).size>0){process.stdout.write(fs.readFileSync(compact,\"utf8\"));process.exit(0)}",
+    "if(fs.existsSync(exitPath)){const raw=fs.readFileSync(exitPath,\"utf8\").trim();console.log(JSON.stringify({ok:false,status:\"blocked\",error:\"remote-runner-exited-without-result\",remoteRunDir:dir,exitCode:Number(raw),stdoutTail:tail(`${dir}/runner.stdout.txt`),stderrTail:tail(`${dir}/runner.stderr.txt`)}));process.exit(0)}",
+    "console.log(JSON.stringify({ok:false,status:\"running\",remoteRunDir:dir}))",
+  ].join(";");
+  return `run_dir="${runDir}"; node -e ${shellQuote(pollJs)} "$run_dir"`;
+}
+
+function buildRemoteCommand(options: HwlabCdOptions, runId: string): string {
+  return buildRemoteInlineCommand(options, runId);
+}
+
+function splitFixed(value: string, size: number): string[] {
+  const chunks: string[] = [];
+  for (let index = 0; index < value.length; index += size) chunks.push(value.slice(index, index + size));
+  return chunks;
+}
+
+function safeRemoteId(value: string): string {
+  return value.replace(/[^A-Za-z0-9_.-]/gu, "-");
+}
+
+function compactTail(value: string, maxLength = 1000): string {
+  return value.slice(Math.max(0, value.length - maxLength));
+}
+
+function uploadCommandShape(providerId = defaultProviderId): string[] {
+  return ["frontend", "/api/dispatch", providerId, "host.ssh", "upload remote runner in small chunks", "start nohup remote runner", "poll ~/.state/unidesk-hwlab-cd/<run-id>/result.compact.json"];
+}
+
+function buildUploadPlan(runId: string): { remoteDir: string; remotePath: string; commands: Array<{ id: string; command: string }> } {
+  const remoteDir = "/tmp/unidesk-hwlab-cd";
+  const remotePath = `${remoteDir}/remote-runner-${safeRemoteId(runId)}.cjs`;
+  const encoded = gzipSync(Buffer.from(remoteScriptSource, "utf8")).toString("base64");
+  const b64Path = `${remotePath}.gz.b64`;
+  const commands = [
+    {
+      id: "init",
+      command: `umask 077; mkdir -p ${shellQuote(remoteDir)}; rm -f ${shellQuote(remotePath)} ${shellQuote(b64Path)}; : > ${shellQuote(b64Path)}`,
+    },
+    ...splitFixed(encoded, 2400).map((chunk, index) => ({
+      id: `chunk-${index + 1}`,
+      command: `printf %s ${shellQuote(chunk)} >> ${shellQuote(b64Path)}`,
+    })),
+    {
+      id: "finalize",
+      command: `base64 -d ${shellQuote(b64Path)} | gzip -d > ${shellQuote(remotePath)}; chmod 700 ${shellQuote(remotePath)}; rm -f ${shellQuote(b64Path)}; test -s ${shellQuote(remotePath)}; wc -c ${shellQuote(remotePath)}`,
+    },
+  ];
+  return { remoteDir, remotePath, commands };
+}
+
+function uploadStepView(step: { id: string; command: string }, result: FrontendHostSshResult): Record<string, unknown> {
+  const remoteOptions = {
+    id: step.id,
+    ok: result.ok,
+    taskId: result.taskId,
+    taskStatus: result.taskStatus,
+    exitCode: result.exitCode,
+    commandBytes: step.command.length,
+    stdoutTail: compactTail(result.stdout, 500),
+    stderrTail: compactTail(result.stderr, 500),
+    error: result.error,
+  };
+  return remoteOptions;
+}
+
 export function buildHwlabCdRemoteCommandForTest(args: string[]): string {
  return buildRemoteCommand(parseOptions(args), "test-run-id");
 }
@@ -325,7 +444,7 @@ function parseRemoteStdout(stdout: string): Record<string, unknown> | null {
 }

 function commandShape(providerId = defaultProviderId): string[] {
-  return ["frontend", "/api/dispatch", providerId, "host.ssh", "UNIDESK_HWLAB_CD_OPTIONS_B64=... node <<'UNIDESK_HWLAB_CD_JS'"];
+  return uploadCommandShape(providerId);
 }

 async function runLocalTransport(options: HwlabCdOptions, remoteCommand: string, runId: string): Promise<Record<string, unknown>> {
@@ -370,65 +489,219 @@ async function runLocalTransport(options: HwlabCdOptions, remoteCommand: string,
  };
 }

-async function runFrontendTransport(options: HwlabCdOptions, remoteCommand: string, config: UniDeskConfig, runId: string): Promise<Record<string, unknown>> {
-  const host = options.mainServerHost ?? config.network.publicHost;
-  const dumpDir = rootPath(".state", "hwlab-cd", runId);
-  mkdirSync(dumpDir, { recursive: true, mode: 0o700 });
-  const session = await loginFrontend(host, config);
+async function dispatchFrontendHostSsh(
+  session: FrontendSession,
+  options: HwlabCdOptions,
+  command: string,
+  timeoutMs: number,
+  source = "cli-hwlab-cd",
+): Promise<FrontendHostSshResult> {
  const dispatch = await frontendJson(session, "/api/dispatch", {
    method: "POST",
    body: JSON.stringify({
      providerId: options.providerId,
      command: "host.ssh",
-      payload: { source: "cli-hwlab-cd", mode: "exec", command: remoteCommand, timeoutMs: options.timeoutMs },
+      payload: { source, mode: "exec", command, timeoutMs },
    }),
  }, 12_000);
  const taskId = String(asRecord(dispatch.body)?.taskId ?? "");
  if (!dispatch.ok || taskId.length === 0) {
+    return {
+      ok: false,
+      taskId: taskId || null,
+      taskStatus: null,
+      stdout: "",
+      stderr: "",
+      exitCode: null,
+      error: asRecord(dispatch.body)?.error as string ?? dispatch.error ?? "provider dispatch failed",
+      raw: dispatch,
+    };
+  }
+  const wait = await waitForFrontendTask(session, taskId, Math.min(timeoutMs + 5000, maxTimeoutMs));
+  const task = asRecord((wait as { task?: unknown }).task);
+  const result = asRecord(task?.result) ?? {};
+  const exitCode = typeof result.exitCode === "number" ? result.exitCode : null;
+  const stdout = typeof result.stdout === "string" ? result.stdout : "";
+  const stderr = typeof result.stderr === "string" ? result.stderr : "";
+  const error = typeof result.error === "string" ? result.error : "";
+  return {
+    ok: task?.status === "succeeded" && (exitCode === null || exitCode === 0),
+    taskId,
+    taskStatus: task?.status ?? null,
+    stdout,
+    stderr,
+    exitCode,
+    error,
+    raw: task,
+  };
+}
+
+async function uploadRemoteRunner(session: FrontendSession, options: HwlabCdOptions, runId: string, dumpDir: string): Promise<{ ok: boolean; path: string; steps: Record<string, unknown>[]; error: string }> {
+  const plan = buildUploadPlan(runId);
+  const steps: Record<string, unknown>[] = [];
+  for (const step of plan.commands) {
+    const result = await dispatchFrontendHostSsh(session, options, step.command, Math.min(options.timeoutMs, 20_000), "cli-hwlab-cd-upload");
+    steps.push(uploadStepView(step, result));
+    if (!result.ok) {
+      await Bun.write(join(dumpDir, "frontend-upload.json"), `${JSON.stringify({ ok: false, path: plan.remotePath, steps }, null, 2)}\n`);
+      return { ok: false, path: plan.remotePath, steps, error: result.error || result.stderr || `upload step ${step.id} failed` };
+    }
+  }
+  await Bun.write(join(dumpDir, "frontend-upload.json"), `${JSON.stringify({ ok: true, path: plan.remotePath, steps }, null, 2)}\n`);
+  return { ok: true, path: plan.remotePath, steps, error: "" };
+}
+
+function parseRemoteRunnerJob(stdout: string): RemoteRunnerJob | null {
+  const parsed = parseRemoteStdout(stdout);
+  if (parsed === null) return null;
+  return {
+    pid: typeof parsed.pid === "number" ? parsed.pid : null,
+    remoteRunDir: typeof parsed.remoteRunDir === "string" ? parsed.remoteRunDir : null,
+    remoteCompactResultPath: typeof parsed.remoteCompactResultPath === "string" ? parsed.remoteCompactResultPath : null,
+    remoteFullResultPath: typeof parsed.remoteFullResultPath === "string" ? parsed.remoteFullResultPath : null,
+  };
+}
+
+async function startRemoteRunner(
+  session: FrontendSession,
+  options: HwlabCdOptions,
+  runId: string,
+  remoteScriptPath: string,
+  dumpDir: string,
+): Promise<{ ok: boolean; job: RemoteRunnerJob | null; result: FrontendHostSshResult; dump: string }> {
+  const result = await dispatchFrontendHostSsh(session, options, buildRemoteStartCommand(options, runId, remoteScriptPath), 8000, "cli-hwlab-cd-start");
+  const stdoutDump = join(dumpDir, "frontend-runner-start.stdout.txt");
+  const stderrDump = join(dumpDir, "frontend-runner-start.stderr.txt");
+  await Bun.write(stdoutDump, result.stdout);
+  await Bun.write(stderrDump, result.stderr);
+  const job = result.ok ? parseRemoteRunnerJob(result.stdout) : null;
+  const dump = join(dumpDir, "frontend-runner-start.json");
+  await Bun.write(dump, `${JSON.stringify({ ok: result.ok && job !== null, job, taskId: result.taskId, taskStatus: result.taskStatus, exitCode: result.exitCode, stdoutDump, stderrDump, error: result.error }, null, 2)}\n`);
+  return { ok: result.ok && job !== null, job, result, dump };
+}
+
+async function pollRemoteRunnerResult(
+  session: FrontendSession,
+  options: HwlabCdOptions,
+  runId: string,
+  dumpDir: string,
+): Promise<{ parsed: Record<string, unknown> | null; stdout: string; stderr: string; result: FrontendHostSshResult | null; polls: Record<string, unknown>[]; timedOut: boolean; dump: string }> {
+  const startedAt = Date.now();
+  const polls: Record<string, unknown>[] = [];
+  const pollCommand = buildRemotePollCommand(runId);
+  let lastResult: FrontendHostSshResult | null = null;
+  while (Date.now() - startedAt < options.timeoutMs) {
+    const result = await dispatchFrontendHostSsh(session, options, pollCommand, 6000, "cli-hwlab-cd-poll");
+    lastResult = result;
+    const parsed = result.ok ? parseRemoteStdout(result.stdout) : null;
+    polls.push({
+      ok: result.ok,
+      taskId: result.taskId,
+      taskStatus: result.taskStatus,
+      exitCode: result.exitCode,
+      parsedStatus: parsed?.status ?? null,
+      parsedError: parsed?.error ?? null,
+      stdoutTail: compactTail(result.stdout, 500),
+      stderrTail: compactTail(result.stderr, 500),
+      error: result.error,
+    });
+    if (parsed !== null && parsed.status !== "running") {
+      const dump = join(dumpDir, "frontend-runner-poll.json");
+      await Bun.write(dump, `${JSON.stringify({ ok: true, polls }, null, 2)}\n`);
+      return { parsed, stdout: result.stdout, stderr: result.stderr, result, polls, timedOut: false, dump };
+    }
+    await Bun.sleep(1000);
+  }
+  const dump = join(dumpDir, "frontend-runner-poll.json");
+  await Bun.write(dump, `${JSON.stringify({ ok: false, timedOut: true, polls }, null, 2)}\n`);
+  return { parsed: null, stdout: lastResult?.stdout ?? "", stderr: lastResult?.stderr ?? "", result: lastResult, polls, timedOut: true, dump };
+}
+
+async function runFrontendTransport(options: HwlabCdOptions, config: UniDeskConfig, runId: string): Promise<Record<string, unknown>> {
+  const host = options.mainServerHost ?? config.network.publicHost;
+  const dumpDir = rootPath(".state", "hwlab-cd", runId);
+  mkdirSync(dumpDir, { recursive: true, mode: 0o700 });
+  const session = await loginFrontend(host, config);
+  const upload = await uploadRemoteRunner(session, options, runId, dumpDir);
+  if (!upload.ok) {
    return {
      ok: false,
      env: options.environment,
      status: "blocked",
-      error: "provider-dispatch-failed",
+      error: "remote-runner-upload-failed",
      remote: {
        host: session.baseUrl,
        providerId: options.providerId,
        transport: "frontend",
        commandShape: commandShape(options.providerId),
        commandCalls: ["scripts/dev-cd-apply.mjs"],
-        dispatch,
+        uploadPath: upload.path,
+        uploadDump: join(dumpDir, "frontend-upload.json"),
      },
      dumpPath: dumpDir,
-      nextSafeCommand: "restore UniDesk frontend/backend-core provider dispatch, then rerun bun scripts/cli.ts hwlab cd status --env dev",
+      upload,
+      nextSafeCommand: "restore D601 host.ssh command dispatch or upload permissions, then rerun bun scripts/cli.ts hwlab cd status --env dev",
    };
  }
-  const wait = await waitForFrontendTask(session, taskId, Math.min(options.timeoutMs + 5000, maxTimeoutMs));
-  const task = asRecord((wait as { task?: unknown }).task);
-  const result = asRecord(task?.result) ?? {};
-  const stdout = typeof result.stdout === "string" ? result.stdout : "";
-  const stderr = typeof result.stderr === "string" ? result.stderr : "";
+
+  const runner = await startRemoteRunner(session, options, runId, upload.path, dumpDir);
+  if (!runner.ok) {
+    return {
+      ok: false,
+      env: options.environment,
+      status: "blocked",
+      error: "remote-runner-start-failed",
+      remote: {
+        host: session.baseUrl,
+        providerId: options.providerId,
+        transport: "frontend",
+        taskId: runner.result.taskId,
+        taskStatus: runner.result.taskStatus,
+        commandShape: commandShape(options.providerId),
+        commandCalls: ["scripts/dev-cd-apply.mjs"],
+        uploadPath: upload.path,
+        uploadDump: join(dumpDir, "frontend-upload.json"),
+        runnerStartDump: runner.dump,
+        exitCode: runner.result.exitCode,
+        error: runner.result.error,
+      },
+      dumpPath: dumpDir,
+      nextSafeCommand: "restore D601 host.ssh background job dispatch, then rerun bun scripts/cli.ts hwlab cd status --env dev",
+    };
+  }
+
+  const poll = await pollRemoteRunnerResult(session, options, runId, dumpDir);
+  const stdout = poll.stdout;
+  const stderr = poll.stderr;
  const stdoutDump = join(dumpDir, "frontend-task.stdout.txt");
  const stderrDump = join(dumpDir, "frontend-task.stderr.txt");
-  Bun.write(stdoutDump, stdout);
-  Bun.write(stderrDump, stderr);
-  const parsed = parseRemoteStdout(stdout);
+  await Bun.write(stdoutDump, stdout);
+  await Bun.write(stderrDump, stderr);
+  const parsed = poll.parsed ?? parseRemoteStdout(stdout);
  if (parsed === null) {
    return {
      ok: false,
      env: options.environment,
      status: "blocked",
-      error: "remote-empty-or-invalid-json",
+      error: poll.timedOut ? "remote-runner-timeout" : "remote-empty-or-invalid-json",
      remote: {
        host: session.baseUrl,
        providerId: options.providerId,
        transport: "frontend",
-        taskId,
-        taskStatus: task?.status ?? null,
+        taskId: poll.result?.taskId ?? null,
+        taskStatus: poll.result?.taskStatus ?? null,
        commandShape: commandShape(options.providerId),
        commandCalls: ["scripts/dev-cd-apply.mjs"],
+        uploadPath: upload.path,
+        uploadDump: join(dumpDir, "frontend-upload.json"),
+        runnerStartDump: runner.dump,
+        runnerPollDump: poll.dump,
+        remoteRunDir: runner.job?.remoteRunDir ?? null,
+        remoteFullResultPath: runner.job?.remoteFullResultPath ?? null,
        stdoutDump,
        stderrDump,
-        exitCode: typeof result.exitCode === "number" ? result.exitCode : null,
+        exitCode: poll.result?.exitCode ?? null,
+        error: poll.result?.error ?? "",
      },
      dumpPath: dumpDir,
    };
@@ -439,13 +712,19 @@ async function runFrontendTransport(options: HwlabCdOptions, remoteCommand: stri
      host: session.baseUrl,
      providerId: options.providerId,
      transport: "frontend",
-      taskId,
-      taskStatus: task?.status ?? null,
+      taskId: poll.result?.taskId ?? null,
+      taskStatus: poll.result?.taskStatus ?? null,
      commandShape: commandShape(options.providerId),
      commandCalls: ["scripts/dev-cd-apply.mjs"],
+      uploadPath: upload.path,
+      uploadDump: join(dumpDir, "frontend-upload.json"),
+      runnerStartDump: runner.dump,
+      runnerPollDump: poll.dump,
+      remoteRunDir: runner.job?.remoteRunDir ?? null,
+      remoteFullResultPath: runner.job?.remoteFullResultPath ?? null,
      stdoutDump,
      stderrDump,
-      exitCode: typeof result.exitCode === "number" ? result.exitCode : null,
+      exitCode: poll.result?.exitCode ?? null,
    },
    localDumpPath: dumpDir,
  };
@@ -502,14 +781,17 @@ export function hwlabHelp(): Record<string, unknown> {
      "bun scripts/cli.ts hwlab cd preflight --env dev",
      "bun scripts/cli.ts hwlab cd apply --env dev --dry-run",
    ],
-    description: "Bounded UniDesk wrapper for the HWLAB DEV CD path. The default transport dispatches a D601 provider host.ssh command and the D601 side calls HWLAB repo-owned scripts/dev-cd-apply.mjs.",
+    description: "Bounded UniDesk wrapper for the HWLAB DEV CD path. The default transport uploads a small remote runner to D601 through short host.ssh commands, creates a one-run HWLAB clone owned by the host.ssh user, then calls HWLAB repo-owned scripts/dev-cd-apply.mjs from the D601 side.",
    boundary: [
      `KUBECONFIG is forced to ${d601NativeKubeconfig}`,
      "docker-desktop, desktop-control-plane, and 127.0.0.1:11700 are refusal signals for the explicit D601 target",
-      `default HWLAB CD repo is ${defaultHwlabCdRepoPath}; ${rejectedRunnerHistoryRepoPath} is rejected as runner history`,
+      `default HWLAB CD repo is ${defaultHwlabCdRepoPath}, materialized from dedicated full bare cache ${defaultHwlabCdCachePath}; ${rejectedRunnerHistoryRepoPath} is rejected as runner history`,
      "deploy/deploy.json remains the authoritative desired-state source",
+      "the default HWLAB repo is an ephemeral one-run clone under the remote dump directory; --hwlab-repo is reserved for explicitly supplied clean diagnostic clones",
+      "the dedicated cache is owner-only and HWLAB-CD-only; it must keep full repo history so deploy/deploy.json promotion commits resolve, and one-run repos are copied with independent object stores rather than --shared alternates",
      "preflight/apply --dry-run check required SecretRef object/key metadata without reading or printing Secret values",
      "audit/status/preflight/apply --dry-run call only HWLAB scripts/dev-cd-apply.mjs read-only/status paths with --skip-live-verify; no apply, rollout, lock mutation, DEV acceptance live verification, DB write, or secret read is executed",
+      "frontend transport keeps each host.ssh command below the provider-gateway short-command limit by uploading the remote runner in chunks, starting it as a remote background job, and polling compact result files",
      "audit adds bounded read-only kubectl get/curl health probes for blocker classification; full stdout/stderr stays in temp dump paths, not reports/",
      "real apply is structured refused and must remain with the host commander or unique CD runner",
    ],
@@ -524,5 +806,5 @@ export async function runHwlabCdCommand(args: string[]): Promise<Record<string,
  const remoteCommand = buildRemoteCommand(options, runId);
  if (options.transport === "local") return runLocalTransport(options, remoteCommand, runId);
  const config = readConfig();
-  return runFrontendTransport(options, remoteCommand, config, runId);
+  return runFrontendTransport(options, config, runId);
 }