superlab 0.1.47 → 0.1.48

package/lib/auto_contracts.cjs

@@ -99,6 +99,23 @@ function normalizeRequestedAutoContract(requested = {}) {
   };
 }
 
+function resolveRequestedAutonomyLevel({
+  requestedAutonomyLevel = "",
+  currentAutonomyLevel = "",
+} = {}) {
+  const requested = normalizeScalar(requestedAutonomyLevel || "");
+  if (isMeaningful(requested)) {
+    return requested;
+  }
+
+  const current = normalizeScalar(currentAutonomyLevel || "");
+  if (isMeaningful(current)) {
+    return current;
+  }
+
+  return "l2";
+}
+
 function sameStageSet(left, right) {
   if (left.length !== right.length) {
     return false;
@@ -530,6 +547,7 @@ module.exports = {
   inferCampaignKind,
   isLocalProcessAlive,
   normalizeRequestedAutoContract,
+  resolveRequestedAutonomyLevel,
   resolveFrozenCoreEntries,
   resolveStageCommand,
   snapshotFrozenCore,

package/lib/auto_runner.cjs

@@ -15,6 +15,7 @@ const {
   inferCampaignKind,
   isLocalProcessAlive,
   normalizeRequestedAutoContract,
+  resolveRequestedAutonomyLevel,
   resolveStageCommand,
   snapshotFrozenCore,
   snapshotPaths,
@@ -164,7 +165,10 @@ function buildRolledOverAutoMode(mode, requested, now) {
     campaignKind,
     campaignStartedAt: now.toISOString(),
     objective: requested.objective || mode.objective,
-    autonomyLevel: requested.autonomyLevel || mode.autonomyLevel || "l2",
+    autonomyLevel: resolveRequestedAutonomyLevel({
+      requestedAutonomyLevel: requested.autonomyLevel,
+      currentAutonomyLevel: mode.autonomyLevel,
+    }),
     approvalStatus: "draft",
     allowedStages,
     successCriteria: "",
@@ -489,6 +493,10 @@ async function startAutoMode({ targetDir, now = new Date(), requestedContract =
     }
     if (fit.classification === "hard-mismatch") {
       const normalizedRequest = normalizeRequestedAutoContract(requestedContract);
+      normalizedRequest.autonomyLevel = resolveRequestedAutonomyLevel({
+        requestedAutonomyLevel: normalizedRequest.autonomyLevel,
+        currentAutonomyLevel: mode.autonomyLevel,
+      });
       const { newMode, archivedPaths } = rolloverAutoCampaign({
         targetDir,
         mode,

package/lib/i18n.cjs

@@ -2168,7 +2168,7 @@ ZH_CONTENT[path.join(".lab", ".managed", "templates", "framing.md")] = `# 论文
 ZH_CONTENT[path.join(".codex", "prompts", "lab.md")] = codexPrompt(
   "查看 /lab 研究工作流总览并选择合适阶段",
   "workflow question 或 stage choice",
-  "# `/lab` for Codex\n\n`/lab` 是严格的研究工作流命令族。每次都使用同一套仓库工件和阶段边界。\n\n## 子命令\n\n- `/lab:idea`\n  先做两轮脑暴和两轮文献检索，再定义问题与 failure case、对比最接近前作，并输出带 approval gate 的 source-backed recommendation。\n\n- `/lab:data`\n  把已批准的 idea 转成数据集与 benchmark 方案，记录数据集年份、使用过该数据集的论文、下载来源、许可或访问限制，以及 classic-public、recent-strong-public、claim-specific 三类 benchmark 的纳入理由，和 canonical baselines、strong historical baselines、recent strong public methods、closest prior work 四类对比方法的纳入理由。\n\n- `/lab:auto`\n  在不改变 mission、framing 和核心 claims 的前提下，读取 eval-protocol 与 auto-mode 契约并自动编排 `run`、`iterate`、`review`、`report`，必要时扩展数据集、benchmark 和 comparison methods，并在满足升格策略时自动升级 primary package。启动前必须选定 autonomy level、声明 terminal goal，并显式写清 primary gate、secondary guard、promotion condition、stop reason 和 escalation reason，再批准契约。\n\n- `/lab:framing`\n  通过审计当前领域与相邻领域的术语，锁定 paper-facing 的方法名、模块名、论文题目和 contribution bullets，并在 section 起草前保留 approval gate。\n\n- `/lab:spec`\n  把已批准的 idea 转成 `.lab/changes/<change-id>/` 下的一个 lab change 目录，并在其中写出 `proposal`、`design`、`spec`、`tasks`。\n\n- `/lab:run`\n  执行最小有意义验证运行，登记 run，并生成第一版标准化评估摘要。\n\n- `/lab:iterate`\n  在冻结 mission、阈值、verification commands 与 `completion_promise` 的前提下执行有边界的实验迭代。\n\n- `/lab:review`\n  以 reviewer mode 审查文档或结果，先给短摘要，再输出 findings、fatal flaws、fix priority 和 residual risks。\n\n- `/lab:report`\n  从 runs 和 iterations 工件生成最终研究报告。\n\n- `/lab:write`\n  使用已安装 `lab` skill 下 vendored 的 paper-writing references，把稳定 report 工件转成论文 section。\n\n## 调度规则\n\n- 始终使用 `skills/lab/SKILL.md` 作为工作流合同。\n- 用户显式调用 `/lab:<stage>` 时，要立刻执行该 stage，而不是只推荐别的 `/lab` stage。\n- 先给简洁的阶段摘要；只要 stage 合同要求受管工件，就应立刻落盘，再回报输出路径和下一步。\n- 如果歧义会影响结论，一次只问一个问题；如果有多条可行路径，先给 2-3 个方案再收敛。\n- `/lab:spec` 前应已有经批准的数据集与 benchmark 方案。\n- `/lab:run`、`/lab:iterate`、`/lab:auto`、`/lab:report` 都应遵循 `.lab/context/eval-protocol.md`。\n- `.lab/context/eval-protocol.md` 不只定义主指标和主表，也应定义指标释义、实验阶梯，以及指标和对比实现的来源。\n- `/lab:auto` 只编排已批准边界内的执行阶段，不替代手动的 idea/data/framing/spec 决策。\n- `/lab:write` 前必须已有经批准的 `/lab:framing` 工件。\n\n## 如何输入 `/lab:auto`\n\n## `/lab:auto` 层级指南\n\n- `L1`：适合安全验证、一轮 bounded 真实运行，或简单 report 刷新。\n- `L2`：默认推荐级别，适合冻结核心边界内的常规实验迭代。\n- `L3`：激进 campaign 级别，只在你明确想做更大范围探索和可选写作时使用。\n- 如果不确定，默认推荐 `L2`。\n- 如果用户输入没写级别，或者把级别和 `paper layer`、`phase`、`table` 混用了，就应先停下来，要求用户明确选 `L1/L2/L3`。\n- 真正的 `/lab:auto` 首个可见输出必须是 `Auto preflight`。\n- 这个首个可见输出必须展示已读取文件，以及 `Autonomy level`、`Allowed stages`、`Terminal goal`、`Primary gate`、`Secondary guard`。\n- 如果无法从 `.lab/context/eval-protocol.md`、`.lab/context/auto-mode.md`、`.lab/context/auto-status.md` 和 `.lab/context/auto-outcome.md` 完成 preflight，就必须停下，而不是假装 loop 已经 armed。\n\n- 把 `Autonomy level L1/L2/L3` 视为执行权限级别，不要和论文里的 layer、phase、table 编号混用。\n- 把 `paper layer`、`phase`、`table` 视为实验目标。例如 `paper layer 3` 或 `Phase 1` 不是 `Autonomy level L3`。\n- 一条好的 `/lab:auto` 输入应至少说清：objective、自治级别、terminal goal、scope、allowed modifications。\n- 如果 workflow language 是中文，摘要、清单条目、任务标签和进度更新都应使用中文，除非文件路径、代码标识符或字面指标名必须保持原样。\n- 示例：`/lab:auto 自治级别 L2。目标：推进 paper layer 3。终止条件：完成 bounded protocol、测试、最小实现和一轮小规模结果。允许修改：配置、数据接入、评估脚本。`\n"
+  "# `/lab` for Codex\n\n`/lab` 是严格的研究工作流命令族。每次都使用同一套仓库工件和阶段边界。\n\n## 子命令\n\n- `/lab:idea`\n  先做两轮脑暴和两轮文献检索，再定义问题与 failure case、对比最接近前作，并输出带 approval gate 的 source-backed recommendation。\n\n- `/lab:data`\n  把已批准的 idea 转成数据集与 benchmark 方案，记录数据集年份、使用过该数据集的论文、下载来源、许可或访问限制，以及 classic-public、recent-strong-public、claim-specific 三类 benchmark 的纳入理由，和 canonical baselines、strong historical baselines、recent strong public methods、closest prior work 四类对比方法的纳入理由。\n\n- `/lab:auto`\n  在不改变 mission、framing 和核心 claims 的前提下，读取 eval-protocol 与 auto-mode 契约并自动编排 `run`、`iterate`、`review`、`report`，必要时扩展数据集、benchmark 和 comparison methods，并在满足升格策略时自动升级 primary package。启动前必须选定 autonomy level、声明 terminal goal，并显式写清 primary gate、secondary guard、promotion condition、stop reason 和 escalation reason，再批准契约。\n\n- `/lab:framing`\n  通过审计当前领域与相邻领域的术语，锁定 paper-facing 的方法名、模块名、论文题目和 contribution bullets，并在 section 起草前保留 approval gate。\n\n- `/lab:spec`\n  把已批准的 idea 转成 `.lab/changes/<change-id>/` 下的一个 lab change 目录，并在其中写出 `proposal`、`design`、`spec`、`tasks`。\n\n- `/lab:run`\n  执行最小有意义验证运行，登记 run，并生成第一版标准化评估摘要。\n\n- `/lab:iterate`\n  在冻结 mission、阈值、verification commands 与 `completion_promise` 的前提下执行有边界的实验迭代。\n\n- `/lab:review`\n  以 reviewer mode 审查文档或结果，先给短摘要，再输出 findings、fatal flaws、fix priority 和 residual risks。\n\n- `/lab:report`\n  从 runs 和 iterations 工件生成最终研究报告。\n\n- `/lab:write`\n  使用已安装 `lab` skill 下 vendored 的 paper-writing references，把稳定 report 工件转成论文 section。\n\n## 调度规则\n\n- 始终使用 `skills/lab/SKILL.md` 作为工作流合同。\n- 用户显式调用 `/lab:<stage>` 时，要立刻执行该 stage，而不是只推荐别的 `/lab` stage。\n- 先给简洁的阶段摘要；只要 stage 合同要求受管工件，就应立刻落盘，再回报输出路径和下一步。\n- 如果歧义会影响结论，一次只问一个问题；如果有多条可行路径，先给 2-3 个方案再收敛。\n- `/lab:spec` 前应已有经批准的数据集与 benchmark 方案。\n- `/lab:run`、`/lab:iterate`、`/lab:auto`、`/lab:report` 都应遵循 `.lab/context/eval-protocol.md`。\n- `.lab/context/eval-protocol.md` 不只定义主指标和主表，也应定义指标释义、实验阶梯，以及指标和对比实现的来源。\n- `/lab:auto` 只编排已批准边界内的执行阶段，不替代手动的 idea/data/framing/spec 决策。\n- `/lab:write` 前必须已有经批准的 `/lab:framing` 工件。\n\n## 如何输入 `/lab:auto`\n\n## `/lab:auto` 层级指南\n\n- 继续当前活动：`/lab-auto: 继续`。\n- 以最高执行级别继续当前活动：`/lab-auto: L3，继续`。\n- `L1`：适合安全验证、一轮 bounded 真实运行，或简单 report 刷新。\n- `L2`：默认推荐级别，适合冻结核心边界内的常规实验迭代。\n- `L3`：继续推进直到命中边界的级别；适合你明确想让 auto 在已批准 envelope 内默认继续，而不是为常规实现分歧停下。\n- 如果用户没写级别，默认按 `L2` 处理。\n- 接受短写：`L1`、`L2`、`L3`，以及小写 `l1/l2/l3`；`最高级别`、`最高自治` 也按 `L3` 处理。\n- 如果用户只写 `继续`，且当前已有 active 或可恢复 campaign，就直接继承当前 campaign 的级别。\n- 只有当级别本身真的有歧义时，才停下来追问，例如 `第三层`、`phase 3`、`table 3`。\n- 已批准的 `L2` 和 `L3` 执行 campaign 默认进入执行模式，而不是重新打开 brainstorming 或 spec-review。\n- 在执行模式里，不要把常规实现路径选择、helper script、路径修正、数据集适配、同 family 候选切换或普通自检重新路由到 brainstorming、spec review，或 reviewer/explorer 子智能体。\n- 只有当用户明确要求设计或审阅帮助、contract fit 需要新 campaign，或 escalation condition 明确要求独立复核时，才从执行模式切到设计或 reviewer 模式。\n- 真正的 `/lab:auto` 首个可见输出必须是 `Auto preflight`。\n- 这个首个可见输出必须展示已读取文件，以及 `Autonomy level`、`Allowed stages`、`Terminal goal`、`Primary gate`、`Secondary guard`。\n- 如果无法从 `.lab/context/eval-protocol.md`、`.lab/context/auto-mode.md`、`.lab/context/auto-status.md` 和 `.lab/context/auto-outcome.md` 完成 preflight，就必须停下，而不是假装 loop 已经 armed。\n\n- 把 `Autonomy level L1/L2/L3` 视为执行权限级别，不要和论文里的 layer、phase、table 编号混用。\n- 把 `paper layer`、`phase`、`table` 视为实验目标。例如 `paper layer 3` 或 `Phase 1` 不是 `Autonomy level L3`。\n- 一条好的 `/lab:auto` 输入应至少说清：objective、terminal goal、scope、allowed modifications；只有在你想覆盖默认或当前 campaign 时才需要额外写自治级别。\n- 如果 workflow language 是中文，摘要、清单条目、任务标签和进度更新都应使用中文，除非文件路径、代码标识符或字面指标名必须保持原样。\n- 示例：`/lab:auto: 继续`\n- 示例：`/lab:auto: L3，继续`\n- 示例：`/lab:auto 目标：推进 paper layer 3。终止条件：完成 bounded protocol、测试、最小实现和一轮小规模结果。允许修改：配置、数据接入、评估脚本。`\n"
 );
 
 ZH_CONTENT[path.join(".codex", "prompts", "lab-data.md")] = codexPrompt(
@@ -2180,14 +2180,14 @@ ZH_CONTENT[path.join(".codex", "prompts", "lab-data.md")] = codexPrompt(
 ZH_CONTENT[path.join(".codex", "prompts", "lab-auto.md")] = codexPrompt(
   "在已批准边界内编排自动实验循环",
   "auto mode objective",
-  "使用已安装的 `lab` 技能：`.codex/skills/lab/SKILL.md`。\n\n立刻针对用户当前给出的参数执行 `/lab:auto`，不要只推荐别的 `/lab` 阶段。只有在缺少阻塞性前提时，才明确指出缺什么，并且一次最多追问一个问题。\n\n先把用户请求规范化成可交给 CLI 的 auto contract 字段：objective、autonomy level、campaign kind、allowed stages，以及任何不改变范围就能明确的 terminal-goal 提示。\n你的第一步执行动作必须是对当前项目运行 `superlab auto start`，而不是自己直接改写 `.lab/context/auto-mode.md`、`.lab/context/auto-status.md`、`.lab/context/auto-ledger.md` 或 `.lab/context/auto-outcome.md`。\n把规范化后的字段通过 CLI 参数传下去，包括 `--objective`、`--campaign-kind`、`--allowed-stages`，以及在用户已明确或已隐含时传 `--autonomy-level`。\nCLI 返回后的 runtime 结果才是事实来源。如果 CLI 报 rollover、conflict、缺字段，或已经成功启动 campaign，就如实回报，不要绕过 CLI 自己做 prompt 侧写回。\n\n本命令运行 `/lab:auto` 阶段。它必须读取 `.lab/context/eval-protocol.md`、`.lab/context/auto-mode.md`、`.lab/context/auto-status.md`、`.lab/context/auto-ledger.md` 与 `.lab/context/auto-outcome.md`，先确认 autonomy level、approval status、terminal goal schema，以及 primary gate、secondary guard、promotion condition、stop reason、escalation reason，再把 eval-protocol 里的指标释义、主表计划、来源约束与结构化实验阶梯当作执行依据，在不修改 mission、framing 和核心 claims 的前提下编排已批准的 `run`、`iterate`、`review`、`report`，轮询长任务完成情况；如果声明了 rung，就保持会话活着并按 rung 转移继续推进。\n首个可见输出块必须是 `Auto preflight`。这个块必须列出已读取文件，并回显 `Autonomy level`、`Approval status`、`Allowed stages`、`Terminal goal`、`Primary gate` 和 `Secondary guard`，然后才能进入执行摘要或动作计划。\n如果 preflight 所需字段缺失、过期或彼此冲突，就必须在执行前停下，并明确指出到底是哪一个字段阻止了 loop 启动。\n当 loop 活着时，必须把当前 owner、观察状态、checkpoint 摘要、继续边界、停止边界和恢复读取集合写进 `.lab/context/auto-ledger.md`。\n如果仓库的 workflow language 是中文，摘要、清单条目、任务标签和进度更新都必须使用中文，除非某个文件路径、代码标识符或字面指标名必须保持原样。\n把 `Layer 3`、`Phase 1`、`Table 2` 这类表达视为论文范围目标；只有显式写成 `Autonomy level L3` 或 `自治级别 L3` 时，才把它当成执行权限级别。\n不要用 `sleep 30`、单次 `pgrep` 或一次性的 `metrics.json` 探针来代替真实长任务命令；当真实实验进程还活着时，只允许发进度更新并继续等待。"
+  "使用已安装的 `lab` 技能：`.codex/skills/lab/SKILL.md`。\n\n继续当前活动：`/lab-auto: 继续`。\n以最高执行级别继续当前活动：`/lab-auto: L3，继续`。\n\n立刻针对用户当前给出的参数执行 `/lab:auto`，不要只推荐别的 `/lab` 阶段。只有在缺少阻塞性前提时，才明确指出缺什么，并且一次最多追问一个问题。\n\n先把用户请求规范化成可交给 CLI 的 auto contract 字段：objective、autonomy level、campaign kind、allowed stages，以及任何不改变范围就能明确的 terminal-goal 提示。\n如果用户没写级别，默认按 `L2` 处理；接受 `L1/L2/L3`、`l1/l2/l3` 这类短写，`最高级别`、`最高自治` 也按 `L3` 处理。\n如果用户只写了 `继续`，且当前已有 active 或可恢复 campaign，就直接继承当前 campaign 的级别，而不是要求用户重复写。\n如果你想沿用 runtime 里已存的 campaign 级别继续，就直接写 `/lab-auto: 继续`。\n只有当级别本身真的有歧义时，才停下来追问，例如 `第三层`、`phase 3`、`table 3`。\n已批准的 `L2` 和 `L3` 执行 campaign 默认进入执行模式。\n在执行模式里，不要进入 brainstorming，不要进入 spec review，也不要为了常规实现路径选择、helper script、路径修正、数据集适配、同 family 候选切换或普通自检而生成 reviewer、explorer 或其他子智能体循环。\n只有当用户明确要求设计或审阅帮助、contract fit 需要新 campaign，或 contract 的 escalation condition 明确要求独立复核时，才从执行模式切到设计或 reviewer 模式。\n你的第一步执行动作必须是对当前项目运行 `superlab auto start`，而不是自己直接改写 `.lab/context/auto-mode.md`、`.lab/context/auto-status.md`、`.lab/context/auto-ledger.md` 或 `.lab/context/auto-outcome.md`。\n把规范化后的字段通过 CLI 参数传下去，包括 `--objective`、`--campaign-kind`、`--allowed-stages`，以及在用户已明确或已隐含时传 `--autonomy-level`。\nCLI 返回后的 runtime 结果才是事实来源。如果 CLI 报 rollover、conflict、缺字段，或已经成功启动 campaign，就如实回报，不要绕过 CLI 自己做 prompt 侧写回。\n\n本命令运行 `/lab:auto` 阶段。它必须读取 `.lab/context/eval-protocol.md`、`.lab/context/auto-mode.md`、`.lab/context/auto-status.md`、`.lab/context/auto-ledger.md` 与 `.lab/context/auto-outcome.md`，先确认 autonomy level、approval status、terminal goal schema，以及 primary gate、secondary guard、promotion condition、stop reason、escalation reason，再把 eval-protocol 里的指标释义、主表计划、来源约束与结构化实验阶梯当作执行依据，在不修改 mission、framing 和核心 claims 的前提下编排已批准的 `run`、`iterate`、`review`、`report`，轮询长任务完成情况；如果声明了 rung，就保持会话活着并按 rung 转移继续推进。\n首个可见输出块必须是 `Auto preflight`。这个块必须列出已读取文件，并回显 `Autonomy level`、`Approval status`、`Allowed stages`、`Terminal goal`、`Primary gate` 和 `Secondary guard`，然后才能进入执行摘要或动作计划。\n如果 preflight 所需字段缺失、过期或彼此冲突，就必须在执行前停下，并明确指出到底是哪一个字段阻止了 loop 启动。\n当 loop 活着时，必须把当前 owner、观察状态、checkpoint 摘要、继续边界、停止边界和恢复读取集合写进 `.lab/context/auto-ledger.md`。\n如果仓库的 workflow language 是中文，摘要、清单条目、任务标签和进度更新都必须使用中文，除非某个文件路径、代码标识符或字面指标名必须保持原样。\n把 `Layer 3`、`Phase 1`、`Table 2` 这类表达视为论文范围目标；只有显式写成 `Autonomy level L3` 或 `自治级别 L3` 时，才把它当成执行权限级别。\n不要用 `sleep 30`、单次 `pgrep` 或一次性的 `metrics.json` 探针来代替真实长任务命令；当真实实验进程还活着时，只允许发进度更新并继续等待。"
 );
 
 ZH_CONTENT[path.join(".claude", "commands", "lab.md")] = claudeCommand(
   "lab",
   "查看 /lab 研究工作流总览并选择合适阶段",
   "[stage] [target]",
-  "# `/lab` for Claude\n\n`/lab` 是 Claude Code 里的 lab 工作流分发入口。调用方式有两种：\n\n- `/lab <stage> ...`\n- `/lab-idea`、`/lab-data`、`/lab-auto`、`/lab-framing`、`/lab-spec`、`/lab-run`、`/lab-iterate`、`/lab-review`、`/lab-report`、`/lab-write`\n\n## 阶段别名\n\n- `/lab idea ...` 或 `/lab-idea`\n- `/lab data ...` 或 `/lab-data`\n- `/lab auto ...` 或 `/lab-auto`\n- `/lab framing ...` 或 `/lab-framing`\n- `/lab spec ...` 或 `/lab-spec`\n- `/lab run ...` 或 `/lab-run`\n- `/lab iterate ...` 或 `/lab-iterate`\n- `/lab review ...` 或 `/lab-review`\n- `/lab report ...` 或 `/lab-report`\n- `/lab write ...` 或 `/lab-write`\n\n## 调度规则\n\n- 始终使用 `skills/lab/SKILL.md` 作为工作流合同。\n- 用户显式调用 `/lab <stage> ...` 或 `/lab-<stage>` 时，要立刻执行该 stage，而不是只推荐别的阶段。\n- 先给简洁的阶段摘要；只要 stage 合同要求受管工件，就应立刻落盘，再回报输出路径和下一步。\n- 如果歧义会影响结论，一次只问一个问题；如果有多条可行路径，先给 2-3 个方案再收敛。\n- `spec` 前应已有经批准的数据集与 benchmark 方案。\n- `run`、`iterate`、`auto`、`report` 都应遵循 `.lab/context/eval-protocol.md`。\n- `auto` 只编排已批准边界内的执行阶段，不替代手动的 idea/data/framing/spec 决策。\n- `write` 前必须已有经批准的 `framing` 工件。\n\n## 如何输入 `/lab auto`\n\n## `/lab auto` 层级指南\n\n- `L1`：适合安全验证、一轮 bounded 真实运行，或简单 report 刷新。\n- `L2`：默认推荐级别，适合冻结核心边界内的常规实验迭代。\n- `L3`：激进 campaign 级别，只在你明确想做更大范围探索和可选写作时使用。\n- 如果不确定，默认推荐 `L2`。\n- 如果用户输入没写级别，或者把级别和 `paper layer`、`phase`、`table` 混用了，就应先停下来，要求用户明确选 `L1/L2/L3`。\n- 真正的 `/lab auto` 首个可见输出必须是 `Auto preflight`。\n- 这个首个可见输出必须展示已读取文件，以及 `Autonomy level`、`Allowed stages`、`Terminal goal`、`Primary gate`、`Secondary guard`。\n- 如果无法从 `.lab/context/eval-protocol.md`、`.lab/context/auto-mode.md`、`.lab/context/auto-status.md` 和 `.lab/context/auto-outcome.md` 完成 preflight，就必须停下，而不是假装 loop 已经 armed。\n\n- 把 `Autonomy level L1/L2/L3` 视为执行权限级别，不要和论文里的 layer、phase、table 编号混用。\n- 把 `paper layer`、`phase`、`table` 视为实验目标。例如 `paper layer 3` 或 `Phase 1` 不是 `Autonomy level L3`。\n- 一条好的 `/lab auto` 输入应至少说清：objective、自治级别、terminal goal、scope、allowed modifications。\n- 如果 workflow language 是中文，摘要、清单条目、任务标签和进度更新都应使用中文，除非文件路径、代码标识符或字面指标名必须保持原样。\n- 示例：`/lab auto 自治级别 L2。目标：推进 paper layer 3。终止条件：完成 bounded protocol、测试、最小实现和一轮小规模结果。允许修改：配置、数据接入、评估脚本。`\n"
+  "# `/lab` for Claude\n\n`/lab` 是 Claude Code 里的 lab 工作流分发入口。调用方式有两种：\n\n- `/lab <stage> ...`\n- `/lab-idea`、`/lab-data`、`/lab-auto`、`/lab-framing`、`/lab-spec`、`/lab-run`、`/lab-iterate`、`/lab-review`、`/lab-report`、`/lab-write`\n\n## 阶段别名\n\n- `/lab idea ...` 或 `/lab-idea`\n- `/lab data ...` 或 `/lab-data`\n- `/lab auto ...` 或 `/lab-auto`\n- `/lab framing ...` 或 `/lab-framing`\n- `/lab spec ...` 或 `/lab-spec`\n- `/lab run ...` 或 `/lab-run`\n- `/lab iterate ...` 或 `/lab-iterate`\n- `/lab review ...` 或 `/lab-review`\n- `/lab report ...` 或 `/lab-report`\n- `/lab write ...` 或 `/lab-write`\n\n## 调度规则\n\n- 始终使用 `skills/lab/SKILL.md` 作为工作流合同。\n- 用户显式调用 `/lab <stage> ...` 或 `/lab-<stage>` 时，要立刻执行该 stage，而不是只推荐别的阶段。\n- 先给简洁的阶段摘要；只要 stage 合同要求受管工件，就应立刻落盘，再回报输出路径和下一步。\n- 如果歧义会影响结论，一次只问一个问题；如果有多条可行路径，先给 2-3 个方案再收敛。\n- `spec` 前应已有经批准的数据集与 benchmark 方案。\n- `run`、`iterate`、`auto`、`report` 都应遵循 `.lab/context/eval-protocol.md`。\n- `auto` 只编排已批准边界内的执行阶段，不替代手动的 idea/data/framing/spec 决策。\n- `write` 前必须已有经批准的 `framing` 工件。\n\n## 如何输入 `/lab auto`\n\n## `/lab auto` 层级指南\n\n- 继续当前活动：`/lab-auto: 继续`。\n- 以最高执行级别继续当前活动：`/lab-auto: L3，继续`。\n- `L1`：适合安全验证、一轮 bounded 真实运行，或简单 report 刷新。\n- `L2`：默认推荐级别，适合冻结核心边界内的常规实验迭代。\n- `L3`：继续推进直到命中边界的级别；适合你明确想让 auto 在已批准 envelope 内默认继续，而不是为常规实现分歧停下。\n- 如果用户没写级别，默认按 `L2` 处理。\n- 接受短写：`L1`、`L2`、`L3`，以及小写 `l1/l2/l3`；`最高级别`、`最高自治` 也按 `L3` 处理。\n- 如果用户只写 `继续`，且当前已有 active 或可恢复 campaign，就直接继承当前 campaign 的级别。\n- 只有当级别本身真的有歧义时，才停下来追问，例如 `第三层`、`phase 3`、`table 3`。\n- 已批准的 `L2` 和 `L3` 执行 campaign 默认进入执行模式，而不是重新打开 brainstorming 或 spec-review。\n- 在执行模式里，不要把常规实现路径选择、helper script、路径修正、数据集适配、同 family 候选切换或普通自检重新路由到 brainstorming、spec review，或 reviewer/explorer 子智能体。\n- 只有当用户明确要求设计或审阅帮助、contract fit 需要新 campaign，或 escalation condition 明确要求独立复核时，才从执行模式切到设计或 reviewer 模式。\n- 真正的 `/lab auto` 首个可见输出必须是 `Auto preflight`。\n- 这个首个可见输出必须展示已读取文件，以及 `Autonomy level`、`Allowed stages`、`Terminal goal`、`Primary gate`、`Secondary guard`。\n- 如果无法从 `.lab/context/eval-protocol.md`、`.lab/context/auto-mode.md`、`.lab/context/auto-status.md` 和 `.lab/context/auto-outcome.md` 完成 preflight，就必须停下，而不是假装 loop 已经 armed。\n\n- 把 `Autonomy level L1/L2/L3` 视为执行权限级别，不要和论文里的 layer、phase、table 编号混用。\n- 把 `paper layer`、`phase`、`table` 视为实验目标。例如 `paper layer 3` 或 `Phase 1` 不是 `Autonomy level L3`。\n- 一条好的 `/lab auto` 输入应至少说清：objective、terminal goal、scope、allowed modifications；只有在你想覆盖默认或当前 campaign 时才需要额外写自治级别。\n- 如果 workflow language 是中文，摘要、清单条目、任务标签和进度更新都应使用中文，除非文件路径、代码标识符或字面指标名必须保持原样。\n- 示例：`/lab-auto: 继续`\n- 示例：`/lab-auto: L3，继续`\n- 示例：`/lab auto 目标：推进 paper layer 3。终止条件：完成 bounded protocol、测试、最小实现和一轮小规模结果。允许修改：配置、数据接入、评估脚本。`\n"
 );
 
 ZH_CONTENT[path.join(".claude", "commands", "lab-data.md")] = claudeCommand(
@@ -2201,7 +2201,40 @@ ZH_CONTENT[path.join(".claude", "commands", "lab-auto.md")] = claudeCommand(
   "lab-auto",
   "在已批准边界内编排自动实验循环",
   "auto mode objective",
-  "使用已安装的 `lab` 技能：`.claude/skills/lab/SKILL.md`。\n\n立刻针对用户当前给出的参数执行 `auto` 阶段，不要只推荐别的 lab 阶段。只有在缺少阻塞性前提时，才明确指出缺什么，并且一次最多追问一个问题。\n\n先把用户请求规范化成可交给 CLI 的 auto contract 字段：objective、autonomy level、campaign kind、allowed stages，以及任何不改变范围就能明确的 terminal-goal 提示。\n你的第一步执行动作必须是对当前项目运行 `superlab auto start`，而不是自己直接改写 `.lab/context/auto-mode.md`、`.lab/context/auto-status.md`、`.lab/context/auto-ledger.md` 或 `.lab/context/auto-outcome.md`。\n把规范化后的字段通过 CLI 参数传下去，包括 `--objective`、`--campaign-kind`、`--allowed-stages`，以及在用户已明确或已隐含时传 `--autonomy-level`。\nCLI 返回后的 runtime 结果才是事实来源。如果 CLI 报 rollover、conflict、缺字段，或已经成功启动 campaign，就如实回报，不要绕过 CLI 自己做 prompt 侧写回。\n\n本命令运行 lab workflow 的 `auto` 阶段。它必须读取 `.lab/context/eval-protocol.md`、`.lab/context/auto-mode.md`、`.lab/context/auto-status.md`、`.lab/context/auto-ledger.md` 与 `.lab/context/auto-outcome.md`，先确认 autonomy level、approval status、terminal goal schema，以及 primary gate、secondary guard、promotion condition、stop reason、escalation reason，再把 eval-protocol 里的指标释义、主表计划、来源约束与结构化实验阶梯当作执行依据，在不修改 mission、framing 和核心 claims 的前提下编排已批准的 `run`、`iterate`、`review`、`report`，轮询长任务完成情况；如果声明了 rung，就保持会话活着并按 rung 转移继续推进。\n首个可见输出块必须是 `Auto preflight`。这个块必须列出已读取文件，并回显 `Autonomy level`、`Approval status`、`Allowed stages`、`Terminal goal`、`Primary gate` 和 `Secondary guard`，然后才能进入执行摘要或动作计划。\n如果 preflight 所需字段缺失、过期或彼此冲突，就必须在执行前停下，并明确指出到底是哪一个字段阻止了 loop 启动。\n当 loop 活着时，必须把当前 owner、观察状态、checkpoint 摘要、继续边界、停止边界和恢复读取集合写进 `.lab/context/auto-ledger.md`。\n如果仓库的 workflow language 是中文，摘要、清单条目、任务标签和进度更新都必须使用中文，除非某个文件路径、代码标识符或字面指标名必须保持原样。\n把 `Layer 3`、`Phase 1`、`Table 2` 这类表达视为论文范围目标；只有显式写成 `Autonomy level L3` 或 `自治级别 L3` 时，才把它当成执行权限级别。\n不要用 `sleep 30`、单次 `pgrep` 或一次性的 `metrics.json` 探针来代替真实长任务命令；当真实实验进程还活着时，只允许发进度更新并继续等待。"
+  "使用已安装的 `lab` 技能：`.claude/skills/lab/SKILL.md`。\n\n继续当前活动：`/lab-auto: 继续`。\n以最高执行级别继续当前活动：`/lab-auto: L3，继续`。\n\n立刻针对用户当前给出的参数执行 `auto` 阶段，不要只推荐别的 lab 阶段。只有在缺少阻塞性前提时，才明确指出缺什么，并且一次最多追问一个问题。\n\n先把用户请求规范化成可交给 CLI 的 auto contract 字段：objective、autonomy level、campaign kind、allowed stages，以及任何不改变范围就能明确的 terminal-goal 提示。\n如果用户没写级别，默认按 `L2` 处理；接受 `L1/L2/L3`、`l1/l2/l3` 这类短写，`最高级别`、`最高自治` 也按 `L3` 处理。\n如果用户只写了 `继续`，且当前已有 active 或可恢复 campaign，就直接继承当前 campaign 的级别，而不是要求用户重复写。\n如果你想沿用 runtime 里已存的 campaign 级别继续，就直接写 `/lab-auto: 继续`。\n只有当级别本身真的有歧义时，才停下来追问，例如 `第三层`、`phase 3`、`table 3`。\n已批准的 `L2` 和 `L3` 执行 campaign 默认进入执行模式。\n在执行模式里，不要进入 brainstorming，不要进入 spec review，也不要为了常规实现路径选择、helper script、路径修正、数据集适配、同 family 候选切换或普通自检而生成 reviewer、explorer 或其他子智能体循环。\n只有当用户明确要求设计或审阅帮助、contract fit 需要新 campaign，或 contract 的 escalation condition 明确要求独立复核时，才从执行模式切到设计或 reviewer 模式。\n你的第一步执行动作必须是对当前项目运行 `superlab auto start`，而不是自己直接改写 `.lab/context/auto-mode.md`、`.lab/context/auto-status.md`、`.lab/context/auto-ledger.md` 或 `.lab/context/auto-outcome.md`。\n把规范化后的字段通过 CLI 参数传下去，包括 `--objective`、`--campaign-kind`、`--allowed-stages`，以及在用户已明确或已隐含时传 `--autonomy-level`。\nCLI 返回后的 runtime 结果才是事实来源。如果 CLI 报 rollover、conflict、缺字段，或已经成功启动 campaign，就如实回报，不要绕过 CLI 自己做 prompt 侧写回。\n\n本命令运行 lab workflow 的 `auto` 阶段。它必须读取 `.lab/context/eval-protocol.md`、`.lab/context/auto-mode.md`、`.lab/context/auto-status.md`、`.lab/context/auto-ledger.md` 与 `.lab/context/auto-outcome.md`，先确认 autonomy level、approval status、terminal goal schema，以及 primary gate、secondary guard、promotion condition、stop reason、escalation reason，再把 eval-protocol 里的指标释义、主表计划、来源约束与结构化实验阶梯当作执行依据，在不修改 mission、framing 和核心 claims 的前提下编排已批准的 `run`、`iterate`、`review`、`report`，轮询长任务完成情况；如果声明了 rung，就保持会话活着并按 rung 转移继续推进。\n首个可见输出块必须是 `Auto preflight`。这个块必须列出已读取文件，并回显 `Autonomy level`、`Approval status`、`Allowed stages`、`Terminal goal`、`Primary gate` 和 `Secondary guard`，然后才能进入执行摘要或动作计划。\n如果 preflight 所需字段缺失、过期或彼此冲突，就必须在执行前停下，并明确指出到底是哪一个字段阻止了 loop 启动。\n当 loop 活着时，必须把当前 owner、观察状态、checkpoint 摘要、继续边界、停止边界和恢复读取集合写进 `.lab/context/auto-ledger.md`。\n如果仓库的 workflow language 是中文，摘要、清单条目、任务标签和进度更新都必须使用中文，除非某个文件路径、代码标识符或字面指标名必须保持原样。\n把 `Layer 3`、`Phase 1`、`Table 2` 这类表达视为论文范围目标；只有显式写成 `Autonomy level L3` 或 `自治级别 L3` 时，才把它当成执行权限级别。\n不要用 `sleep 30`、单次 `pgrep` 或一次性的 `metrics.json` 探针来代替真实长任务命令；当真实实验进程还活着时，只允许发进度更新并继续等待。"
+);
+
+const zhAutoPriorityCodexLine =
+  "显式的 `/lab:auto` 或 `/lab-auto` 请求，其优先级高于 brainstorming、spec review 这类更宽的创作或审阅技能路径。";
+const zhAutoPriorityClaudeLine =
+  "显式的 `/lab auto` 或 `/lab-auto` 请求，其优先级高于 brainstorming、spec review 这类更宽的创作或审阅技能路径。";
+
+ZH_CONTENT[path.join(".codex", "prompts", "lab.md")] = ZH_CONTENT[
+  path.join(".codex", "prompts", "lab.md")
+].replace(
+  "- 已批准的 `L2` 和 `L3` 执行 campaign 默认进入执行模式，而不是重新打开 brainstorming 或 spec-review。",
+  `- ${zhAutoPriorityCodexLine}\n- 已批准的 \`L2\` 和 \`L3\` 执行 campaign 默认进入执行模式，而不是重新打开 brainstorming 或 spec-review。`
+);
+
+ZH_CONTENT[path.join(".codex", "prompts", "lab-auto.md")] = ZH_CONTENT[
+  path.join(".codex", "prompts", "lab-auto.md")
+].replace(
+  "已批准的 `L2` 和 `L3` 执行 campaign 默认进入执行模式。",
+  `${zhAutoPriorityCodexLine}\n已批准的 \`L2\` 和 \`L3\` 执行 campaign 默认进入执行模式。`
+);
+
+ZH_CONTENT[path.join(".claude", "commands", "lab.md")] = ZH_CONTENT[
+  path.join(".claude", "commands", "lab.md")
+].replace(
+  "- 已批准的 `L2` 和 `L3` 执行 campaign 默认进入执行模式，而不是重新打开 brainstorming 或 spec-review。",
+  `- ${zhAutoPriorityClaudeLine}\n- 已批准的 \`L2\` 和 \`L3\` 执行 campaign 默认进入执行模式，而不是重新打开 brainstorming 或 spec-review。`
+);
+
+ZH_CONTENT[path.join(".claude", "commands", "lab-auto.md")] = ZH_CONTENT[
+  path.join(".claude", "commands", "lab-auto.md")
+].replace(
+  "已批准的 `L2` 和 `L3` 执行 campaign 默认进入执行模式。",
+  `${zhAutoPriorityClaudeLine}\n已批准的 \`L2\` 和 \`L3\` 执行 campaign 默认进入执行模式。`
 );
 
 const zhRecipeQuickPathLine =
@@ -2763,7 +2796,7 @@ ZH_CONTENT[path.join(".codex", "skills", "lab", "stages", "auto.md")] = `# \`/la
 - 级别含义固定为：
   - \`L1\`：safe run，只允许 \`run\`、\`review\`、\`report\`
   - \`L2\`：bounded iteration，允许 \`run\`、\`iterate\`、\`review\`、\`report\`
-  - \`L3\`：aggressive campaign，才允许额外编排 \`write\`
+  - \`L3\`：在已批准 envelope 内持续推进直到命中边界；framing 已批准时才可额外编排 \`write\`
 - 默认只编排 \`run\`、\`iterate\`、\`review\`、\`report\`；只有 framing 已批准时才可选 \`write\`。
 - 不要自动修改 mission、paper-facing framing 或核心 claims。
 - 可以在 exploration envelope 内增加数据集、benchmark 和 comparison methods。
@@ -2797,10 +2830,16 @@ ZH_CONTENT[path.join(".codex", "skills", "lab", "stages", "auto.md")] = `# \`/la
 
 ## 层级指南
 
+- 继续当前活动：\`/lab-auto: 继续\`。
+- 以最高执行级别继续当前活动：\`/lab-auto: L3，继续\`。
 - \`L1\` = safe validation
 - \`L2\` = 默认推荐的 bounded iteration
-- \`L3\` = aggressive campaign
+- \`L3\` = 在已批准 envelope 内持续推进直到命中边界
 - 如果不确定，默认推荐 \`L2\`。
+- 显式的 \`/lab:auto\` 或 \`/lab-auto\` 请求，其优先级高于 brainstorming、spec review 这类更宽的创作或审阅技能路径。
+- 已批准的 \`L2\` 和 \`L3\` 执行 campaign 默认进入执行模式，而不是重新打开 brainstorming 或 spec-review。
+- 在执行模式里，不要把常规实现路径选择、helper script、路径修正、数据集适配、同 family 候选切换或普通自检重新路由到 brainstorming、spec review，或 reviewer/explorer 子智能体。
+- 只有当用户明确要求设计或审阅帮助、contract fit 需要新 campaign，或 escalation condition 明确要求独立复核时，才从执行模式切到设计或 reviewer 模式。
 
 - 开始前先简洁说明：objective、frozen core 和下一自动阶段。
 - 如果契约本身不完整，一次只追问一个问题。
@@ -2808,12 +2847,11 @@ ZH_CONTENT[path.join(".codex", "skills", "lab", "stages", "auto.md")] = `# \`/la
 - 只有当下一步会离开已批准的 exploration envelope、超出选定 autonomy level，或实质改变 frozen core 时，才保留人工 approval gate。
 - 每次进入 \`/lab:auto\` 都要先给出这份层级指南。
 - 先做输入归一化：把 \`Autonomy level L1/L2/L3\` 视为执行权限级别，把 \`Layer 3\`、\`Phase 1\`、\`Table 2\` 视为论文范围目标。
-- 如果用户没有写自治级别，或者把自治级别和论文层、phase、table 混用了，就必须先给一版更详细的层级说明，至少解释：
-  - \`L1/L2/L3\` 的典型适用场景
-  - 每个级别允许改什么
-  - 每个级别通常在什么 stop boundary 停下
-  - 如果不确定，默认推荐 \`L2\`
-- 给完这版详细说明后，再追问一个明确的 \`L1/L2/L3\` 选择；在用户明确选级别前不要启动循环。
+- 如果用户没有写自治级别，默认按 \`L2\` 处理，不要为了“只差一个级别”先停下来。
+- 接受 \`L1/L2/L3\`、\`l1/l2/l3\` 这类短写，\`最高级别\`、\`最高自治\` 也按 \`L3\` 处理。
+- 如果用户只写 \`继续\`，且当前已有 active 或可恢复 campaign，就直接继承当前 campaign 的级别。
+- 只有当级别本身真的有歧义时，才停下来追问，例如 \`第三层\`、\`phase 3\`、\`table 3\`。
+- 如果用户显式调用 \`/lab:auto\` 或 \`/lab-auto\`，就保持在 auto 执行路径里；只要请求仍在已批准 execution envelope 内，即使目标听起来像 feature selection、baseline selection、离散化或 candidate sweep，也不要重新路由到 brainstorming 或 spec review。
 - 如果用户同时提了论文层、实验 phase 和自治级别，先用一句话重述：objective、自治级别、terminal goal、scope、allowed modifications。
 - 如果 workflow language 是中文，摘要、清单条目、任务标签和进度更新都应使用中文，除非文件路径、代码标识符或字面指标名必须保持原样。
 - 当循环进入 \`report\` 时，要主动给出用户可读的白话总结，解释主指标、次级指标和主表作用；不要等用户额外发一句“解释这些指标”。
@@ -2839,6 +2877,17 @@ ZH_CONTENT[path.join(".codex", "skills", "lab", "stages", "auto.md")] = `# \`/la
 - 当真实实验进程还活着时，只允许发进度更新并继续等待，不能把这一 rung 当作已经完成。
 `;
 
+ZH_CONTENT[path.join(".claude", "skills", "lab", "stages", "auto.md")] =
+  ZH_CONTENT[path.join(".codex", "skills", "lab", "stages", "auto.md")];
+ZH_CONTENT[path.join(".codex", "skills", "lab", "stages", "auto.md")] = ZH_CONTENT[
+  path.join(".codex", "skills", "lab", "stages", "auto.md")
+].replace(
+  "- 已批准的 \\`L2\\` 和 \\`L3\\` 执行 campaign 默认进入执行模式，而不是重新打开 brainstorming 或 spec-review。",
+  "- 显式的 \\`/lab:auto\\` 或 \\`/lab-auto\\` 请求，其优先级高于 brainstorming、spec review 这类更宽的创作或审阅技能路径。\\n- 已批准的 \\`L2\\` 和 \\`L3\\` 执行 campaign 默认进入执行模式，而不是重新打开 brainstorming 或 spec-review。"
+).replace(
+  "- 只有当级别本身真的有歧义时，才停下来追问，例如 \\`第三层\\`、\\`phase 3\\`、\\`table 3\\`。",
+  "- 只有当级别本身真的有歧义时，才停下来追问，例如 \\`第三层\\`、\\`phase 3\\`、\\`table 3\\`。\n- 如果用户显式调用 \\`/lab:auto\\` 或 \\`/lab-auto\\`，就保持在 auto 执行路径里；只要请求仍在已批准 execution envelope 内，即使目标听起来像 feature selection、baseline selection、离散化或 candidate sweep，也不要重新路由到 brainstorming 或 spec review。"
+);
 ZH_CONTENT[path.join(".claude", "skills", "lab", "stages", "auto.md")] =
   ZH_CONTENT[path.join(".codex", "skills", "lab", "stages", "auto.md")];
 ZH_CONTENT[path.join(".claude", "skills", "lab", "stages", "report.md")] =

package/package-assets/claude/commands/lab/auto.md

@@ -6,8 +6,19 @@ argument-hint: autonomous campaign target
 
 Use the installed `lab` skill at `.claude/skills/lab/SKILL.md`.
 
+Quick continue: `/lab-auto: 继续`.
+Quick continue at the highest execution level: `/lab-auto: L3，继续`.
+
 Execute the requested `/lab-auto` command against the user's argument now. Do not only recommend another lab stage. If a blocking prerequisite is missing, say exactly what is missing and ask at most one clarifying question.
 First normalize the user's request into CLI-ready auto contract fields: objective, autonomy level, campaign kind, allowed stages, and any explicit terminal-goal hints that can be inferred without changing scope.
+If the user omits the level, default to `L2`. Accept short forms like `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts such as `最高级别` or `最高自治` for `L3`.
+If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+To continue the current campaign with the stored runtime level, say `/lab-auto: 继续`.
+Only stop for a level clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+An explicit `/lab auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+Approved `L2` and `L3` execution campaigns default to execution mode.
+In execution mode, do not enter brainstorming, do not enter spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 Your first execution action must be to run `superlab auto start` against the current project instead of directly rewriting `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, or `.lab/context/auto-outcome.md` yourself.
 Pass the normalized fields through CLI flags, including `--objective`, `--campaign-kind`, `--allowed-stages`, and `--autonomy-level` when the user specified or implied them.
 After that CLI handoff, treat the CLI/runtime result as the source of truth. If it reports rollover, conflict, missing contract fields, or a started campaign, reflect that result back to the user instead of bypassing it with prompt-side edits.

package/package-assets/claude/commands/lab-auto.md

@@ -6,8 +6,19 @@ argument-hint: autonomous campaign target
 
 Use the installed `lab` skill at `.claude/skills/lab/SKILL.md`.
 
+Quick continue: `/lab-auto: 继续`.
+Quick continue at the highest execution level: `/lab-auto: L3，继续`.
+
 Execute the requested `/lab-auto` command against the user's argument now. Do not only recommend another lab stage. If a blocking prerequisite is missing, say exactly what is missing and ask at most one clarifying question.
 First normalize the user's request into CLI-ready auto contract fields: objective, autonomy level, campaign kind, allowed stages, and any explicit terminal-goal hints that can be inferred without changing scope.
+If the user omits the level, default to `L2`. Accept short forms like `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts such as `最高级别` or `最高自治` for `L3`.
+If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+To continue the current campaign with the stored runtime level, say `/lab-auto: 继续`.
+Only stop for a level clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+An explicit `/lab auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+Approved `L2` and `L3` execution campaigns default to execution mode.
+In execution mode, do not enter brainstorming, do not enter spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 Your first execution action must be to run `superlab auto start` against the current project instead of directly rewriting `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, or `.lab/context/auto-outcome.md` yourself.
 Pass the normalized fields through CLI flags, including `--objective`, `--campaign-kind`, `--allowed-stages`, and `--autonomy-level` when the user specified or implied them.
 After that CLI handoff, treat the CLI/runtime result as the source of truth. If it reports rollover, conflict, missing contract fields, or a started campaign, reflect that result back to the user instead of bypassing it with prompt-side edits.

package/package-assets/claude/commands/lab.md

@@ -81,11 +81,19 @@ Treat all of these as equivalent stage requests:
 
 ## Level Guide for `/lab auto`
 
+- Continue the current campaign: `/lab-auto: 继续`.
+- Continue the current campaign at the highest execution level: `/lab-auto: L3，继续`.
 - `L1` is the safe validation level. Use it for a smoke run, one bounded real run, or a simple review/report refresh when you do not want automatic iteration.
 - `L2` is the default recommended level. Use it for bounded experiment iteration inside a frozen core when you want auto to keep running until a gate, stop condition, or terminal goal is hit.
-- `L3` is the aggressive campaign level. Use it only when you explicitly want broad exploration, larger search space changes, and optional manuscript-writing work.
-- If you are unsure, choose `L2`.
-- If the request omits the level or mixes it with a paper layer, phase, or table target, `/lab auto` should stop and ask for an explicit autonomy level before arming the loop.
+- `L3` is the continue-until-boundary level. Use it when you want auto to keep pushing inside the approved envelope without stopping for routine implementation-path choices.
+- If the request omits the level, default to `L2`.
+- Accept short level forms such as `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts like `最高级别` or `最高自治` for `L3`.
+- If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+- Only stop and ask when a level reference is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+- An explicit `/lab auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+- Approved `L2` and `L3` execution campaigns default to execution mode instead of reopening brainstorming or spec-review loops.
+- In execution mode, do not route routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks into brainstorming, spec review, or reviewer or explorer subagents.
+- Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 - The first visible output of a real `/lab auto` run must be `Auto preflight`.
 - That first visible output must show files read plus `Autonomy level`, `Allowed stages`, `Terminal goal`, `Primary gate`, and `Secondary guard`.
 - If the preflight block cannot be completed from `.lab/context/eval-protocol.md`, `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, and `.lab/context/auto-outcome.md`, `/lab auto` should stop instead of acting like the loop is armed.
@@ -95,10 +103,12 @@ Treat all of these as equivalent stage requests:
 - Treat `paper layer`, `phase`, and `table` as experiment targets. For example, `paper layer 3` or `Phase 1` should not be interpreted as `Autonomy level L3`.
 - A good `/lab auto` request should name:
   - the objective
-  - the autonomy level
+  - the autonomy level when you want to override the default or current campaign
   - the terminal goal
   - the scope or phase to advance
   - the allowed modifications
 - If the repository workflow language is Chinese, summaries, checklist items, task labels, and progress updates should be written in Chinese unless a code identifier or file path must stay literal.
-- Good example:
-  - `/lab auto Autonomy level L2. Objective: advance paper layer 3 through one bounded protocol improvement. Terminal goal: task-completion. Scope: bounded protocol, tests, one minimal implementation, and one small run. Allowed modifications: configuration, evaluation script, and data-loading logic only.`
+- Good examples:
+  - `/lab-auto: 继续`
+  - `/lab-auto: L3，继续`
+  - `/lab auto Objective: advance paper layer 3 through one bounded protocol improvement. Terminal goal: task-completion. Scope: bounded protocol, tests, one minimal implementation, and one small run. Allowed modifications: configuration, evaluation script, and data-loading logic only.`

package/package-assets/claude/commands/lab:auto.md

@@ -6,8 +6,19 @@ argument-hint: autonomous campaign target
 
 Use the installed `lab` skill at `.claude/skills/lab/SKILL.md`.
 
+Quick continue: `/lab-auto: 继续`.
+Quick continue at the highest execution level: `/lab-auto: L3，继续`.
+
 Execute the requested `/lab-auto` command against the user's argument now. Do not only recommend another lab stage. If a blocking prerequisite is missing, say exactly what is missing and ask at most one clarifying question.
 First normalize the user's request into CLI-ready auto contract fields: objective, autonomy level, campaign kind, allowed stages, and any explicit terminal-goal hints that can be inferred without changing scope.
+If the user omits the level, default to `L2`. Accept short forms like `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts such as `最高级别` or `最高自治` for `L3`.
+If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+To continue the current campaign with the stored runtime level, say `/lab-auto: 继续`.
+Only stop for a level clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+An explicit `/lab auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+Approved `L2` and `L3` execution campaigns default to execution mode.
+In execution mode, do not enter brainstorming, do not enter spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 Your first execution action must be to run `superlab auto start` against the current project instead of directly rewriting `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, or `.lab/context/auto-outcome.md` yourself.
 Pass the normalized fields through CLI flags, including `--objective`, `--campaign-kind`, `--allowed-stages`, and `--autonomy-level` when the user specified or implied them.
 After that CLI handoff, treat the CLI/runtime result as the source of truth. If it reports rollover, conflict, missing contract fields, or a started campaign, reflect that result back to the user instead of bypassing it with prompt-side edits.

package/package-assets/claude/commands/lab/357/274/232auto.md

@@ -6,8 +6,19 @@ argument-hint: autonomous campaign target
 
 Use the installed `lab` skill at `.claude/skills/lab/SKILL.md`.
 
+Quick continue: `/lab-auto: 继续`.
+Quick continue at the highest execution level: `/lab-auto: L3，继续`.
+
 Execute the requested `/lab-auto` command against the user's argument now. Do not only recommend another lab stage. If a blocking prerequisite is missing, say exactly what is missing and ask at most one clarifying question.
 First normalize the user's request into CLI-ready auto contract fields: objective, autonomy level, campaign kind, allowed stages, and any explicit terminal-goal hints that can be inferred without changing scope.
+If the user omits the level, default to `L2`. Accept short forms like `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts such as `最高级别` or `最高自治` for `L3`.
+If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+To continue the current campaign with the stored runtime level, say `/lab-auto: 继续`.
+Only stop for a level clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+An explicit `/lab auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+Approved `L2` and `L3` execution campaigns default to execution mode.
+In execution mode, do not enter brainstorming, do not enter spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 Your first execution action must be to run `superlab auto start` against the current project instead of directly rewriting `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, or `.lab/context/auto-outcome.md` yourself.
 Pass the normalized fields through CLI flags, including `--objective`, `--campaign-kind`, `--allowed-stages`, and `--autonomy-level` when the user specified or implied them.
 After that CLI handoff, treat the CLI/runtime result as the source of truth. If it reports rollover, conflict, missing contract fields, or a started campaign, reflect that result back to the user instead of bypassing it with prompt-side edits.

package/package-assets/codex/prompts/lab/auto.md

@@ -5,8 +5,19 @@ argument-hint: autonomous campaign target
 
 Use the installed `lab` skill at `.codex/skills/lab/SKILL.md`.
 
+Quick continue: `/lab-auto: 继续`.
+Quick continue at the highest execution level: `/lab-auto: L3，继续`.
+
 Execute the requested `/lab:auto` stage against the user's argument now. Do not only recommend another lab stage. If a blocking prerequisite is missing, say exactly what is missing and ask at most one clarifying question.
 First normalize the user's request into CLI-ready auto contract fields: objective, autonomy level, campaign kind, allowed stages, and any explicit terminal-goal hints that can be inferred without changing scope.
+If the user omits the level, default to `L2`. Accept short forms like `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts such as `最高级别` or `最高自治` for `L3`.
+If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+To continue the current campaign with the stored runtime level, say `/lab-auto: 继续`.
+Only stop for a level clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+An explicit `/lab:auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+Approved `L2` and `L3` execution campaigns default to execution mode.
+In execution mode, do not enter brainstorming, do not enter spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 Your first execution action must be to run `superlab auto start` against the current project instead of directly rewriting `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, or `.lab/context/auto-outcome.md` yourself.
 Pass the normalized fields through CLI flags, including `--objective`, `--campaign-kind`, `--allowed-stages`, and `--autonomy-level` when the user specified or implied them.
 After that CLI handoff, treat the CLI/runtime result as the source of truth. If it reports rollover, conflict, missing contract fields, or a started campaign, reflect that result back to the user instead of bypassing it with prompt-side edits.

package/package-assets/codex/prompts/lab-auto.md

@@ -5,8 +5,19 @@ argument-hint: autonomous campaign target
 
 Use the installed `lab` skill at `.codex/skills/lab/SKILL.md`.
 
+Quick continue: `/lab-auto: 继续`.
+Quick continue at the highest execution level: `/lab-auto: L3，继续`.
+
 Execute the requested `/lab:auto` stage against the user's argument now. Do not only recommend another lab stage. If a blocking prerequisite is missing, say exactly what is missing and ask at most one clarifying question.
 First normalize the user's request into CLI-ready auto contract fields: objective, autonomy level, campaign kind, allowed stages, and any explicit terminal-goal hints that can be inferred without changing scope.
+If the user omits the level, default to `L2`. Accept short forms like `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts such as `最高级别` or `最高自治` for `L3`.
+If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+To continue the current campaign with the stored runtime level, say `/lab-auto: 继续`.
+Only stop for a level clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+An explicit `/lab:auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+Approved `L2` and `L3` execution campaigns default to execution mode.
+In execution mode, do not enter brainstorming, do not enter spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 Your first execution action must be to run `superlab auto start` against the current project instead of directly rewriting `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, or `.lab/context/auto-outcome.md` yourself.
 Pass the normalized fields through CLI flags, including `--objective`, `--campaign-kind`, `--allowed-stages`, and `--autonomy-level` when the user specified or implied them.
 After that CLI handoff, treat the CLI/runtime result as the source of truth. If it reports rollover, conflict, missing contract fields, or a started campaign, reflect that result back to the user instead of bypassing it with prompt-side edits.

package/package-assets/codex/prompts/lab.md

@@ -75,11 +75,19 @@ Treat all of these as equivalent stage requests:
 
 ## Level Guide for `/lab:auto`
 
+- Continue the current campaign: `/lab-auto: 继续`.
+- Continue the current campaign at the highest execution level: `/lab-auto: L3，继续`.
 - `L1` is the safe validation level. Use it for a smoke run, one bounded real run, or a simple review/report refresh when you do not want automatic iteration.
 - `L2` is the default recommended level. Use it for bounded experiment iteration inside a frozen core when you want auto to keep running until a gate, stop condition, or terminal goal is hit.
-- `L3` is the aggressive campaign level. Use it only when you explicitly want broad exploration, larger search space changes, and optional manuscript-writing work.
-- If you are unsure, choose `L2`.
-- If the request omits the level or mixes it with a paper layer, phase, or table target, `/lab:auto` should stop and ask for an explicit autonomy level before arming the loop.
+- `L3` is the continue-until-boundary level. Use it when you want auto to keep pushing inside the approved envelope without stopping for routine implementation-path choices.
+- If the request omits the level, default to `L2`.
+- Accept short level forms such as `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts like `最高级别` or `最高自治` for `L3`.
+- If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+- Only stop and ask when a level reference is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+- An explicit `/lab:auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+- Approved `L2` and `L3` execution campaigns default to execution mode instead of reopening brainstorming or spec-review loops.
+- In execution mode, do not route routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks into brainstorming, spec review, or reviewer or explorer subagents.
+- Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 - The first visible output of a real `/lab:auto` run must be `Auto preflight`.
 - That first visible output must show files read plus `Autonomy level`, `Allowed stages`, `Terminal goal`, `Primary gate`, and `Secondary guard`.
 - If the preflight block cannot be completed from `.lab/context/eval-protocol.md`, `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, and `.lab/context/auto-outcome.md`, `/lab:auto` should stop instead of acting like the loop is armed.
@@ -89,10 +97,12 @@ Treat all of these as equivalent stage requests:
 - Treat `paper layer`, `phase`, and `table` as experiment targets. For example, `paper layer 3` or `Phase 1` should not be interpreted as `Autonomy level L3`.
 - A good `/lab:auto` request should name:
   - the objective
-  - the autonomy level
+  - the autonomy level when you want to override the default or current campaign
   - the terminal goal
   - the scope or phase to advance
   - the allowed modifications
 - If the repository workflow language is Chinese, summaries, checklist items, task labels, and progress updates should be written in Chinese unless a code identifier or file path must stay literal.
-- Good example:
-  - `/lab:auto Autonomy level L2. Objective: advance paper layer 3 through one bounded protocol improvement. Terminal goal: task-completion. Scope: bounded protocol, tests, one minimal implementation, and one small run. Allowed modifications: configuration, evaluation script, and data-loading logic only.`
+- Good examples:
+  - `/lab:auto: 继续`
+  - `/lab:auto: L3，继续`
+  - `/lab:auto Objective: advance paper layer 3 through one bounded protocol improvement. Terminal goal: task-completion. Scope: bounded protocol, tests, one minimal implementation, and one small run. Allowed modifications: configuration, evaluation script, and data-loading logic only.`

package/package-assets/codex/prompts/lab:auto.md

@@ -5,8 +5,19 @@ argument-hint: autonomous campaign target
 
 Use the installed `lab` skill at `.codex/skills/lab/SKILL.md`.
 
+Quick continue: `/lab-auto: 继续`.
+Quick continue at the highest execution level: `/lab-auto: L3，继续`.
+
 Execute the requested `/lab:auto` stage against the user's argument now. Do not only recommend another lab stage. If a blocking prerequisite is missing, say exactly what is missing and ask at most one clarifying question.
 First normalize the user's request into CLI-ready auto contract fields: objective, autonomy level, campaign kind, allowed stages, and any explicit terminal-goal hints that can be inferred without changing scope.
+If the user omits the level, default to `L2`. Accept short forms like `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts such as `最高级别` or `最高自治` for `L3`.
+If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+To continue the current campaign with the stored runtime level, say `/lab-auto: 继续`.
+Only stop for a level clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+An explicit `/lab:auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+Approved `L2` and `L3` execution campaigns default to execution mode.
+In execution mode, do not enter brainstorming, do not enter spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 Your first execution action must be to run `superlab auto start` against the current project instead of directly rewriting `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, or `.lab/context/auto-outcome.md` yourself.
 Pass the normalized fields through CLI flags, including `--objective`, `--campaign-kind`, `--allowed-stages`, and `--autonomy-level` when the user specified or implied them.
 After that CLI handoff, treat the CLI/runtime result as the source of truth. If it reports rollover, conflict, missing contract fields, or a started campaign, reflect that result back to the user instead of bypassing it with prompt-side edits.

package/package-assets/codex/prompts/lab/357/274/232auto.md

@@ -5,8 +5,19 @@ argument-hint: autonomous campaign target
 
 Use the installed `lab` skill at `.codex/skills/lab/SKILL.md`.
 
+Quick continue: `/lab-auto: 继续`.
+Quick continue at the highest execution level: `/lab-auto: L3，继续`.
+
 Execute the requested `/lab:auto` stage against the user's argument now. Do not only recommend another lab stage. If a blocking prerequisite is missing, say exactly what is missing and ask at most one clarifying question.
 First normalize the user's request into CLI-ready auto contract fields: objective, autonomy level, campaign kind, allowed stages, and any explicit terminal-goal hints that can be inferred without changing scope.
+If the user omits the level, default to `L2`. Accept short forms like `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts such as `最高级别` or `最高自治` for `L3`.
+If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+To continue the current campaign with the stored runtime level, say `/lab-auto: 继续`.
+Only stop for a level clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+An explicit `/lab:auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+Approved `L2` and `L3` execution campaigns default to execution mode.
+In execution mode, do not enter brainstorming, do not enter spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 Your first execution action must be to run `superlab auto start` against the current project instead of directly rewriting `.lab/context/auto-mode.md`, `.lab/context/auto-status.md`, `.lab/context/auto-ledger.md`, or `.lab/context/auto-outcome.md` yourself.
 Pass the normalized fields through CLI flags, including `--objective`, `--campaign-kind`, `--allowed-stages`, and `--autonomy-level` when the user specified or implied them.
 After that CLI handoff, treat the CLI/runtime result as the source of truth. If it reports rollover, conflict, missing contract fields, or a started campaign, reflect that result back to the user instead of bypassing it with prompt-side edits.

package/package-assets/shared/skills/lab/SKILL.md

@@ -115,8 +115,17 @@ Use this skill when the user invokes `/lab:*` or asks for the structured researc
 - Start every `/lab:auto` run with a visible `Auto preflight` summary that reports files read plus `Autonomy level`, `Approval status`, `Allowed stages`, `Terminal goal`, `Primary gate`, and `Secondary guard`.
 - If any required preflight field is missing or inconsistent, stop before any loop action. Do not present a fake auto summary as if the loop were armed.
 - Keep `.lab/context/auto-ledger.md` updated with the active owner, observed state, and resume boundary while the loop is live.
-- Treat `L1` as safe-run validation, `L2` as bounded iteration, and `L3` as aggressive campaign mode.
-- Surface the level guide every time `/lab:auto` starts, and make the detailed guide mandatory when the user omits the level or mixes it with a paper layer, phase, or table target.
+- Treat `L1` as safe-run validation, `L2` as bounded iteration, and `L3` as continue-until-boundary execution inside the approved envelope.
+- Surface the level guide every time `/lab:auto` starts.
+- If the user omits the level, default to `L2`.
+- Accept short forms such as `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts like `最高级别` or `最高自治` for `L3`.
+- If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level.
+- Only require a clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+- Treat an explicit `/lab:auto` or `/lab-auto` request as higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+- Treat approved `L2` and `L3` campaigns as execution mode by default.
+- In execution mode, do not reopen brainstorming, do not reopen spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+- Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
+- If an explicit `/lab:auto` request stays inside the approved execution envelope, keep it in the auto execution path even when it mentions design-like tasks such as feature selection, baseline selection, discretization, or candidate sweeps.
 - Reuse `/lab:run`, `/lab:iterate`, `/lab:review`, `/lab:report`, and optional `/lab:write` instead of inventing a second workflow.
 - Do not automatically change the research mission, paper-facing framing, or core claims.
 - You may add exploratory datasets, benchmarks, and comparison methods inside the approved exploration envelope.

package/package-assets/shared/skills/lab/stages/auto.md

@@ -66,7 +66,11 @@
 - Recommended level meanings:
   - `L1`: safe run validation over `run`, `review`, and `report`
   - `L2`: bounded iteration over `run`, `iterate`, `review`, and `report`
-  - `L3`: aggressive campaign that may also include `write`
+- `L3`: continue-until-boundary execution over the approved envelope, and it may also include `write` when framing is already approved
+- An explicit `/lab:auto` or `/lab-auto` request has higher priority than broader creative-work or review skills such as brainstorming or spec-review.
+- Approved `L2` and `L3` execution campaigns default to execution mode.
+- In execution mode, do not reopen brainstorming, do not reopen spec review, and do not spawn reviewer, explorer, or other subagent loops for routine implementation-path choices, helper scripts, path fixes, dataset adaptation, same-family candidate sweeps, or ordinary self-checks.
+- Only switch from execution mode into design or reviewer mode when the user explicitly asks for design or review help, the contract fit requires a new campaign, or the contract's escalation condition explicitly requires independent review.
 - Default allowed stages are `run`, `iterate`, `review`, and `report`. Only include `write` when framing is already approved and manuscript drafting is within scope.
 - Do not automatically change the research mission, paper-facing framing, or core claims.
 - You may add exploratory datasets, benchmarks, and comparison methods inside the exploration envelope.
@@ -127,9 +131,11 @@
 
 ## Level Guide
 
+- Continue the current campaign: `/lab-auto: 继续`.
+- Continue the current campaign at the highest execution level: `/lab-auto: L3，继续`.
 - `L1` = safe validation
 - `L2` = default recommended bounded iteration
-- `L3` = aggressive campaign
+- `L3` = continue-until-boundary execution inside the approved envelope
 - If you are unsure, choose `L2`.
 
 - Start with a concise summary of the objective, the frozen core, and the next automatic stage.
@@ -144,15 +150,14 @@
 - Normalize ambiguous user requests before arming the loop.
 - Treat `Autonomy level L1/L2/L3` as execution privilege only.
 - Treat `Layer`, `Phase`, and `Table` references as paper-structure or experiment-scope targets, not as autonomy levels.
-- If the user does not name an autonomy level, or mixes it with a paper layer, phase, or table target, stop and deliver a detailed level guide before execution. That detailed guide should explain:
-  - the typical use case for `L1`, `L2`, and `L3`
-  - what kinds of modifications each level allows
-  - what kind of stop boundary each level is meant for
-  - that `L2` is the default recommendation when the user is unsure
-- Ask for one explicit level choice before arming the loop after that detailed guide.
+- If the user does not name an autonomy level, default to `L2` instead of stopping for a level-only clarification.
+- Accept short level forms such as `L1`, `L2`, `L3`, lowercase `l1/l2/l3`, and explicit Chinese shortcuts like `最高级别` or `最高自治` for `L3`.
+- If the user says `继续` and there is an active or resumable campaign, reuse the current campaign level instead of asking them to repeat it.
+- Only stop for a level clarification when the level reference itself is genuinely ambiguous, such as `第三层`, `phase 3`, or `table 3`.
+- If the user explicitly invokes `/lab:auto` or `/lab-auto`, keep the request in the auto execution path even when the objective sounds design-heavy, such as feature selection, baseline selection, discretization, or candidate sweeps, so long as the requested work stays inside the approved execution envelope.
 - Example:
   - `Layer 3` means a paper layer or experiment target.
-  - `Autonomy level L3` means the aggressive campaign permission envelope.
+  - `Autonomy level L3` means the continue-until-boundary permission envelope.
 - If the user mixes framework work and experiment work in one request, restate a normalized contract with:
   - objective
   - autonomy level

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "superlab",
-  "version": "0.1.47",
+  "version": "0.1.48",
   "description": "Strict /lab research workflow installer for Codex and Claude",
   "keywords": [
     "codex",