superlab 0.1.24 → 0.1.26

package/README.md

@@ -180,7 +180,7 @@ superlab auto stop
 
 - `run` and `iterate` must change persistent outputs under `results_root`
 - `review` must update canonical review context
-- `report` must write `<deliverables_root>/report.md` and `<deliverables_root>/main-tables.md`
+- `report` must write `<deliverables_root>/report.md`, `<deliverables_root>/main-tables.md`, and `<deliverables_root>/artifact-status.md`
 - `write` must produce LaTeX output under `<deliverables_root>/paper/`
 - a successful promotion must write back into `.lab/context/data-decisions.md`, `.lab/context/decisions.md`, `.lab/context/state.md`, and `.lab/context/session-brief.md`
 - every run must end with `.lab/context/auto-outcome.md`, including why it stopped, whether the terminal goal was reached, and which artifact is the final outcome
@@ -255,7 +255,7 @@ Stages should follow that file rather than guess language locally.
 - Python 3.10+
 - Git
 
-`/lab:write` ships with vendored paper-writing references and the upstream example bank for `abstract`, `introduction`, and `method` under the installed `lab` skill, so it does not depend on an extra runtime skill installation.
+`/lab:write` ships with vendored paper-writing references, the upstream example bank for `abstract`, `introduction`, and `method`, and local manuscript-delivery examples for `related work`, `experiments`, and `conclusion` under the installed `lab` skill, so it does not depend on an extra runtime skill installation.
 
 ## Command Set
 
@@ -309,8 +309,12 @@ See the source command docs in [commands/codex/lab.md](/Users/zhouhao119/coding/
 
 - `docs/research/report.md`
 - `docs/research/main-tables.md`
+- `docs/research/artifact-status.md`
 - `docs/research/paper/main.tex`
+- `docs/research/paper/references.bib`
 - `docs/research/paper/sections/*.tex`
+- `docs/research/paper/tables/*.tex`
+- `docs/research/paper/figures/*.tex`
 
 Internal writing-control artifacts stay under:
 
@@ -322,6 +326,7 @@ If `paper_template_root` is configured, `/lab:write` should inspect that templat
 If no template is configured, the first manuscript-writing round should ask once whether to continue with the managed default LaTeX scaffold or attach a template directory first.
 If the user approves the default scaffold, persist that choice in `.lab/config/workflow.json` and stop asking on ordinary rounds.
 At the final export or final-draft boundary, if the project is still on the default scaffold and no attached template exists, ask one final reminder question before finalizing.
+For final-draft or export rounds, `/lab:write` should materialize real LaTeX tables, figure placeholders with figure intent, a non-empty `references.bib`, and pass `.lab/.managed/scripts/validate_manuscript_delivery.py --paper-dir <deliverables_root>/paper` before stopping.
 
 `/lab` treats `.lab/` as the workflow-control layer only. Durable outputs should use natural project roots:
 
@@ -335,6 +340,7 @@ At the final export or final-draft boundary, if the project is still on the defa
 - `scripts/eval_report.py` normalizes metrics into a comparable summary.
 - `scripts/summarize_iterations.py` rolls multiple iteration summaries into one file.
 - `scripts/validate_results.py` verifies required report fields before publication.
+- `scripts/validate_manuscript_delivery.py` verifies that the paper deliverable contains basic manuscript-ready tables, figures, citations, and bibliography.
 
 ## Packaging
 

package/README.zh-CN.md

@@ -178,7 +178,7 @@ superlab auto stop
 
 - `run` 和 `iterate` 必须更新 `results_root` 下的持久输出
 - `review` 必须更新规范的审查上下文
-- `report` 必须写出 `<deliverables_root>/report.md` 和 `<deliverables_root>/main-tables.md`
+- `report` 必须写出 `<deliverables_root>/report.md`、`<deliverables_root>/main-tables.md` 和 `<deliverables_root>/artifact-status.md`
 - `write` 必须写出 `<deliverables_root>/paper/` 下的 LaTeX 论文产物
 - promotion 成功后必须写回 `.lab/context/data-decisions.md`、`.lab/context/decisions.md`、`.lab/context/state.md` 和 `.lab/context/session-brief.md`
 - 每次运行都必须写出 `.lab/context/auto-outcome.md`，记录为什么停止、是否达到终止目标，以及哪一个工件是最终结果
@@ -253,7 +253,7 @@ superlab init --lang en
 - Python 3.10+
 - Git
 
-`/lab:write` 自带 vendored 的 paper-writing 章节参考，以及 `abstract`、`introduction`、`method` 对应的 upstream example bank，不再依赖额外安装一个运行时写作 skill。
+`/lab:write` 自带 vendored 的 paper-writing 章节参考、`abstract`、`introduction`、`method` 对应的 upstream example bank，以及为 `related work`、`experiments`、`conclusion` 补齐的 manuscript-delivery examples，不再依赖额外安装一个运行时写作 skill。
 
 ## 命令集合
 
@@ -294,8 +294,12 @@ Codex 和 Claude 的命令入口不一样：
 
 - `docs/research/report.md`
 - `docs/research/main-tables.md`
+- `docs/research/artifact-status.md`
 - `docs/research/paper/main.tex`
+- `docs/research/paper/references.bib`
 - `docs/research/paper/sections/*.tex`
+- `docs/research/paper/tables/*.tex`
+- `docs/research/paper/figures/*.tex`
 
 内部写作控制工件放在：
 
@@ -307,6 +311,7 @@ Codex 和 Claude 的命令入口不一样：
 如果没有配置模板，第一次进入论文 `.tex` 写作时应先追问一次：继续使用内置默认 LaTeX scaffold，还是先接入模板目录。
 如果用户确认先用默认 scaffold，就把这个决定持久化到 `.lab/config/workflow.json`，后续普通轮次不再重复追问。
 但在最终导出或最终定稿节点，如果项目仍在使用默认 scaffold 且没有接入模板，应再提醒一次，给用户最后切换模板的机会。
+在最终定稿或导出轮次里，`/lab:write` 还应物化真正的 LaTeX 表格、带图意图的 figure placeholders、非空的 `references.bib`，并在停止前通过 `.lab/.managed/scripts/validate_manuscript_delivery.py --paper-dir <deliverables_root>/paper`。
 
 `/lab` 把 `.lab/` 视为工作流控制层，不是正式结果目录。持久输出应按自然根目录放置：
 
@@ -320,7 +325,7 @@ Codex 和 Claude 的命令入口不一样：
 - `package-assets/` 是 npm 安装时真正分发到目标项目的命令与 skill 资产。
 - `skills/lab/` 是仓库开发时使用的共享 workflow 定义。
 - `templates/` 是研究工件模板。
-- `scripts/` 是运行登记、评估汇总和结果校验脚本。
+- `scripts/` 是运行登记、评估汇总和结果校验脚本，其中也包括论文交付校验器。
 - `results/` 和 `figures/` 是默认的自然结果目录。
 - `tests/` 是脚本与安装器测试。
 - `examples/` 是最小端到端示例。

package/lib/auto_contracts.cjs

@@ -34,6 +34,7 @@ const FROZEN_CORE_ALIASES = {
 const REVIEW_CONTEXT_FILES = [
   path.join(".lab", "context", "decisions.md"),
   path.join(".lab", "context", "state.md"),
+  path.join(".lab", "context", "workflow-state.md"),
   path.join(".lab", "context", "open-questions.md"),
   path.join(".lab", "context", "evidence-index.md"),
 ];
@@ -288,6 +289,7 @@ function stageContractSnapshot(targetDir, stage) {
     report: [
       path.join(deliverablesRoot, "report.md"),
       path.join(deliverablesRoot, "main-tables.md"),
+      path.join(deliverablesRoot, "artifact-status.md"),
     ],
     write: [
       path.join(deliverablesRoot, "paper", "main.tex"),
@@ -318,7 +320,7 @@ function verifyStageContract({ stage, snapshot }) {
   if (stage === "review") {
     if (changedPaths.length === 0) {
       throw new Error(
-        "review stage did not update canonical review context (.lab/context/decisions.md, state.md, open-questions.md, or evidence-index.md)"
+        "review stage did not update canonical review context (.lab/context/decisions.md, state.md, workflow-state.md, open-questions.md, or evidence-index.md)"
       );
     }
     return;
@@ -327,7 +329,7 @@ function verifyStageContract({ stage, snapshot }) {
   if (stage === "report") {
     const missing = Array.from(snapshot.keys()).filter((absolutePath) => !changedPaths.includes(absolutePath));
     if (missing.length > 0) {
-      throw new Error("report stage did not produce the deliverable report.md and main-tables.md under deliverables_root");
+      throw new Error("report stage did not produce report.md, main-tables.md, and artifact-status.md under deliverables_root");
     }
     return;
   }

package/lib/context.cjs

@@ -206,6 +206,30 @@ function isMeaningful(value) {
   return !PLACEHOLDER_VALUES.has((value || "").trim().toLowerCase());
 }
 
+function hasWorkflowStateShape(text) {
+  if (!text) {
+    return false;
+  }
+  return [
+    extractValue(text, ["Active stage", "当前阶段", "Stage"]),
+    extractValue(text, ["Current objective", "当前目标"]),
+    extractValue(text, ["Next required output", "Next required artifact", "下一项必要输出"]),
+    extractValue(text, ["Immediate action", "立即要做的动作"]),
+  ].some((value) => isMeaningful(value));
+}
+
+function readWorkflowStateContext(targetDir) {
+  const workflowState = readFileIfExists(contextFile(targetDir, "workflow-state.md"));
+  if (hasWorkflowStateShape(workflowState)) {
+    return workflowState;
+  }
+  const legacyState = readFileIfExists(contextFile(targetDir, "state.md"));
+  if (hasWorkflowStateShape(legacyState)) {
+    return legacyState;
+  }
+  return workflowState || legacyState;
+}
+
 function readWorkflowConfig(targetDir) {
   const configPath = path.join(targetDir, ".lab", "config", "workflow.json");
   if (!fs.existsSync(configPath)) {
@@ -232,6 +256,7 @@ function getCollaboratorDeliverablePaths(targetDir) {
     deliverablesRoot,
     reportPath: path.join(deliverablesRoot, "report.md"),
     mainTablesPath: path.join(deliverablesRoot, "main-tables.md"),
+    artifactStatusPath: path.join(deliverablesRoot, "artifact-status.md"),
   };
 }
 
@@ -610,11 +635,16 @@ function labelValue(text, englishLabels, chineseLabels = []) {
 
 function collectHydrationSources(targetDir) {
   const { reportPath, mainTablesPath } = getCollaboratorDeliverablePaths(targetDir);
+  const workflowStatePath = hasWorkflowStateShape(readFileIfExists(contextFile(targetDir, "workflow-state.md")))
+    ? ".lab/context/workflow-state.md"
+    : hasWorkflowStateShape(readFileIfExists(contextFile(targetDir, "state.md")))
+      ? ".lab/context/state.md"
+      : "";
   return [
     fs.existsSync(reportPath) ? path.relative(targetDir, reportPath) : "",
     fs.existsSync(mainTablesPath) ? path.relative(targetDir, mainTablesPath) : "",
     readFileIfExists(contextFile(targetDir, "data-decisions.md")) ? ".lab/context/data-decisions.md" : "",
-    readFileIfExists(contextFile(targetDir, "state.md")) ? ".lab/context/state.md" : "",
+    workflowStatePath,
     readFileIfExists(contextFile(targetDir, "evidence-index.md")) ? ".lab/context/evidence-index.md" : "",
   ].filter(Boolean);
 }
@@ -626,7 +656,7 @@ function hydrateMissionContext(targetDir) {
 
   const lang = readWorkflowLanguage(targetDir);
   const missionText = readFileIfExists(contextFile(targetDir, "mission.md"));
-  const stateText = readFileIfExists(contextFile(targetDir, "state.md"));
+  const workflowStateText = readWorkflowStateContext(targetDir);
   const evidenceText = readFileIfExists(contextFile(targetDir, "evidence-index.md"));
   const dataDecisions = readFileIfExists(contextFile(targetDir, "data-decisions.md"));
   const reportText = readFileIfExists(getCollaboratorDeliverablePaths(targetDir).reportPath);
@@ -637,7 +667,7 @@ function hydrateMissionContext(targetDir) {
     problem: mergePreferred(
       extractValue(missionText, ["One-sentence problem", "一句话问题"]),
       extractReportValue(reportText, "problem"),
-      extractValue(stateText, ["Current objective", "当前目标", "Current objective"])
+      extractValue(workflowStateText, ["Current objective", "当前目标", "Current objective"])
     ),
     whyItMatters: mergePreferred(
       extractValue(missionText, ["Why it matters", "为什么重要"]),
@@ -680,7 +710,7 @@ function hydrateMissionContext(targetDir) {
     currentOwner: extractValue(missionText, ["Current owner or session", "当前 owner 或会话"]),
     latestStage: mergePreferred(
       extractValue(missionText, ["Latest stage to update this mission", "最近一次允许更新 mission 的阶段"]),
-      extractValue(stateText, ["Active stage", "当前阶段", "Stage"])
+      extractValue(workflowStateText, ["Active stage", "当前阶段", "Stage"])
     ),
   };
 
@@ -935,6 +965,110 @@ function hydrateCanonicalContext(targetDir) {
   };
 }
 
+function renderResearchState(lang, data) {
+  if (lang === "zh") {
+    return `# 研究状态
+
+## 已批准方向
+
+- One-sentence problem: ${data.problem || "待补充"}
+- Approved direction: ${data.direction || "待补充"}
+- Strongest supported claim: ${data.claim || "待补充"}
+
+## 证据边界
+
+- What the current evidence really supports: ${data.reportModeReason || data.direction || "待补充"}
+- What is still outside the boundary: ${data.question || "待补充"}
+- Biggest research risk: ${data.risk || "待补充"}
+
+## 当前研究主线
+
+- Current research focus: ${data.immediateAction || data.direction || "待补充"}
+- Primary metric: ${data.evalPrimaryMetrics || data.threshold || "待补充"}
+- Dataset or benchmark scope: ${data.datasetPackage || data.benchmarkRole || "待补充"}
+
+## 当前研究约束
+
+- Hard constraints: ${data.boundary || "待补充"}
+- Claim boundary: ${data.evalClaimBoundary || "待补充"}
+- Conditions that require reopening the direction: ${data.humanDecision || "待补充"}
+`;
+  }
+
+  return `# Research State
+
+## Approved Direction
+
+- One-sentence problem: ${data.problem || "TBD"}
+- Approved direction: ${data.direction || "TBD"}
+- Strongest supported claim: ${data.claim || "TBD"}
+
+## Evidence Boundary
+
+- What the current evidence really supports: ${data.reportModeReason || data.direction || "TBD"}
+- What is still outside the boundary: ${data.question || "TBD"}
+- Biggest research risk: ${data.risk || "TBD"}
+
+## Active Research Track
+
+- Current research focus: ${data.immediateAction || data.direction || "TBD"}
+- Primary metric: ${data.evalPrimaryMetrics || data.threshold || "TBD"}
+- Dataset or benchmark scope: ${data.datasetPackage || data.benchmarkRole || "TBD"}
+
+## Current Research Constraints
+
+- Hard constraints: ${data.boundary || "TBD"}
+- Claim boundary: ${data.evalClaimBoundary || "TBD"}
+- Conditions that require reopening the direction: ${data.humanDecision || "TBD"}
+`;
+}
+
+function renderWorkflowState(lang, data) {
+  if (lang === "zh") {
+    return `# 工作流状态
+
+## 当前阶段
+
+- Active stage: ${data.stage || "待补充"}
+- 当前目标：${data.workflowObjective || "待补充"}
+- 下一项必要输出：${data.nextArtifact || "待补充"}
+
+## 最近更新
+
+- 最近完成动作：${data.latestAction || "待补充"}
+- 最新工件路径：${data.latestArtifactPath || "待补充"}
+- 最新 run 或 report id：${data.latestRunOrReportId || "待补充"}
+
+## 下一步
+
+- 立即要做的动作：${data.immediateAction || "待补充"}
+- 当前阻塞：${data.blocker || "待补充"}
+- 是否需要人工决策：${data.humanDecision || "待补充"}
+`;
+  }
+
+  return `# Workflow State
+
+## Current Stage
+
+- Active stage: ${data.stage || "TBD"}
+- Current objective: ${data.workflowObjective || "TBD"}
+- Next required output: ${data.nextArtifact || "TBD"}
+
+## Latest Update
+
+- Last completed action: ${data.latestAction || "TBD"}
+- Latest artifact path: ${data.latestArtifactPath || "TBD"}
+- Latest run or report id: ${data.latestRunOrReportId || "TBD"}
+
+## Next Step
+
+- Immediate next action: ${data.immediateAction || "TBD"}
+- Blocking issue: ${data.blocker || "TBD"}
+- Human decision needed: ${data.humanDecision || "TBD"}
+`;
+}
+
 function renderSummary(lang, data) {
   if (lang === "zh") {
     return `# 研究摘要
@@ -1220,7 +1354,7 @@ ${data.problem || "待补充"}
 ## 先读这些文件
 
 1. \`.lab/context/mission.md\`
-2. \`.lab/context/state.md\`
+2. \`.lab/context/workflow-state.md\`
 3. \`.lab/context/evidence-index.md\`
 
 ## 不要静默修改
@@ -1309,7 +1443,7 @@ ${data.problem || "TBD"}
 ## Read First
 
 1. \`.lab/context/mission.md\`
-2. \`.lab/context/state.md\`
+2. \`.lab/context/workflow-state.md\`
 3. \`.lab/context/evidence-index.md\`
 
 ## Do Not Change Silently
@@ -1322,7 +1456,7 @@ ${data.problem || "TBD"}
 function buildContextSnapshot(targetDir) {
   const reportStatus = getCollaboratorReportStatus(targetDir);
   const mission = readFileIfExists(contextFile(targetDir, "mission.md"));
-  const state = readFileIfExists(contextFile(targetDir, "state.md"));
+  const workflowState = readWorkflowStateContext(targetDir);
   const evidence = readFileIfExists(contextFile(targetDir, "evidence-index.md"));
   const questions = readFileIfExists(contextFile(targetDir, "open-questions.md"));
   const dataDecisions = readFileIfExists(contextFile(targetDir, "data-decisions.md"));
@@ -1390,16 +1524,22 @@ function buildContextSnapshot(targetDir) {
   return {
     problem: extractValue(mission, ["One-sentence problem", "一句话问题"]),
     direction: extractValue(mission, ["Approved direction", "已批准方向"]),
-    stage: extractValue(state, ["Active stage", "当前阶段", "Stage"]),
-    nextArtifact: extractValue(state, ["Next required artifact", "下一项必要输出"]),
-    immediateAction: extractValue(state, ["Immediate action", "立即要做的动作"]),
-    blocker: extractValue(state, ["Current blocker", "当前阻塞"]),
-    humanDecision: extractValue(state, ["Human decision needed", "是否需要人工决策"]),
+    stage: extractValue(workflowState, ["Active stage", "当前阶段", "Stage"]),
+    workflowObjective: extractValue(workflowState, ["Current objective", "当前目标"]),
+    nextArtifact: extractValue(workflowState, ["Next required artifact", "下一项必要输出"]),
+    latestAction: extractValue(workflowState, ["Latest completed action", "Last completed action", "最近完成动作"]),
+    latestArtifactPath: extractValue(workflowState, ["Latest artifact path", "最新工件路径"]),
+    latestRunOrReportId: extractValue(workflowState, ["Latest run or report id", "最新 run 或 report id"]),
+    immediateAction: extractValue(workflowState, ["Immediate action", "立即要做的动作"]),
+    blocker: extractValue(workflowState, ["Current blocker", "Blocking issue", "当前阻塞"]),
+    humanDecision: extractValue(workflowState, ["Human decision needed", "是否需要人工决策"]),
     threshold: extractValue(mission, ["Success threshold", "成功阈值"]),
     boundary: extractValue(mission, ["Hard constraints", "硬约束"]),
     claim: extractClaim(evidence),
     question: extractOpenQuestion(questions),
-    risk: extractValue(questions, ["Why it matters", "为什么重要"]) || extractValue(state, ["Current blocker", "当前阻塞"]),
+    risk:
+      extractValue(questions, ["Why it matters", "为什么重要"]) ||
+      extractValue(workflowState, ["Current blocker", "Blocking issue", "当前阻塞"]),
     datasetPackage:
       extractValue(dataDecisions, ["Approved dataset package", "已批准数据集包"]) ||
       extractValue(dataDecisions, ["Approved datasets", "已批准数据集"]),
@@ -1583,6 +1723,8 @@ function refreshContext({ targetDir }) {
   hydrateCanonicalContext(targetDir);
   const lang = readWorkflowLanguage(targetDir);
   const snapshot = buildContextSnapshot(targetDir);
+  writeContextFile(targetDir, "state.md", renderResearchState(lang, snapshot));
+  writeContextFile(targetDir, "workflow-state.md", renderWorkflowState(lang, snapshot));
   writeContextFile(targetDir, "summary.md", renderSummary(lang, snapshot));
   writeContextFile(targetDir, "next-action.md", renderNextAction(lang, snapshot));
   writeContextFile(targetDir, "session-brief.md", renderSessionBrief(lang, snapshot));

package/lib/i18n.cjs

@@ -55,12 +55,12 @@ const ZH_CONTENT = {
   [path.join(".codex", "prompts", "lab-report.md")]: codexPrompt(
     "基于验证后的迭代工件生成最终报告",
     "report context",
-    "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:report`，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本命令运行 `/lab:report` 阶段。它必须生成给用户直接阅读的最终实验报告和受管的 `main-tables.md`，明确写出主指标、次级指标和必要终局证据，并用白话解释这些指标分别衡量什么、哪些只是健康度或支持性指标、以及每张主表到底证明了什么和没证明什么。"
+    "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:report`，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本命令运行 `/lab:report` 阶段。它必须生成给用户直接阅读的最终实验报告、受管的 `main-tables.md`，以及单独的内部 `artifact-status.md`。主报告要明确写出主指标、次级指标和必要终局证据，并用白话解释这些指标分别衡量什么、哪些只是健康度或支持性指标、以及每张主表到底证明了什么和没证明什么。"
   ),
   [path.join(".codex", "prompts", "lab-write.md")]: codexPrompt(
     "把验证过的研究工件转成论文 section，并按小步方式修订",
     "section or writing target",
-    "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:write`，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本命令运行 `/lab:write` 阶段。它必须先有来自 `/lab:framing` 的已批准 framing artifact，再读取 `.codex/skills/lab/references/paper-writing/` 下与当前 section 对应的参考文件；如果当前是 `abstract`、`introduction` 或 `method`，还必须继续读取 `.codex/skills/lab/references/paper-writing/examples/index.md`、对应的 examples index，以及 1-2 个具体 example 文件。然后结合 `paper-review.md` 与 `does-my-writing-flow-source.md`，先写 mini-outline，再只修改一个 section。第一次进入会产出论文 `.tex` 的路径时，如果 `paper_template_root` 为空，必须先问一次：继续使用默认 LaTeX scaffold，还是先接入模板目录。"
+    "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:write`，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本命令运行 `/lab:write` 阶段。它必须先有来自 `/lab:framing` 的已批准 framing artifact，再读取 `.codex/skills/lab/references/paper-writing/` 下与当前 section 对应的参考文件；如果当前 section 有 bundled example bank，还必须继续读取 `.codex/skills/lab/references/paper-writing/examples/index.md`、对应的 examples index，以及 1-2 个具体 example 文件。然后结合 `paper-review.md` 与 `does-my-writing-flow-source.md`，先写 mini-outline，先规划该 section 需要的表、图和 citation，再只修改一个 section。最终定稿或导出轮次必须运行 `.lab/.managed/scripts/validate_manuscript_delivery.py --paper-dir <deliverables_root>/paper`。第一次进入会产出论文 `.tex` 的路径时，如果 `paper_template_root` 为空，必须先问一次：继续使用默认 LaTeX scaffold，还是先接入模板目录。"
   ),
   [path.join(".claude", "commands", "lab-idea.md")]: claudeCommand(
     "lab-idea",
@@ -102,7 +102,7 @@ const ZH_CONTENT = {
     "lab-report",
     "基于验证后的迭代工件生成最终报告",
     "report context",
-    "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 `report` 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本命令运行 lab workflow 的 `report` 阶段。它必须生成给用户直接阅读的最终实验报告和受管的 `main-tables.md`，明确写出主指标、次级指标和必要终局证据，并用白话解释这些指标分别衡量什么、哪些只是健康度或支持性指标、以及每张主表到底证明了什么和没证明什么。"
+    "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 `report` 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本命令运行 lab workflow 的 `report` 阶段。它必须生成给用户直接阅读的最终实验报告、受管的 `main-tables.md`，以及单独的内部 `artifact-status.md`。主报告要明确写出主指标、次级指标和必要终局证据，并用白话解释这些指标分别衡量什么、哪些只是健康度或支持性指标、以及每张主表到底证明了什么和没证明什么。"
   ),
   [path.join(".claude", "commands", "lab-write.md")]: claudeCommand(
     "lab-write",
@@ -300,6 +300,7 @@ const ZH_SKILL_FILES = {
 - 实验设置
 - 已验证主结果
 - 位于 \`<deliverables_root>/main-tables.md\` 的受管主表工件
+- 位于 \`<deliverables_root>/artifact-status.md\` 的内部工件状态
 - 怎么看主表的阅读指引
 - 消融
 - 失败尝试
@@ -311,6 +312,7 @@ const ZH_SKILL_FILES = {
 
 - \`.lab/context/mission.md\`
 - \`.lab/context/state.md\`
+- \`.lab/context/workflow-state.md\`
 - \`.lab/context/decisions.md\`
 - \`.lab/context/evidence-index.md\`
 - \`.lab/context/eval-protocol.md\`
@@ -320,6 +322,7 @@ const ZH_SKILL_FILES = {
 - \`.lab/context/mission.md\`
 - \`.lab/context/eval-protocol.md\`
 - \`.lab/context/state.md\`
+- \`.lab/context/workflow-state.md\`
 - \`.lab/context/evidence-index.md\`
 
 ## 证据规则
@@ -337,14 +340,14 @@ const ZH_SKILL_FILES = {
 - 方法概述必须用协作者能读懂的话说明：我们的方法大致怎么做、相对 closest prior work 或 strongest baseline 改了什么、这些 prior 方法各自做了什么，以及它们为什么在当前 claim 下仍然不够。
 - 只保留少量最关键的 prior work/baseline 锚点；每个锚点都要用一句话交代它做了什么和它的局限。
 - 在“背景来源”“方法与基线来源”“指标来源”里，每个锚点都必须包含：引用、它做了什么或衡量什么、以及至少一个局限或 caveat。
-- 内部 provenance 只能放到 \`工件状态\` 或 \`.lab/context/evidence-index.md\`，不能塞进来源章节。
+- 内部 provenance 只能放到 \`<deliverables_root>/artifact-status.md\` 或 \`.lab/context/evidence-index.md\`，不能塞进来源章节。
 - 在起草报告前，先检查 \`.lab/context/mission.md\` 和 \`.lab/context/eval-protocol.md\` 是否仍是模板空壳。
 - 如果 canonical context 还是空壳，要先根据 frozen result artifacts、data-decisions、evidence-index 和已批准上下文回填“最小可信版本”，再写报告。
 - 如果回填后仍缺少协作者可读所需的关键字段，就必须把输出降级成 \`artifact-anchored interim report\`，不能冒充最终协作者报告。
 - 如果现有的 \`report.md\` 或 \`main-tables.md\` 缺少受管模板要求的协作者可读章节，也必须视为报告缺陷；rerun 需要补齐这些缺失块，不能直接宣称“正文无变化”或把这次 rerun 当成 no-op。
 - 报告起草或 rerun 完成后，必须运行 \`.lab/.managed/scripts/validate_collaborator_report.py --report <deliverables_root>/report.md --main-tables <deliverables_root>/main-tables.md\`。如果校验失败，就继续补正文，不能停在“只新增审计痕迹”的状态。
 - 如果报告依赖了对原始指标或原始实现的偏差，必须明确写出这个偏差。
-- workflow 工件状态、rerun id 或 LaTeX 骨架状态不能混进“已验证主结果”；这些内容必须单列到工件状态部分。
+- workflow 工件状态、rerun id 或 LaTeX 骨架状态不能混进“已验证主结果”；这些内容必须单列到 \`<deliverables_root>/artifact-status.md\`。
 - 如果 workflow language 是中文，\`report.md\` 和 \`<deliverables_root>/main-tables.md\` 也应使用中文，除非文件路径、代码标识符或字面指标名必须保持原样。
 - 解释优先保守，不要写成营销文案。
 - 要给 \`/lab:write\` 留下清晰 handoff，尤其是 section draft 可以直接引用的证据链接。
@@ -814,11 +817,6 @@ const ZH_SKILL_FILES = {
 - 最终表现摘要：
 - 主表覆盖情况：
 
-## 工件状态
-
-- 已就绪的交付物或工作流工件：
-- 这些工件状态为什么不是科学结论：
-
 ## 主要结果
 
 - 主要发现 1：
@@ -833,6 +831,36 @@ const ZH_SKILL_FILES = {
 
 - Claim：
 - 缺失支持：
+`,
+  [path.join(".lab", ".managed", "templates", "artifact-status.md")]:
+`# 工件状态
+
+## 交付物状态
+
+- 协作者报告路径：
+- 受管主表路径：
+- 当前报告模式：
+- 为什么当前状态是合理的：
+
+## 工作流审计
+
+- 最近完成动作：
+- 最新工件路径：
+- 最新 run 或 report id：
+- rerun 或校验备注：
+
+## 内部溯源
+
+- 使用的冻结结果工件：
+- 已刷新 canonical context：
+- Evidence index 锚点：
+
+## 论文交接
+
+- 已可进入 \`/lab:write\` 的 sections：
+- 可引用的证据包：
+- 仍需要更强支持的 claims：
+- 仍未完成的 paper-finishing 项：
 `,
   [path.join(".lab", ".managed", "templates", "main-tables.md")]:
 `# 主表工件
@@ -1056,6 +1084,33 @@ const ZH_SKILL_FILES = {
 - 协作者可读状态：
 `,
   [path.join(".lab", "context", "state.md")]:
+`# 研究状态
+
+## 已批准方向
+
+- One-sentence problem:
+- Approved direction:
+- Strongest supported claim:
+
+## 证据边界
+
+- What the current evidence really supports:
+- What is still outside the boundary:
+- Biggest research risk:
+
+## 当前研究主线
+
+- Current research focus:
+- Primary metric:
+- Dataset or benchmark scope:
+
+## 当前研究约束
+
+- Hard constraints:
+- Claim boundary:
+- Conditions that require reopening the direction:
+`,
+  [path.join(".lab", "context", "workflow-state.md")]:
 `# 工作流状态
 
 ## 当前阶段
@@ -1172,8 +1227,8 @@ const ZH_SKILL_FILES = {
 
 - Run stage contract: write persistent outputs under \`results_root\`.
 - Iterate stage contract: update persistent outputs under \`results_root\`.
-- Review stage contract: update canonical review context such as \`.lab/context/decisions.md\`、\`state.md\`、\`open-questions.md\` or \`evidence-index.md\`.
-- Report stage contract: write the final report to \`<deliverables_root>/report.md\`.
+- Review stage contract: update canonical review context such as \`.lab/context/decisions.md\`、\`state.md\`、\`workflow-state.md\`、\`open-questions.md\` or \`evidence-index.md\`.
+- Report stage contract: write \`<deliverables_root>/report.md\`、\`<deliverables_root>/main-tables.md\` and \`<deliverables_root>/artifact-status.md\`.
 - Write stage contract: write LaTeX output under \`<deliverables_root>/paper/\`.
 
 ## 升格策略
@@ -1284,7 +1339,7 @@ ZH_CONTENT[path.join(".lab", "system", "core.md")] = `# Lab 系统核心
 
 1. \`.lab/context/session-brief.md\`
 2. \`.lab/context/mission.md\`
-3. \`.lab/context/state.md\`
+3. \`.lab/context/workflow-state.md\`
 4. \`.lab/context/evidence-index.md\`
 5. \`.lab/context/data-decisions.md\`（当问题涉及数据集、benchmark 或对比方法时）
 6. \`.lab/context/auto-mode.md\` 和 \`.lab/context/auto-status.md\`（当任务涉及自动模式时）
@@ -1293,13 +1348,15 @@ ZH_CONTENT[path.join(".lab", "system", "core.md")] = `# Lab 系统核心
 
 ## 工作流边界
 
-- \`.lab/context/\` 保存持久研究状态。
+- \`.lab/context/\` 同时保存持久研究状态和轻量工作流状态。
 - \`.lab/changes/\`、\`.lab/iterations/\`、\`.lab/writing/\` 保存工作流控制工件、轻量 manifest 和 change-local harness。
 - \`.lab/.managed/\` 保存工具托管模板和脚本。
 - 持久 run 输出应写到 \`results_root\`，不要写进 \`.lab/changes/\`。
 - 图表和可视化应写到 \`figures_root\`，不要写进 \`.lab/changes/\`。
 - 最终交付物应写到 \`deliverables_root\`，不要写进 \`.lab/context/\`。
 - change-local 的 \`data/\` 目录只应用来放轻量 manifest 或 batch spec，不要当正式数据集入口。
+- \`.lab/context/state.md\` 保存持久研究状态；\`.lab/context/workflow-state.md\` 保存当前工作流状态。
+- \`.lab/context/summary.md\` 是长期项目摘要；\`.lab/context/session-brief.md\` 是下一次会话启动简报。
 - \`.lab/context/auto-mode.md\` 定义自动模式边界，\`.lab/context/auto-status.md\` 记录自动运行状态，二者都属于项目状态。
 - 如果用户提供了 LaTeX 模板目录，先校验并通过 \`paper_template_root\` 接入，再开始写作。
 - 已接入的模板目录视为用户资产，默认不要改模板文件，除非用户明确要求。
@@ -1391,7 +1448,7 @@ ZH_CONTENT[path.join(".lab", "context", "session-brief.md")] = `# 会话简报
 ## 先读这些文件
 
 1. \`.lab/context/mission.md\`
-2. \`.lab/context/state.md\`
+2. \`.lab/context/workflow-state.md\`
 3. \`.lab/context/evidence-index.md\`
 
 ## 不要静默修改
@@ -1526,8 +1583,20 @@ ZH_CONTENT[path.join(".codex", "skills", "lab", "stages", "write.md")] = `# \`/l
 - 如果用户在最终提醒里仍确认继续使用默认 scaffold，就把 \`paper_template_final_reminder_acknowledged\` 持久化为 \`true\`。
 - 只加载当前 section guide，不要一次加载全部章节参考。
 - 如果当前 section 是 \`abstract\`、\`introduction\` 或 \`method\`，还必须继续读取本地 example bank：\`references/paper-writing/examples/index.md\`、对应的 examples index，以及 1-2 个具体 example 文件。
+- 如果当前 section 是 \`related work\`、\`experiments\` 或 \`conclusion\`，也要读取对应的本地 example bank：\`references/paper-writing/examples/index.md\`、对应的 examples index，以及 1-2 个具体 example 文件。
 - 例子只能复用结构、段落角色和句法逻辑，不能直接复用原句。
 - 先写 mini-outline 再写 prose。
+- 如果当前 section 带方法或实验 claim，先规划需要的表格、figure placeholders 和 citations，再写 prose。
+- 最终定稿或导出轮次里，必须维护非空的 \`references.bib\`，并把关键实验结果物化成真正的 LaTeX 表格：
+  - \`<deliverables_root>/paper/tables/main-results.tex\`
+  - \`<deliverables_root>/paper/tables/ablations.tex\`
+- 方法或实验章节还必须物化带 caption、label 和图意图的 figure placeholders：
+  - \`<deliverables_root>/paper/figures/method-overview.tex\`
+  - \`<deliverables_root>/paper/figures/results-overview.tex\`
+- 不要把纯 prose section 当成论文交付完成。
+- 论文源码里不要出现绝对本地路径、workflow provenance、shell 日志或 \`+/-\` 这类工程记法；应改成 paper-facing LaTeX 表达。
+- 在最终定稿或导出轮次结束前，必须运行 \`.lab/.managed/scripts/validate_manuscript_delivery.py --paper-dir <deliverables_root>/paper\`；如果失败，就继续补表、图、引用和正文。
+- 如果本地有 LaTeX 工具链，就再跑一次 compile smoke test；如果没有，也要在 write iteration artifact 里记录该限制。
 - 如果缺少 framing artifact，不要继续写作，直接回到 \`/lab:framing\`。
 - 如果某个 claim 没有证据支撑，就削弱或删除。
 
@@ -1537,7 +1606,12 @@ ZH_CONTENT[path.join(".codex", "skills", "lab", "stages", "write.md")] = `# \`/l
 - \`.lab/writing/plan.md\`
 - \`.lab/writing/iterations/<n>.md\`
 - \`<deliverables_root>/paper/main.tex\`
+- \`<deliverables_root>/paper/references.bib\`
 - \`<deliverables_root>/paper/sections/<section>.tex\`
+- \`<deliverables_root>/paper/tables/main-results.tex\`
+- \`<deliverables_root>/paper/tables/ablations.tex\`
+- \`<deliverables_root>/paper/figures/method-overview.tex\`
+- \`<deliverables_root>/paper/figures/results-overview.tex\`
 `;
 
 ZH_CONTENT[path.join(".claude", "skills", "lab", "stages", "write.md")] =

package/lib/install.cjs

@@ -36,6 +36,7 @@ const PROJECT_OWNED_LOCALIZED_PATHS = [
   path.join(".lab", "config", "workflow.json"),
   path.join(".lab", "context", "mission.md"),
   path.join(".lab", "context", "state.md"),
+  path.join(".lab", "context", "workflow-state.md"),
   path.join(".lab", "context", "decisions.md"),
   path.join(".lab", "context", "evidence-index.md"),
   path.join(".lab", "context", "open-questions.md"),
@@ -542,6 +543,7 @@ function localizeInstalledAssets(targetDir, lang, { newlyCreatedProjectOwnedPath
     path.join(".lab", ".managed", "templates", "review-checklist.md"),
     path.join(".lab", ".managed", "templates", "final-report.md"),
     path.join(".lab", ".managed", "templates", "main-tables.md"),
+    path.join(".lab", ".managed", "templates", "artifact-status.md"),
     path.join(".lab", ".managed", "templates", "paper-plan.md"),
     path.join(".lab", ".managed", "templates", "paper-section.md"),
     path.join(".lab", ".managed", "templates", "write-iteration.md"),

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

@@ -7,4 +7,4 @@ argument-hint: section or writing target
 Use the installed `lab` skill at `.claude/skills/lab/SKILL.md`.
 
 Execute the requested `/lab-write` 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.
-This command runs the `write` stage of the lab workflow. It requires an approved framing artifact from the `framing` stage, must read the matching section reference from `.claude/skills/lab/references/paper-writing/`, and for `abstract`, `introduction`, or `method` it must also read `.claude/skills/lab/references/paper-writing/examples/index.md` plus the matching examples index and 1-2 concrete example files. Then it should run `paper-review.md` and `does-my-writing-flow-source.md`, build a mini-outline, and revise only one section.
+This command runs the `write` stage of the lab workflow. It requires an approved framing artifact from the `framing` stage, must read the matching section reference from `.claude/skills/lab/references/paper-writing/`, and for any section with a bundled example bank it must also read `.claude/skills/lab/references/paper-writing/examples/index.md` plus the matching examples index and 1-2 concrete example files. Then it should run `paper-review.md` and `does-my-writing-flow-source.md`, build a mini-outline, plan the section's paper-facing tables/figures/citations, and revise only one section. Final-draft or export rounds must run `.lab/.managed/scripts/validate_manuscript_delivery.py --paper-dir <deliverables_root>/paper` before stopping.