@archsight/aios 1.3.1 → 1.4.0

package/prompts/prompt-registry.md

@@ -27,8 +27,10 @@ Prompt 修改必须记录原因、预期行为变化和评估结果。
 | 名称 | 版本 | 适用场景 | 输入 | 输出 | 禁止场景 | 评估方式 | 路径 |
 |---|---|---|---|---|---|---|---|
 | 招标文件解析与技术标响应矩阵基础提示词 | 0.1 | 工程招投标、技术标响应、AI 标书工具复核问题整理 | 招标文件、评分办法、技术标要求、人工检查问题清单 | 资料来源清单、输入类型判断、问题回应矩阵、评分点响应矩阵、资料缺口 | 中标概率、评标结论、采购承诺、串标规避建议 | 是否先判断资料状态，是否避免编造废标项和评分点 | `skills/aios-commercial-tender/prompts/basic-prompt.md` |
+| 工程技术标生成与改写基础提示词 | 0.1 | 工程标书、技术标、投标响应章节生成 / 改写和历史标书素材复用 | 招标文件、评分办法、技术标要求、历史标书素材、类似项目案例、用户初稿 | 资料来源清单、写作 brief、历史素材匹配表、评分点到章节映射、章节初稿、审核门禁交接 | 中标概率、评标结论、最终投标决策、旧项目事实套用 | 是否保留 Markdown 工作母版、素材复用判断、待补占位和 `aios-commercial-tender` 门禁 | `skills/aios-tender-write/prompts/basic-prompt.md` |
 | 合同条款与履约节点基础提示词 | 0.1 | 工程分包、采购、租赁和补充协议履约资料整理 | 合同条款、协议片段、合同摘要 | 资料来源清单、空白字段核对表、履约节点、付款结算条件、风险提示 | 法律意见、违约定性、索赔或结算金额结论 | 是否提取可执行节点，是否保留法务和商务复核边界 | `skills/aios-commercial-contract/prompts/basic-prompt.md` |
 | 项目日报问题跟踪基础提示词 | 0.1 | 施工日报、项目日报、周报素材和项目群日报整理 | 日报、周报、现场记录 | 资料来源清单、管理摘要、问题台账、模板质量诊断、需确认事项 | 工期偏差终局判断、质量安全验收结论、签证或索赔成立判断 | 是否区分现场事实和模板空白，是否避免把未提及写成不合格 | `skills/aios-construction-daily/prompts/basic-prompt.md` |
 | 会议纪要待办闭环基础提示词 | 0.1 | 工程会议、协调会、专题会、交底会记录整理 | 会议纪要、会议记录、录音转写摘要 | 资料来源清单、会议结论、待办闭环、责任线索、遗留问题 | 正式责任归属、合同通知效力、审批通过结论 | 是否区分发言人与最终责任人，是否标注未明确期限和责任人 | `skills/aios-construction-meeting/prompts/basic-prompt.md` |
 | 变更签证资料链基础提示词 | 0.1 | 工程变更、签证、联系单、会议纪要和合同条款链路整理 | 签证单、联系单、合同条款、现场记录、公开样表字段 | 资料来源清单、资料链完整度、字段结构、资料断点、复核分流 | 签证成立、索赔成立、责任归属、最终金额 | 是否区分过程线索和正式依据，是否避免金额和责任判断 | `skills/aios-commercial-variation/prompts/basic-prompt.md` |
 | 施工方案辅助复核基础提示词 | 0.1 | 专项施工方案、技术方案、专家修改说明和 AI 方案试用反馈整理 | 施工方案片段、工程概况、安全措施、专家修改说明、计算书目录 | 资料来源清单、失准复盘、关键工序、危险源、交底要点、回查清单 | 方案合格结论、计算正确结论、专家论证通过结论 | 是否使用辅助复核口径，是否保留总工 / 专家 / 计算书复核边界 | `skills/aios-construction-scheme/prompts/basic-prompt.md` |
+| 专项施工方案生成与改写基础提示词 | 0.1 | 专项施工方案、施工技术措施、交底材料生成 / 改写和历史方案素材复用 | 方案初稿、工程概况、施工工艺、安全措施、历史方案素材、专家意见、计算书目录 | 资料来源清单、写作 brief、历史素材匹配表、方案章节结构、方案章节初稿、危险源占位表、审核门禁交接 | 方案合格、计算正确、专家论证通过、审批通过、旧项目参数套用 | 是否保留 Markdown 工作母版、素材复用判断、危险源占位、待补资料和 `aios-construction-scheme` 门禁 | `skills/aios-scheme-write/prompts/basic-prompt.md` |

package/runtime/archsight-aios.manifest.json

@@ -1,7 +1,7 @@
 {
   "schema": 1,
   "name": "archsight-aios",
-  "version": "1.3.1",
+  "version": "1.4.0",
   "description": "ArchSight AIOS 的 Agent、Skill、Workflow、Runtime 和项目模板注册表，面向建筑行业知识工作者与 AI 工程团队。",
   "agents": [
     {
@@ -195,6 +195,13 @@
       "path": "skills/aios-commercial-tender/SKILL.md",
       "openaiConfigPath": "skills/aios-commercial-tender/agents/openai.yaml"
     },
+    {
+      "id": "aios-tender-write",
+      "primaryAgent": "mason",
+      "defaultWorkflow": "feature-development",
+      "path": "skills/aios-tender-write/SKILL.md",
+      "openaiConfigPath": "skills/aios-tender-write/agents/openai.yaml"
+    },
     {
       "id": "aios-commercial-contract",
       "primaryAgent": "themis",
@@ -229,6 +236,13 @@
       "defaultWorkflow": "review",
       "path": "skills/aios-construction-scheme/SKILL.md",
       "openaiConfigPath": "skills/aios-construction-scheme/agents/openai.yaml"
+    },
+    {
+      "id": "aios-scheme-write",
+      "primaryAgent": "vitruvius",
+      "defaultWorkflow": "feature-development",
+      "path": "skills/aios-scheme-write/SKILL.md",
+      "openaiConfigPath": "skills/aios-scheme-write/agents/openai.yaml"
     }
   ],
   "workflows": [
@@ -374,6 +388,12 @@
       "agent": "mason",
       "workflow": "review"
     },
+    {
+      "taskType": "工程标书、技术标、投标响应章节生成、改写和历史素材复用",
+      "skill": "aios-tender-write",
+      "agent": "mason",
+      "workflow": "feature-development"
+    },
     {
       "taskType": "工程合同履约节点、付款条件、责任边界和资料缺口",
       "skill": "aios-commercial-contract",
@@ -403,6 +423,12 @@
       "skill": "aios-construction-scheme",
       "agent": "vitruvius",
       "workflow": "review"
+    },
+    {
+      "taskType": "专项施工方案、施工技术措施、交底材料生成、改写和历史方案复用",
+      "skill": "aios-scheme-write",
+      "agent": "vitruvius",
+      "workflow": "feature-development"
     }
   ],
   "projectTemplate": {
@@ -476,13 +502,16 @@
     "governance/ai-review-policy.md",
     "governance/security-policy.md",
     "governance/agent-boundary.md",
-    "governance/context-policy.md",
-    "governance/memory-policy.md",
-    "governance/delivery-policy.md",
-    "delivery/release-checklist.md",
-    "delivery/ai-generated-code-checklist.md",
-    "delivery/rollback-policy.md",
-    "memory/decision-records.md",
+    "governance/context-policy.md",
+    "governance/memory-policy.md",
+    "governance/delivery-policy.md",
+    "delivery/release-checklist.md",
+    "delivery/v1.4.0-release-readiness.md",
+    "delivery/ai-generated-code-checklist.md",
+    "delivery/rollback-policy.md",
+    "docs/v1.4.0-writing-workflow-quickstart.md",
+    "docs/v1.4.0-writing-boundary.md",
+    "memory/decision-records.md",
     "memory/project-memory-policy.md",
     "memory/cleanup-policy.md",
     "knowledge/domain-taxonomy.md",
@@ -503,6 +532,16 @@
     "prompts/evaluations/engineering-business-basic-prompts-2026-06-16.md",
     "prompts/evaluations/engineering-business-basic-scorecard.json",
     "prompts/evaluations/engineering-business-public-advisory-fixtures.json",
+    "prompts/evaluations/engineering-document-writing-fixtures.json",
+    "prompts/evaluations/engineering-document-writing-scorecard.json",
+    "prompts/evaluations/skill-runtime/README.md",
+    "prompts/evaluations/skill-runtime/v1.4.0-writing-host-validation.json",
+    "prompts/evaluations/skill-runtime/v1.4.0-writing-host-validation.md",
+    "prompts/evaluations/skill-runtime/v1.4.0-writing-host-scorecard-review.md",
+    "prompts/evaluations/skill-runtime/raw/codex-tender-writing-sample.md",
+    "prompts/evaluations/skill-runtime/raw/codex-scheme-writing-sample.md",
+    "prompts/evaluations/skill-runtime/raw/workbuddy-tender-writing-sample.md",
+    "prompts/evaluations/skill-runtime/raw/workbuddy-scheme-writing-sample.md",
     "prompts/evaluations/public-advisory-md/01-technical-bid.md",
     "prompts/evaluations/public-advisory-md/02-contract.md",
     "prompts/evaluations/public-advisory-md/03-daily.md",
@@ -514,17 +553,40 @@
     "infra/service-boundaries.md",
     "skills/engineering-business-starter-kit.md",
     "skills/aios-commercial-tender/prompts/basic-prompt.md",
+    "skills/aios-tender-write/prompts/basic-prompt.md",
     "skills/aios-commercial-contract/prompts/basic-prompt.md",
     "skills/aios-construction-daily/prompts/basic-prompt.md",
     "skills/aios-construction-meeting/prompts/basic-prompt.md",
     "skills/aios-commercial-variation/prompts/basic-prompt.md",
     "skills/aios-construction-scheme/prompts/basic-prompt.md",
+    "skills/aios-scheme-write/prompts/basic-prompt.md",
+    "templates/document-writing/source-normalized.md",
+    "templates/document-writing/material-index.md",
+    "templates/document-writing/writing-brief.md",
+    "templates/document-writing/draft.md",
+    "templates/document-writing/review-notes.md",
+    "templates/document-writing/final.md",
+    "templates/document-writing-samples/tender/README.md",
+    "templates/document-writing-samples/tender/source-normalized.md",
+    "templates/document-writing-samples/tender/material-index.md",
+    "templates/document-writing-samples/tender/writing-brief.md",
+    "templates/document-writing-samples/tender/draft.md",
+    "templates/document-writing-samples/tender/review-notes.md",
+    "templates/document-writing-samples/tender/final.md",
+    "templates/document-writing-samples/scheme/README.md",
+    "templates/document-writing-samples/scheme/source-normalized.md",
+    "templates/document-writing-samples/scheme/material-index.md",
+    "templates/document-writing-samples/scheme/writing-brief.md",
+    "templates/document-writing-samples/scheme/draft.md",
+    "templates/document-writing-samples/scheme/review-notes.md",
+    "templates/document-writing-samples/scheme/final.md",
     "scripts/analyze-prompt-run-results.mjs",
     "scripts/build-prompt-run-pack.mjs",
     "scripts/validate-prompt-fixtures.mjs",
     "scripts/validate-prompt-model-outputs.mjs",
     "scripts/validate-prompt-run-results.mjs",
     "scripts/validate-prompt-scorecard.mjs",
+    "scripts/validate-skill-runtime-evidence.mjs",
     "runtime/hermes/sync-policy.md",
     "runtime/hermes/sync-record-template.md",
     "runtime/capability-adapters.json",

package/runtime/skill-routing.md

@@ -33,11 +33,13 @@
 | 用户明确调用 `aios-compare` 时，比较两份文档、两个版本或两个 AI 输出哪份更专业 | `aios-compare` | Daedalus | `review` |
 | 建筑行业项目中的受控代码修改、文档、脚本、测试 | `aios-exec` | Hephaestus | `feature-development` |
 | 工程招投标响应、评分点、废标风险和技术标资料矩阵 | `aios-commercial-tender` | Mason | `review` |
+| 工程标书、技术标和投标响应章节生成 / 改写 / 历史素材复用 | `aios-tender-write` | Mason | `feature-development` |
 | 工程合同履约节点、付款条件、责任边界和资料缺口 | `aios-commercial-contract` | Themis | `review` |
 | 施工日报、现场异常、项目群记录和问题追踪台账 | `aios-construction-daily` | Mason | `site-daily-loop` |
 | 工程会议纪要、待办闭环、遗留争议和下次追踪 | `aios-construction-meeting` | Mason | `site-daily-loop` |
 | 工程变更签证资料链、联系单、图纸变更和索赔线索 | `aios-commercial-variation` | Plutus | `site-daily-loop` |
 | 专项施工方案、危险源、交底要点和规范 / 计算书复核清单 | `aios-construction-scheme` | Vitruvius | `review` |
+| 专项施工方案、施工技术措施和交底材料生成 / 改写 / 历史方案复用 | `aios-scheme-write` | Vitruvius | `feature-development` |
 
 ## 路由原则
 
@@ -51,7 +53,8 @@
 - `aios-design` 用于实现前判断界面方案是否支撑建筑行业审查、定位、复核、追溯和交付；不替代 `frontend-generation` 的 UI 实现、布局验证和交互验证，也不替代通用 `frontend-design` 的视觉风格和前端代码美化评审。
 - `aios-arch` 应补足通用架构评审缺失的建筑行业平台视角，包括 BIM / IFC、规范知识链路、审图证据链、RAG / GraphRAG、任务编排、审计和后端运行可靠性。
 - `aios-structural` 用于结构力学、荷载、边界条件、FEM 和求解器接口评审；它不能替代结构工程师签审，关键数值必须来自 Capability 或项目已有求解器证据。
-- `aios-commercial-tender`、`aios-commercial-contract`、`aios-construction-daily`、`aios-construction-meeting`、`aios-commercial-variation` 和 `aios-construction-scheme` 属于工程业务管理增强；它们只处理建筑工程资料的抽取、证据链整理、风险提示和人工复核分流，不扩展为通用 HR、行政、财务 Skill。
+- `aios-commercial-tender`、`aios-tender-write`、`aios-commercial-contract`、`aios-construction-daily`、`aios-construction-meeting`、`aios-commercial-variation`、`aios-construction-scheme` 和 `aios-scheme-write` 属于工程业务管理增强；它们只处理建筑工程资料的抽取、生成初稿、证据链整理、风险提示和人工复核分流，不扩展为通用 HR、行政、财务 Skill。
+- `aios-tender-write` 和 `aios-scheme-write` 是写作型 Skill，默认使用 Markdown 工作母版；生成后必须分别交回 `aios-commercial-tender` 和 `aios-construction-scheme` 做审核门禁。
 - 工程业务 Agent 分工：技术标以 Mason 为主；合同法律边界以 Themis 为主；变更签证、工程款、结算和成本线索以 Plutus 为主；会议纪要中的行政、人事、证照和组织协同事项由 Hestia 辅助分流；施工方案以 Vitruvius 为主，涉及结构计算时升级给 Euclid，涉及现场组织和交付时由 Mason 协同。
 - 工程业务管理基础场景可先参考 `skills/engineering-business-starter-kit.md` 和各 Skill 目录下的 `prompts/basic-prompt.md`，形成矩阵、清单、台账和复核问题；涉及金额、工期、责任、合规、质量安全、结构计算或法律意见时，再按对应 Skill 的证据链和人工复核规则升级。
 - `aios-compare` 用于普通两份文档 / 两个版本 / 两个 AI 输出的专业度对比；不做 weak / portable / skill-runtime 提示词评测。

package/scripts/build-prompt-run-pack.mjs

@@ -121,6 +121,31 @@ function runPackName() {
   return `${fixture?.name ?? "prompt-fixture"}-run-pack`;
 }
 
+function runPackDataBoundary() {
+  if (fixture?.name === "engineering-document-writing-fixtures") {
+    return "De-identified engineering document writing run pack. Inputs are synthetic or abstracted writing-task shapes; do not add real customer names, contacts, project names, amounts, dates, locations, or raw source documents.";
+  }
+
+  return "De-identified weak/basic prompt run pack. Public advisory fixtures use Markdown-normalized synthetic inputs; do not add real customer names, contacts, project names, amounts, dates, locations, or raw source documents.";
+}
+
+function runPackInstructions() {
+  const instructions = [
+    "For each item, use prompt as the instruction and sampleInput as the user-provided material.",
+    "When inputFormat is markdown, pass the Markdown text as the material under review.",
+    "Run weak and basic variants separately for the same caseId.",
+    "Save model outputs into the model-output JSON schema and validate with validate-prompt-model-outputs.mjs."
+  ];
+
+  if (fixture?.name === "engineering-document-writing-fixtures") {
+    instructions.push("Compare weak and basic outputs against expectedStrongSections, bannedClaims, source provenance, material reuse judgment, and review-gate handoff.");
+  } else {
+    instructions.push("Compare weak and basic outputs using engineering-business-basic-scorecard.json.");
+  }
+
+  return instructions;
+}
+
 function buildRunPack() {
   if (!fixture) return undefined;
 
@@ -163,15 +188,8 @@ function buildRunPack() {
     name: runPackName(),
     version: fixture.version,
     fixture: args.fixture,
-    dataBoundary:
-      "De-identified weak/basic prompt run pack. Public advisory fixtures use Markdown-normalized synthetic inputs; do not add real customer names, contacts, project names, amounts, dates, locations, or raw source documents.",
-    runInstructions: [
-      "For each item, use prompt as the instruction and sampleInput as the user-provided material.",
-      "When inputFormat is markdown, pass the Markdown text as the material under review.",
-      "Run weak and basic variants separately for the same caseId.",
-      "Save model outputs into the model-output JSON schema and validate with validate-prompt-model-outputs.mjs.",
-      "Compare weak and basic outputs using engineering-business-basic-scorecard.json."
-    ],
+    dataBoundary: runPackDataBoundary(),
+    runInstructions: runPackInstructions(),
     runs
   };
 }

package/scripts/validate-prompt-fixtures.mjs

@@ -8,7 +8,8 @@ const errors = [];
 
 const fixturePaths = [
   "prompts/evaluations/engineering-business-basic-fixtures.json",
-  "prompts/evaluations/engineering-business-public-advisory-fixtures.json"
+  "prompts/evaluations/engineering-business-public-advisory-fixtures.json",
+  "prompts/evaluations/engineering-document-writing-fixtures.json"
 ];
 const fixtures = fixturePaths.map((fixturePath) => ({
   path: fixturePath,
@@ -143,9 +144,10 @@ function validateFixture(fixturePath, fixture) {
     return;
   }
 
+  const expectations = fixtureExpectations(fixturePath, fixture);
   check(fixture.schema === 1, `${fixturePath}: schema must be 1`);
   check(Array.isArray(fixture.cases), `${fixturePath}: cases must be an array`);
-  check(fixture.cases?.length === 6, `${fixturePath}: must cover 6 engineering-business cases`);
+  check(fixture.cases?.length === expectations.caseCount, `${fixturePath}: must cover ${expectations.caseCount} ${expectations.label} cases`);
 
   const seenIds = new Set();
   const seenSkills = new Set();
@@ -201,17 +203,36 @@ function validateFixture(fixturePath, fixture) {
     }
   }
 
-  const expectedSkills = [
-    "aios-commercial-tender",
-    "aios-commercial-contract",
-    "aios-construction-daily",
-    "aios-construction-meeting",
-    "aios-commercial-variation",
-    "aios-construction-scheme"
-  ].sort();
+  const expectedSkills = expectations.skillIds.sort();
   check(JSON.stringify([...seenSkills].sort()) === JSON.stringify(expectedSkills), `${fixturePath}: skill coverage mismatch`);
 }
 
+function fixtureExpectations(fixturePath, fixture) {
+  if (fixture.name === "engineering-document-writing-fixtures") {
+    return {
+      label: "engineering-document-writing",
+      caseCount: 2,
+      skillIds: [
+        "aios-tender-write",
+        "aios-scheme-write"
+      ]
+    };
+  }
+
+  return {
+    label: "engineering-business",
+    caseCount: 6,
+    skillIds: [
+      "aios-commercial-tender",
+      "aios-commercial-contract",
+      "aios-construction-daily",
+      "aios-construction-meeting",
+      "aios-commercial-variation",
+      "aios-construction-scheme"
+    ]
+  };
+}
+
 for (const fixture of fixtures) {
   validateFixture(fixture.path, fixture.data);
 }

package/scripts/validate-prompt-scorecard.mjs

@@ -6,10 +6,16 @@ import path from "node:path";
 const root = fs.realpathSync(process.cwd());
 const errors = [];
 
-const fixturePath = repoPath("prompts/evaluations/engineering-business-basic-fixtures.json");
-const scorecardPath = repoPath("prompts/evaluations/engineering-business-basic-scorecard.json");
-const fixture = readJson(fixturePath);
-const scorecard = readJson(scorecardPath);
+const scorecardSpecs = [
+  {
+    fixturePath: "prompts/evaluations/engineering-business-basic-fixtures.json",
+    scorecardPath: "prompts/evaluations/engineering-business-basic-scorecard.json"
+  },
+  {
+    fixturePath: "prompts/evaluations/engineering-document-writing-fixtures.json",
+    scorecardPath: "prompts/evaluations/engineering-document-writing-scorecard.json"
+  }
+];
 
 const sensitiveTerms = [
   "立信",
@@ -64,10 +70,16 @@ function weightedScore(scores, criteria) {
   );
 }
 
-if (fixture && scorecard) {
+function validateScorecard(fixturePathValue, scorecardPathValue) {
+  const fixture = readJson(repoPath(fixturePathValue));
+  const scorecard = readJson(repoPath(scorecardPathValue));
+  if (!fixture || !scorecard) {
+    return;
+  }
+
   check(scorecard.schema === 1, "scorecard: schema must be 1");
   check(scorecard.version === fixture.version, `scorecard: version must match fixture version ${fixture.version}`);
-  check(scorecard.fixture === "prompts/evaluations/engineering-business-basic-fixtures.json", "scorecard: fixture path mismatch");
+  check(scorecard.fixture === fixturePathValue, `${scorecardPathValue}: fixture path mismatch`);
   check(Array.isArray(scorecard.criteria) && scorecard.criteria.length > 0, "scorecard: criteria must be a non-empty array");
   check(Array.isArray(scorecard.cases) && scorecard.cases.length === fixture.cases.length, "scorecard: case count mismatch");
 
@@ -124,6 +136,10 @@ if (fixture && scorecard) {
   check(typeof scorecard.overallDecision?.notAClaim === "string", "scorecard: overallDecision.notAClaim is required");
 }
 
+for (const spec of scorecardSpecs) {
+  validateScorecard(spec.fixturePath, spec.scorecardPath);
+}
+
 if (errors.length > 0) {
   console.error(`Prompt scorecard validation failed with ${errors.length} error(s):`);
   for (const error of errors) console.error(`- ${error}`);

package/scripts/validate-skill-runtime-evidence.mjs

@@ -0,0 +1,125 @@
+#!/usr/bin/env node
+
+import fs from "node:fs";
+import path from "node:path";
+
+const root = fs.realpathSync(process.cwd());
+const errors = [];
+const evidencePath = "prompts/evaluations/skill-runtime/v1.4.0-writing-host-validation.json";
+const reportPath = "prompts/evaluations/skill-runtime/v1.4.0-writing-host-validation.md";
+
+const sensitiveTerms = [
+  "立信",
+  "费敏",
+  "闻总",
+  "谭总",
+  "客户内部",
+  "培训演示",
+  "基础内测",
+  "内测模式",
+  "内测包"
+];
+
+function repoPath(...parts) {
+  const target = path.join(root, ...parts);
+  const relative = path.relative(root, target);
+  if (relative.startsWith("..") || path.isAbsolute(relative)) {
+    throw new Error(`Path traversal detected: ${target}`);
+  }
+  return target;
+}
+
+function readJson(relativePath) {
+  try {
+    return JSON.parse(fs.readFileSync(repoPath(relativePath), "utf8"));
+  } catch (error) {
+    errors.push(`${relativePath}: invalid JSON (${error.message})`);
+    return undefined;
+  }
+}
+
+function readText(relativePath) {
+  try {
+    return fs.readFileSync(repoPath(relativePath), "utf8");
+  } catch (error) {
+    errors.push(`${relativePath}: ${error.message}`);
+    return "";
+  }
+}
+
+function exists(relativePath) {
+  return fs.existsSync(repoPath(relativePath));
+}
+
+function check(condition, message) {
+  if (!condition) errors.push(message);
+}
+
+function includesSensitiveTerm(value) {
+  const raw = typeof value === "string" ? value : JSON.stringify(value);
+  return sensitiveTerms.filter((term) => raw.includes(term));
+}
+
+const evidence = readJson(evidencePath);
+const report = readText(reportPath);
+
+if (evidence) {
+  check(evidence.schema === 1, "evidence: schema must be 1");
+  check(evidence.name === "v1.4.0-writing-host-validation", "evidence: unexpected name");
+  check(typeof evidence.aiosVersion === "string" && evidence.aiosVersion.length > 0, "evidence: aiosVersion is required");
+  check(Array.isArray(evidence.hostChecks) && evidence.hostChecks.length >= 2, "evidence: hostChecks must cover at least two hosts");
+  check(Array.isArray(evidence.cases) && evidence.cases.length >= 4, "evidence: cases must cover host x writing skill matrix");
+  check(typeof evidence.scorecardReview?.reviewPath === "string", "evidence: scorecardReview.reviewPath is required");
+  check(exists(evidence.scorecardReview?.scorecardPath ?? ""), "evidence: scorecardReview.scorecardPath missing");
+  check(exists(evidence.scorecardReview?.reviewPath ?? ""), "evidence: scorecardReview.reviewPath missing");
+  check(typeof evidence.releaseGate?.notReadyUntil === "string", "evidence: releaseGate.notReadyUntil is required");
+  check(report.includes(evidence.name), "report: must reference evidence name");
+  const scorecardReview = evidence.scorecardReview?.reviewPath ? readText(evidence.scorecardReview.reviewPath) : "";
+
+  const sensitiveHits = includesSensitiveTerm(evidence);
+  check(sensitiveHits.length === 0, `evidence: sensitive terms leaked (${sensitiveHits.join(", ")})`);
+
+  const expectedPairs = new Set([
+    "codex::aios-tender-write",
+    "codex::aios-scheme-write",
+    "workbuddy::aios-tender-write",
+    "workbuddy::aios-scheme-write"
+  ]);
+  const actualPairs = new Set();
+  const allowedStatuses = new Set(["yes", "no", "blocked", "uncertain"]);
+
+  for (const item of evidence.cases ?? []) {
+    check(typeof item.caseId === "string" && item.caseId.length > 0, "case: caseId is required");
+    check(typeof item.host === "string" && item.host.length > 0, `${item.caseId}: host is required`);
+    check(typeof item.skillId === "string" && item.skillId.startsWith("aios-"), `${item.caseId}: skillId is invalid`);
+    check(typeof item.expectedGateSkill === "string" && item.expectedGateSkill.startsWith("aios-"), `${item.caseId}: expectedGateSkill is invalid`);
+    check(typeof item.sampleWorkspacePath === "string" && exists(item.sampleWorkspacePath), `${item.caseId}: sampleWorkspacePath missing`);
+    check(typeof item.triggerPrompt === "string" && item.triggerPrompt.includes(item.skillId), `${item.caseId}: triggerPrompt must mention skillId`);
+    check(allowedStatuses.has(item.skillRuntimeConfirmed), `${item.caseId}: invalid skillRuntimeConfirmed`);
+    check(Array.isArray(item.evidence) && item.evidence.length > 0, `${item.caseId}: evidence must be non-empty`);
+    actualPairs.add(`${item.host}::${item.skillId}`);
+
+    if (item.skillRuntimeConfirmed === "yes") {
+      check(typeof item.rawOutputPath === "string" && exists(item.rawOutputPath), `${item.caseId}: confirmed runtime needs rawOutputPath`);
+      const raw = readText(item.rawOutputPath);
+      check(raw.includes(item.skillId), `${item.caseId}: raw output must include skillId`);
+      check(raw.includes(item.expectedGateSkill), `${item.caseId}: raw output must include expected gate skill`);
+      check(scorecardReview.includes(item.caseId), `${item.caseId}: scorecard review must mention caseId`);
+    } else {
+      check(item.rawOutputPath === null, `${item.caseId}: non-confirmed runtime must not point to raw output`);
+      check(typeof item.blocker === "string" && item.blocker.length > 0, `${item.caseId}: blocker is required when runtime is not confirmed`);
+    }
+  }
+
+  for (const expected of expectedPairs) {
+    check(actualPairs.has(expected), `evidence: missing case ${expected}`);
+  }
+}
+
+if (errors.length > 0) {
+  console.error(`Skill runtime evidence validation failed with ${errors.length} error(s):`);
+  for (const error of errors) console.error(`- ${error}`);
+  process.exit(1);
+}
+
+console.log("Skill runtime evidence validation passed.");

package/skills/README.md

@@ -64,6 +64,8 @@ Skill 可以继续用 `SKILL.md` 表达操作方法，但涉及确定性工具
 | `aios-runtime` | Prompt、Context、Memory、MCP/Tool、RAG/GraphRAG 和多 Agent Runtime 设计。 |
 | `aios-compare` | 文档专业度对比：比较两份文档、两个版本或两个 AI 输出哪份更专业、更可复核、更适合交付。 |
 | `aios-prompt-compare` | 内部 Prompt / Skill 测试工具：仅开发者明确调用时，对同一输入分别评估弱提示词、便携强提示词和真实 Skill 触发结果，判断是否应沉淀为 Skill。 |
+| `aios-tender-write` | 工程标书 / 技术标生成与改写：基于招标要求、历史标书素材和用户初稿生成 Markdown 工作母版，并交回 `aios-commercial-tender` 复核。 |
+| `aios-scheme-write` | 专项施工方案生成与改写：基于方案初稿、历史方案素材、工程概况和专家意见生成 Markdown 工作母版，并交回 `aios-construction-scheme` 复核。 |
 
 工程业务管理技能包 (Engineering Project Management)：
 
@@ -72,10 +74,12 @@ Skill 可以继续用 `SKILL.md` 表达操作方法，但涉及确定性工具
 | Skill | 用途 |
 | --- | --- |
 | `aios-commercial-tender` | 工程招投标响应证据链，用于提取评分点、资格条件、废标风险、资料缺口和人工复核事项。 |
+| `aios-tender-write` | 工程标书 / 技术标生成与改写，用于生成目录、章节初稿、评分点响应内容和历史素材复用表。 |
 | `aios-commercial-contract` | 工程分包、采购和补充协议履约证据链，用于提取节点、责任边界、付款条件和合同资料缺口。 |
 | `aios-construction-daily` | 现场施工日报证据链，用于提取管理摘要、异常、问题台账、计划偏差和需补充确认事项。 |
 | `aios-construction-meeting` | 工程现场会议待办闭环，用于将会议讨论转化为责任人、期限、争议点和下次追踪清单。 |
 | `aios-commercial-variation` | 工程变更签证资料链审查，用于梳理联系单、纪要、图纸变更、合同流程和资料断点。 |
 | `aios-construction-scheme` | 专项施工方案证据链辅审，用于提取危险源、交底要点、规范核验点、计算书缺口和专家复核事项。 |
+| `aios-scheme-write` | 专项施工方案生成与改写，用于生成方案章节、工艺流程、危险源控制措施和交底材料初稿。 |
 
-工程业务管理 Skill 只处理建筑工程资料抽取、证据链整理、风险提示和人工复核分流，不替代法务、造价、监理、安全、项目经理、总工或专家签审。涉及规范、制度、结构计算、质量安全、金额、工期索赔或责任归属时，必须输出中文化的 `判断事项 / 证据 / 工具结果 / 处理建议`；没有工具或人工证据时只能标注 `需核验` 或 `转人工复核`。
+工程业务管理 Skill 只处理建筑工程资料抽取、生成初稿、证据链整理、风险提示和人工复核分流，不替代法务、造价、监理、安全、项目经理、总工或专家签审。涉及规范、制度、结构计算、质量安全、金额、工期索赔或责任归属时，必须输出中文化的 `判断事项 / 证据 / 工具结果 / 处理建议`；没有工具或人工证据时只能标注 `需核验` 或 `转人工复核`。写作型 Skill 必须使用 Markdown 工作母版，保留素材来源、复用判断、待补占位和审核门禁，不得把历史项目事实直接套入当前项目。

package/skills/aios/SKILL.md

@@ -31,11 +31,13 @@ description: ArchSight AIOS 总路由入口。用户只说“请用 AIOS 技能
 | 资料 / 任务线索 | 路由到 | 默认输出方向 |
 |---|---|---|
 | 招标文件、评分办法、技术标、商务标、资格条件、AI 技术标问题、废标风险 | `aios-commercial-tender` | 响应矩阵、评分点、资料清单、风险提示、人工复核事项 |
+| 标书编写、技术标生成、投标响应章节改写、历史标书素材复用、用户提供技术标初稿 | `aios-tender-write` | Markdown 工作母版、写作 brief、素材匹配表、章节初稿、审核门禁 |
 | 合同、协议、分包、采购、租赁、付款、结算、履约、违约、争议解决 | `aios-commercial-contract` | 合同基本事实、空白字段、履约节点、付款结算、责任边界、资料缺口 |
 | 项目日报、施工日报、周报素材、现场记录、材料进场、机械、劳务、进度 | `aios-construction-daily` | 管理摘要、问题台账、异常事项、模板质量诊断、待追踪事项 |
 | 会议纪要、例会、协调会、专题会、交底会、待办、责任人、截止时间 | `aios-construction-meeting` | 会议结论、待办闭环、责任线索、遗留问题、下次追踪 |
 | 变更、签证、联系单、洽商、索赔、停工窝工、图纸变更、工程量变化 | `aios-commercial-variation` | 资料链、断点、流程依据、疑点、复核分流 |
 | 专项施工方案、施工组织、危大工程、深基坑、高支模、脚手架、吊装、危险源、交底、专家论证、计算书 | `aios-construction-scheme` | 工程概况、关键工序、危险源、交底要点、规范 / 计算书 / 专家复核清单 |
+| 方案编写、方案生成、方案改写、历史方案素材复用、专家意见回写、施工方案初稿优化 | `aios-scheme-write` | Markdown 工作母版、写作 brief、素材匹配表、方案章节初稿、审核门禁 |
 | 结构计算、荷载、边界条件、挠度、稳定、FEM、结构求解器 | `aios-structural` | 结构力学输入检查、求解链路、转 Euclid 结构复核、人工签审边界 |
 | BIM、IFC、Revit、CAD、建筑规范、审图规则、知识结构化 | `aios-knowledge` | 行业语义、规范知识、资料来源、知识结构化边界 |
 | 用户明确调用 `aios-compare`，或明确要求用 AIOS 判断两份文档 / 两个版本 / 两个 AI 输出哪份更专业 | `aios-compare` | 对比范围、可比性判断、结构差异、内容差异、专业度评分、边界风险、合并建议 |
@@ -54,12 +56,15 @@ description: ArchSight AIOS 总路由入口。用户只说“请用 AIOS 技能
 2. 未指定时，根据文件名、标题、正文高频词、表格字段和用户任务判断资料类型。
 3. 命中单一高置信场景时，按对应 Skill 的 `SKILL.md` 工作流执行。
 4. 同一资料包含多个场景时，先输出资料类型判断，再按主任务路由；必要时列出次级 Skill 作为后续复核建议。
-5. 类型不清时，不要泛泛总结；先输出“资料状态判断”和“需要用户补充的资料类型 / 目标”。
-6. 不因为用户说“AIOS”就强行套建筑行业规则；非建筑任务优先说明 AIOS 不适用，并建议使用宿主工具通用能力。
+5. 当用户动词是“生成、编写、改写、优化初稿、套用历史素材”时，优先选择写作型 Skill；当用户动词是“审核、复核、检查、抽取风险、列缺口”时，优先选择审核型 Skill。
+6. 类型不清时，不要泛泛总结；先输出“资料状态判断”和“需要用户补充的资料类型 / 目标”。
+7. 不因为用户说“AIOS”就强行套建筑行业规则；非建筑任务优先说明 AIOS 不适用，并建议使用宿主工具通用能力。
 
 ## 输出要求
 
-默认输出应包括：
+当用户只说“用 AIOS 分析该文档 / 用 AIOS 技能包看一下”且没有明确要求摘要时，默认输出“标准详版报告”，不是短摘要。路由判断可以简短，但专项分析必须展开到可复核的表格、清单或台账。
+
+标准详版报告必须包括：
 
 1. 资料类型和路由判断。
 2. 资料来源清单或资料来源说明。
@@ -67,9 +72,16 @@ description: ArchSight AIOS 总路由入口。用户只说“请用 AIOS 技能
 4. 资料缺口和需补充确认事项。
 5. 人工复核岗位。
 6. 不能由 AI 直接下结论的事项。
+7. 输出自检。
 
 如果已路由到专项 Skill，应遵守该 Skill 的输出格式；本 Skill 不要求额外生成一份冗长的路由报告。
 
+最小详度要求：
+
+- 资料足够时，主输出表 / 清单 / 台账不应少于 3 条可定位事项；资料不足时，列出全部可定位事项，并说明不足原因。
+- 不得只输出几段概括性文字；必须把关键事项拆成表格、清单或台账字段。
+- 输出自检必须逐项确认是否包含资料来源、主分析表 / 清单 / 台账、资料缺口、人工复核岗位、AI 不应下结论事项；缺一项时先补齐再结束。
+
 ## 约束
 
 - 不输出最终法律意见、合规结论、安全结论、结构计算结论、结算金额或责任归属。

package/skills/aios/agents/openai.yaml

@@ -1,4 +1,4 @@
 interface:
   display_name: "AIOS Router"
   short_description: "按资料类型自动路由到对应 ArchSight AIOS 技能"
-  default_prompt: "当用户只说使用 AIOS 或 AIOS 技能包分析文档时，先识别资料类型，再路由到合适的 aios-* 技能；不确定时先做资料状态判断，保留证据链、人工复核和不能下结论的边界。"
+  default_prompt: "当用户只说使用 AIOS 或 AIOS 技能包分析文档时，先识别资料类型，再路由到合适的 aios-* 技能；默认输出标准详版报告，不要压缩成摘要；必须保留资料来源、主分析表或台账、资料缺口、人工复核、AI 不应下结论事项和输出自检。"

package/skills/aios-commercial-contract/SKILL.md

@@ -46,6 +46,16 @@ description: 工程合同履约证据链工作流。用于从工程分包、采
 3. 所有金额、期限、责任方、验收、付款、违约条款必须带原文依据位置或原文关键词。
 4. 输出只作为履约管理辅助，不构成法律意见、违约定性、索赔或结算结论。
 
+## 标准详版报告与输出自检
+
+短指令触发本 Skill 时，默认输出标准详版报告，不要压缩成摘要。标准详版报告必须包含：资料来源清单、主分析表 / 清单 / 台账、资料缺口、人工复核岗位、AI 不应下结论事项、输出自检。
+
+最小详度要求：
+
+- 资料足够时，履约节点表、付款与结算条件或责任边界风险提示至少展开 3 条可定位事项；资料不足时，列出全部可定位事项并说明不足原因。
+- 主分析表必须保留“节点 / 条款或事项 / 证据 / 缺口 / 处理建议 / 人工复核岗位”等字段，不得只写概括段落。
+- 输出自检必须逐项确认是否包含资料来源、主分析表 / 清单 / 台账、资料缺口、人工复核岗位、AI 不应下结论事项；缺一项时先补齐再结束。
+
 ## Capability 与证据
 
 - 所有抽取结论必须带 `证据`，至少包括来源文件、页码/章节/行号或原文短摘。

package/skills/aios-commercial-contract/agents/openai.yaml

@@ -1,4 +1,4 @@
 interface:
   display_name: "AIOS Contract"
   short_description: "整理工程合同履约节点、责任边界和证据链风险"
-  default_prompt: "当任务涉及建筑工程合同、分包、采购、租赁或补充协议履约资料时，使用该技能抽取履约节点、付款条件、责任边界、原文证据和人工复核事项；不得输出最终法律或结算结论。"
+  default_prompt: "当任务涉及建筑工程合同、分包、采购、租赁或补充协议履约资料时，使用该技能抽取履约节点、付款条件、责任边界、原文证据和人工复核事项；默认输出标准详版报告，不要压缩成摘要；必须包含资料来源、主分析表或台账、资料缺口、人工复核、AI 不应下结论事项和输出自检；不得输出最终法律或结算结论。"

package/skills/aios-commercial-tender/SKILL.md

@@ -46,6 +46,16 @@ description: 工程招投标响应证据链工作流。用于从招标文件、
 3. 输出 AI 技术标复核问题回应矩阵、评分点响应矩阵、资料缺口和人工复核事项。
 4. 废标、资格条件和评分分值缺证据时标为 `需核验`，不得补猜。
 
+## 标准详版报告与输出自检
+
+短指令触发本 Skill 时，默认输出标准详版报告，不要压缩成摘要。标准详版报告必须包含：资料来源清单、主分析表 / 清单 / 台账、资料缺口、人工复核岗位、AI 不应下结论事项、输出自检。
+
+最小详度要求：
+
+- 资料足够时，废标 / 否决风险清单、资格条件清单或评分点响应矩阵至少展开 3 条可定位事项；资料不足时，列出全部可定位事项并说明不足原因。
+- 主分析表必须保留“事项 / 要求 / 证据 / 缺口 / 处理建议 / 人工复核岗位”等字段，不得只写概括段落。
+- 输出自检必须逐项确认是否包含资料来源、主分析表 / 清单 / 台账、资料缺口、人工复核岗位、AI 不应下结论事项；缺一项时先补齐再结束。
+
 ## Capability 与证据
 
 - 每个废标风险、评分点和资格条件必须带原文定位。

package/skills/aios-commercial-tender/agents/openai.yaml

@@ -1,4 +1,4 @@
 interface:
   display_name: "AIOS Tender"
   short_description: "解析工程招投标文件、评分点和响应矩阵"
-  default_prompt: "当任务涉及建筑工程招标文件、评分办法或投标响应准备时，使用该技能抽取废标风险、资格条件、评分点、资料缺口和人工复核事项；不得输出中标概率或最终投标决策。"
+  default_prompt: "当任务涉及建筑工程招标文件、评分办法或投标响应准备时，使用该技能抽取废标风险、资格条件、评分点、资料缺口和人工复核事项；默认输出标准详版报告，不要压缩成摘要；必须包含资料来源、主分析表或台账、资料缺口、人工复核、AI 不应下结论事项和输出自检；不得输出中标概率或最终投标决策。"

package/skills/aios-commercial-variation/SKILL.md

@@ -45,6 +45,16 @@ description: 工程变更签证证据链审查工作流。用于整理联系单
 3. 公开样表只能提供字段结构，不代表项目事实。
 4. 不判断最终金额、责任归属、签证成立或索赔成立。
 
+## 标准详版报告与输出自检
+
+短指令触发本 Skill 时，默认输出标准详版报告，不要压缩成摘要。标准详版报告必须包含：资料来源清单、主分析表 / 清单 / 台账、资料缺口、人工复核岗位、AI 不应下结论事项、输出自检。
+
+最小详度要求：
+
+- 资料足够时，资料链完整性清单、合同和流程依据对照或资料断点清单至少展开 3 条可定位事项；资料不足时，列出全部可定位事项并说明不足原因。
+- 主分析表必须保留“节点 / 资料 / 签发或签收 / 状态 / 证据 / 缺口 / 处理建议 / 人工复核岗位”等字段，不得只写概括段落。
+- 输出自检必须逐项确认是否包含资料来源、主分析表 / 清单 / 台账、资料缺口、人工复核岗位、AI 不应下结论事项；缺一项时先补齐再结束。
+
 ## Capability 与证据
 
 - 每个事实链节点必须带 `证据`：来源文件、日期、编号、页码/章节/行号和签认状态。