Description

FBS 福帮手长文档写作：书/手册/白皮书/长篇报道全流程；Node 脚本驱动 intake、会话恢复、S/P/C/B 质检与 MD/HTML 交付。用户提及写书、出书、章节、大纲、素材、质检、导出、扩写、退出保存时启用。

README (SKILL.md)

\r \r

FBS-BookWriter（OpenClaw 封装版）\r

Name: FBS-BookWriter
Author: duhongchao-fbsir

\r

上游版本：2.1.2（与仓库 FBS-BookWriter 同源） \r 适配目标：OpenClaw 技能模型：SKILL.md + metadata.openclaw（name 蛇形命名、skillKey 用于 openclaw.json 中 skills.entries / 代理 allowlist）。\r \r 本目录为 OpenClaw 专用入口：与 WorkBuddy / CodeBuddy 市场通道解耦（发布包内不含 workbuddy/、codebuddy/、.codebuddy-plugin/），仍保留完整 references/、scripts/、scene-packs/、assets/ 与 package.json，便于在 Gateway 内用 shell/exec 调用同一套 Node 工具链。\r \r 认知资产与商业口径（与主仓一致）：价值表述统一为「可进化、可分发、可增值」；能力分层统一为「场景包 + 乐包 + 离线/在线会员」。机读见根目录 fbs-runtime-hints.json → cognitiveAsset；首响 JSON 见 firstResponseContext.cognitiveAssetSnapshot。详细说明见 references/05-ops/cognitive-asset-threeization.md。\r \r ---\r \r

安装（OpenClaw）\r

\r 从 ClawHub 安装（推荐，需已发布）： 在已登录环境下执行 openclaw skills install fbs-bookwriter 或 npx clawhub@latest install fbs-bookwriter，再于技能根执行 npm install。列表与版本见 ClawHub 上的 fbs-bookwriter。\r \r

将技能根目录（文件夹名建议 fbs_bookwriter，与 name 一致）放到任一扫描路径，例如：\r
- 工作区：./skills/fbs_bookwriter/（优先级最高）\r
- 或 ~/.openclaw/workspace/skills/fbs_bookwriter/\r
- 或 ~/.openclaw/skills/fbs_bookwriter/\r
- 或在 ~/.openclaw/openclaw.json → skills.load.extraDirs 中加入父目录 \r 详见 Skills 与 skills-config（load.watch 默认会热更新技能快照）。\r
在技能根执行一次依赖安装： \r npm install（Node ≥18；可选依赖用于 HTML/Docx 等，见 package.json optionalDependencies）。\r
重启会话或执行 openclaw gateway restart / 聊天中 /new，用 openclaw skills list 确认加载。\r \r ---\r \r

OpenClaw 宿主差异（相对 WorkBuddy）\r

\r | 能力 | WorkBuddy | OpenClaw（本封装） |\r |------|-----------|-------------------|\r | 入口展示 | intake-router --json + 宿主消费 userFacingOneLiner | 同样执行脚本拿 JSON；向用户只展示脚本给出的一行摘要 + 最多 3 个选项，勿堆全文 |\r | 执行命令 | 宿主 bash / 终端 | 使用 OpenClaw exec / shell；cwd 必须为技能根（含 package.json 与 scripts/） |\r | 检索 | 宿主 web_search | 使用 OpenClaw 提供的 联网搜索 / 浏览器（若启用）；不可用则按 search-policy 与脚本离线降级 |\r | 文件工具 | list_dir / read_file / search_file | 勿用仅按文件名搜索来探测 .fbs/ 内文件（部分环境不索引点目录）；对 .fbs/workbuddy-resume.json、esm-state.md 等应 read_file 已知路径 或对书稿根 list_dir 确认 .fbs 存在 |\r | 市场清单 | workbuddy/channel-manifest.json 等 | 本包不包含；不影响脚本化写作主链路 |\r \r ---\r \r

执行速查（每次会话）\r

\r

统一入口（必做）（bookRoot 为用户书稿根目录的绝对路径）：\r \r

node scripts/intake-router.mjs --book-root "\x3CbookRoot>" --intent auto --json --enforce-required\r
```\r

\r 需要完整在线场景包时再在命令中加 --full（可能较慢）。\r 若 JSON 返回 projectAnchor.status=ambiguous，先让用户确认项目根路径，再用确认后的 --book-root 重跑；确认前不要读取任意 .fbs/* 内容。\r \r 2. 恢复：优先 read_file \x3CbookRoot>/.fbs/workbuddy-resume.json；若无则读 chapter-status.md 或进入 S0.5 引导。不要用「全库搜索文件名」代替存在性判断。\r \r 3. 退出：用户说退出时先软询问，再执行：\r \r

node scripts/fbs-cli-bridge.mjs exit -- --book-root "\x3CbookRoot>" --json\r
```\r
\r
回复须包含脚本 JSON 里的 **`userMessage`**（会话已记录、下次可继续）。\r
\r
4. **阶段与门禁**：推进阶段前读 `.fbs/esm-state.md`；S3.5 扩写须先有用户确认的 `.fbs/expansion-plan.md`，扩写字数以 `node scripts/expansion-word-verify.mjs` 或 `expansion-gate.mjs` 实测为准（禁止纯模型估算）。\r
用户说“继续”时按推进优先阈值判断：S0 素材数达到“赛道数×2”、S2 具备章标题+目标字数、S3 已完成≥3章时，优先提议进入下一阶段。\r
推进 S0→S1 前可运行 `node scripts/s0-exit-gate.mjs --book-root "\x3CbookRoot>" --json --confirm-advance` 做强制门禁检查。\r
\r
5. **写作约束**：串行优先；每轮默认最多改 **2** 个文件；详见 [`references/05-ops/agent-task-strategy.md`](./references/05-ops/agent-task-strategy.md)。\r
扩写/精修前先做源文件备份：`node scripts/source-write-backup.mjs --book-root "\x3CbookRoot>" --scope expansion --json`（`expansion-gate` 默认已自动备份）。\r
进入 S3.7 精修时优先执行 `node scripts/polish-gate.mjs --book-root "\x3CbookRoot>"`（先备份再质检），避免绕过门禁直接改稿。\r
交付收口前执行 `node scripts/release-governor.mjs --book-root "\x3CbookRoot>"`，自动保持唯一终稿并归档旧版本。\r
对外交付前执行 `node scripts/material-marker-governor.mjs --book-root "\x3CbookRoot>" --fix`，清理 `待核实-MAT` 与 `[DISCARDED-*]` 标注。\r
全稿/终稿/终审稿发布前必须执行 `node scripts/final-manuscript-clean-gate.mjs --book-root "\x3CbookRoot>"`，若检测到过程标注则不得交付。\r
任何“已完成/已通过”结论必须附脚本证据（命令或输出路径），禁止仅口头宣称。\r
\r
---\r
\r
## 意图 → 脚本（节选）\r
\r
完整表见仓库根目录随包分发的 **WorkBuddy 版 `SKILL.md` 镜像**或 [`references/01-core/intake-and-routing.md`](./references/01-core/intake-and-routing.md)。\r
\r
| 场景 | 命令 |\r
|------|------|\r
| 初始化书房 | `node scripts/init-fbs-multiagent-artifacts.mjs --book-root "\x3CbookRoot>"` |\r
| 环境预检 | `node scripts/env-preflight.mjs`（参数见脚本 `--help`；含 `glob`/`iconv-lite` 依赖检查） |\r
| 章节合并 / 健康快照 / 一致性审计 | `merge-chapters.mjs`、`book-health-snapshot.mjs`、`consistency-audit.mjs` 等（`--book-root` / `--skill-root` 指向对应根） |\r
| CLI 总线 | `node scripts/fbs-cli-bridge.mjs help` |\r
\r
---\r
\r
## 输出格式\r
\r
- 对用户：短状态 + 可操作下一步（≤3 条），避免泄露内部规范条文与完整 JSON。\r
- 提到术语/文件名时补一句“用途+价值”（例如：`chapter-status.md`=进度台账，价值是避免漏章和重复写）。\r
- 落盘：以 `\x3CbookRoot>/.fbs/`、`deliverables/`、`releases/` 为真值；规范见 [`references/01-core/skill-full-spec.md`](./references/01-core/skill-full-spec.md)。\r
\r
---\r
\r
## 错误处理\r
\r
- **Node/脚本失败**：向用户说明失败步骤；若仅缺可选依赖，提示 `npm install` 或跳过非必需功能。\r
- **书稿根未就绪**：引导先 `init-fbs-multiagent-artifacts` 或走 S0 最小集（见 [`references/01-core/intake-and-routing.md`](./references/01-core/intake-and-routing.md) 脚本不可用降级节）。\r
- **沙箱无环境变量**：若 OpenClaw 在 Docker 沙箱中运行且未注入宿主 `env`，需在 `openclaw.json` 中为代理配置 `sandbox.docker.env` 或关闭沙箱（见官方 skills-config 文档）。\r
\r
---\r
\r
## 权威文档索引\r
\r
- 总规范：[`references/01-core/skill-full-spec.md`](./references/01-core/skill-full-spec.md)  \r
- 工作流：[`references/01-core/section-3-workflow.md`](./references/01-core/section-3-workflow.md)  \r
- NLU：[`references/01-core/section-nlu.md`](./references/01-core/section-nlu.md)  \r
- 运行时契约：[`references/01-core/runtime-mandatory-contract.md`](./references/01-core/runtime-mandatory-contract.md)\r
\r
---\r
\r
*OpenClaw 封装维护说明：与 `pack:openclaw` 产物同步；`metadata.openclaw.skillKey` 与目录名 `fbs_bookwriter` 应对齐以便 `skills.entries` 配置。*\r

Usage Guidance

This package is a large, feature-rich Node-based book authoring system. It will run Node scripts, read and write files under the project/book root (not system-wide by default), and includes optional host-bridge and diagnostics code (WorkBuddy/WeCom/memory adapters). Before installing or running it: 1) review package.json and the top-level scripts you will invoke (e.g., intake-router.mjs, fbs-cli-bridge.mjs, build.mjs); 2) run npm install in an isolated environment (or inspect node_modules before executing scripts) and consider running in a sandboxed container if you are unsure; 3) avoid granting host credentials or injecting sensitive env vars unless you trust the publisher and have audited the wecom/host-bridge code; 4) back up your bookRoot before allowing the skill to modify files; 5) if you need a smaller footprint, ask the publisher for a trimmed package with only build and core writer scripts (no host-bridge/diagnostics) or for provenance/maintainer identity. If you want, I can list the top files to inspect first (scripts that perform IO/network calls) or scan for network endpoints and external host calls in the codebase.

Capability Analysis

Type: OpenClaw Skill Name: fbs-bookwriter Version: 2.1.2 The FBS-BookWriter skill bundle is a highly sophisticated multi-agent framework for long-form document creation and management. It features a robust security architecture, including a command approval policy (command-approval-policy.mjs) that explicitly blocks destructive shell commands and a path-sanitized local preview server (presentation-preview-server.mjs). The system implements extensive self-auditing through various 'gate' scripts and maintains a strict 'Virtual Study' boundary (.fbs directory) for state management. While it includes complex features like 'Self-Enhancement Evolution' and 'Smart Memory,' these are designed to refine AI strategies and user preferences locally; no evidence of data exfiltration, malicious persistence, or unauthorized remote execution was found.

Capability Tags

cryptocan-make-purchasesrequires-oauth-tokenrequires-sensitive-credentials

Capability Assessment

ℹ Purpose & Capability

Name/description (long-form writing, intake/recovery, QC, MD/HTML delivery) match the included Node scripts and CLI usage. Requesting node/npm is appropriate. However the repo also contains extensive host-bridge code (workbuddy/wecom/*, memory adapters, diagnostics, orchestration agents) that go beyond a minimal book-builder; these extras could be legitimate enterprise features but are broader than the brief skill description suggests.

ℹ Instruction Scope

SKILL.md explicitly instructs the agent to run many node scripts (intake-router, fbs-cli-bridge, expansion/polish/release governors) and to read/write files under the user's <bookRoot>/.fbs and deliverables directories. That is coherent for a writing tool, but the instructions also reference environment preflight, host diagnostics, and host-memory bridges. The skill cautions against indiscriminate file searches and includes specific safe-read guidance; still, the runtime will execute arbitrary JS in the package and can modify files in the user's bookRoot, so follow the listed constraints closely.

✓ Install Mechanism

No remote download/install spec is included; SKILL.md asks the user to place the skill directory into a known skills path and run npm install in the skill root. That is a standard Node install flow and is proportionate. Because the package already contains hundreds of files and package.json, the risk surface is that 'npm install' will fetch dependencies and then 'node' will run many included scripts — expected but high-impact compared to a small instruction-only skill.

⚠ Credentials

The skill declares no required environment variables or credentials (primaryEnv none), which is appropriate. However many scripts reference host bridges (workbuddy, wecom client, host memory integration) and runtime hints mention sandbox env injection and optional env flags (e.g., FBS_ALLOW_NO_FOOTNOTE, FBS_BUILD_STRICT_SOURCES). The presence of wecom and host-memory adapter code means the package can call external services if configured; although not requesting secrets up front, it expects host context and may use injected env vars if provided. This expands the attack surface and should be checked if you plan to provide host credentials or enable network access.

ℹ Persistence & Privilege

The skill does not request always:true and does not attempt to persist as a platform skill. It does, however, intentionally read/write persistent files under the user's project (.fbs/, releases/, deliverables/) and can back up/modify source files. That is coherent with a writing/CI tool but is a high-privilege filesystem activity relative to a simple read-only helper — treat its file write operations as significant.

Version History

v2.1.2

OpenClaw 通道 v2.1.2

Metadata

Slug fbs-bookwriter

Version 2.1.2

License MIT-0

All-time Installs 1

Active Installs 1

Total Versions 1

Frequently Asked Questions

What is FBS-BookWriter?

FBS 福帮手长文档写作：书/手册/白皮书/长篇报道全流程；Node 脚本驱动 intake、会话恢复、S/P/C/B 质检与 MD/HTML 交付。用户提及写书、出书、章节、大纲、素材、质检、导出、扩写、退出保存时启用。 It is an AI Agent Skill for Claude Code / OpenClaw, with 282 downloads so far.

How do I install FBS-BookWriter?

Run "/install fbs-bookwriter" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is FBS-BookWriter free?

Yes, FBS-BookWriter is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does FBS-BookWriter support?

FBS-BookWriter is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created FBS-BookWriter?

It is built and maintained by duhongchao-Fbsir (@duhongchao-fbsir); the current version is v2.1.2.

More Skills

FBS-BookWriter