Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Novel Workshop

v2.0.0

多模型命题小说创作工坊。用户给出写作命题,自动完成:AI 写初稿 → 三路并行审阅(逻辑/文学/锐评)→ AI 改稿 → 飞书文档完整存档。 一键启动,全程自动,零手动干预。支持飞书实时进度推送。 触发词:命题写作、写一篇小说、命题小说、创作工坊、novel workshop

4· 647·3 current·3 all-time
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
high confidence
!
Purpose & Capability
The skill's stated purpose (multi-model novel workshop + Feishu storage) matches the code's behavior (calls OpenRouter models and Feishu APIs). However the registry metadata declares no required environment variables or config paths while SKILL.md and workflow.py require an OpenRouter API key and Feishu app credentials in ~/.openclaw/openclaw.json (or via env vars). That mismatch between declared requirements and actual needs is incoherent and potentially misleading.
!
Instruction Scope
SKILL.md and workflow.py instruct the agent to: read ~/.openclaw/openclaw.json, call OpenRouter chat completions (sending full user prompt and generated drafts), create/write Feishu documents, and write full archives to ~/.openclaw/workspace. The instructions send entire initial prompts, drafts, reviews and final text to external services (OpenRouter and Feishu) without redaction. This is expected for the feature but is broader than the manifest indicates and has privacy/exfiltration implications.
Install Mechanism
There is no install spec; the skill is instruction + a Python script that uses only standard libs and requests. No external binaries or remote downloads are performed by an installer, which is low-risk from an installation mechanism perspective.
!
Credentials
The code requires sensitive credentials (OpenRouter API key and Feishu appId/appSecret) and reads a config file at ~/.openclaw/openclaw.json, but the registry lists no required env vars or config paths. Additionally, workflow.py defines hardcoded defaults for FEISHU_CHAT_ID, FEISHU_FOLDER_TOKEN and FEISHU_OWNER_OPEN_ID — these defaults could cause messages/doc writes to unintended recipients/locations if not overridden. The skill therefore requests more credential/config access than the manifest declares.
Persistence & Privilege
The skill is not always-enabled and does not request special platform privileges. It writes files under ~/.openclaw/workspace (its own area) and can make outbound network calls. Autonomous invocation is allowed (platform default) — combined with the credential mismatches above this increases blast radius but on its own is expected.
What to consider before installing
Key things to check before installing or running: - The manifest omits required credentials/config; SKILL.md and workflow.py expect an OpenRouter API key and Feishu appId/appSecret in ~/.openclaw/openclaw.json (alternatively set OPENROUTER_API_KEY and FEISHU_* env vars). Provide your own credentials only. - Review and (if needed) override the hardcoded FEISHU_CHAT_ID, FEISHU_FOLDER_TOKEN, and FEISHU_OWNER_OPEN_ID defaults to avoid sending content to an unintended chat/folder. Ensure FEISHU_CHAT_ID matches a destination you control. - Understand privacy impact: the script sends the full user prompt, the full draft, three full reviews, and the final manuscript to external services (OpenRouter, and Feishu). Do not use with sensitive or private content unless you accept that flow. - Inspect the remaining functions (create_feishu_doc, write_feishu_doc_content) in the truncated portion of workflow.py to confirm no hidden endpoints or surprising behavior. Consider running with test/dummy prompts first. - Because the registry metadata is inconsistent, treat this skill as untrusted until you verify configuration and endpoints; consider running in an isolated account or environment and limiting credentials to least privilege.

Like a lobster shell, security has layers — review code before you run it.

latestvk97a17haph8kf0c46hdv0w5tvd82ajhp
647downloads
4stars
1versions
Updated 21h ago
v2.0.0
MIT-0

🎲 命题小说多模型创作工坊

工作流

用户命题 → MiMo 写初稿 → Gemini+Claude 三路并行审阅 → Gemini 改稿 → 飞书文档完整存档

全流程约 2-5 分钟,飞书群聊实时进度推送。

使用方式

用户给出写作命题后,执行脚本:

python3 skills/novel-workshop/workflow.py "用户的原始命题" "文档标题"

重要

  • 第一个参数是用户的原始命题原话,不要修改、不要补充解释
  • 第二个参数是飞书文档标题(简短的标识名)
  • 脚本后台运行,自动推送进度到飞书群聊
  • 脚本完成后输出 __SUMMARY__: JSON,包含文档链接和评分

前置要求

  1. OpenRouter API Key:在 openclaw.jsonmodels.providers.openrouter.apiKey 中配置
  2. 飞书应用:在 openclaw.jsonchannels.feishu 中配置 appIdappSecret
  3. Python 依赖requests(通常已预装)
  4. 模型注册(在 openclaw.jsonmodels.registered 中添加):
    • xiaomi/mimo-v2-flash(写初稿)
    • google/gemini-2.5-pro(审阅+改稿)
    • anthropic/claude-opus-4.6(锐评)

环境变量(可选覆盖)

变量说明默认值
FEISHU_CHAT_ID飞书群聊 ID(进度推送目标)从 openclaw.json 读取
FEISHU_FOLDER_TOKEN飞书文件夹 token(文档创建位置)无(创建到根目录)
FEISHU_OWNER_OPEN_ID文档所有者 open_id(自动授权)
OPENROUTER_API_KEYOpenRouter API Key从 openclaw.json 读取

输出

本地文件

~/.openclaw/workspace/novels/{标题}.md — 完整 markdown(初稿+审阅+终稿)

飞书文档

自动创建到指定文件夹,包含完整内容:

  • Part 1:初稿全文
  • Part 2:三路审阅全文(逻辑检阅 + 文学性分析 + 锐评)
  • Part 3:终稿全文

进度消息(飞书群聊)

[░░░░░] 0/5 收到命题!工作流启动中 🎲
[█░░░░] 1/5 初稿完成 ✅《标题》(N字) 三路审阅启动中…
[██░░░] 2/5 审阅完成 ✅ 逻辑 X/10 | 文学 X/10 | 改稿启动中…
[███░░] 3/5 改稿完成 ✅ 保存中…
[████░] 4/5 存档完成 ✅ 正在写入飞书文档…
[█████] 5/5 全部完成!🎲 链接 + 评分

模型角色

角色默认模型备选
初稿写作xiaomi/mimo-v2-flashgoogle/gemini-2.5-pro
逻辑审阅google/gemini-2.5-proanthropic/claude-opus-4.6
文学审阅google/gemini-2.5-proanthropic/claude-opus-4.6
锐评anthropic/claude-opus-4.6google/gemini-2.5-pro
改稿google/gemini-2.5-proanthropic/claude-opus-4.6

⚠️ 铁律

  1. Prompt 原封不动:用户怎么说的就怎么传给模型,不要添加解释或修改
  2. 飞书文档不省略任何内容:初稿全文、审阅全文、终稿全文,一个字都不能少
  3. 改稿不用写初稿的模型:避免"自己改自己"导致的保守倾向
  4. 全程自动:用户只需给命题,不需要说"继续"

Comments

Loading comments...