| name | six-writing-cats |
| description | Develops self-media long-form articles, opinion essays, newsletters, blogs, and columns through Socratic inquiry, six independent writing-cat subagents, public fact checking, article starter packs, and draft skeletons. Use when users ask for AI-assisted writing, topic expansion, writing angles, article structure, public fact support, multi-perspective drafts, self-media posts, deep commentary, or Chinese/English long-form writing. Triggers include AI协作写作、写作猫、六只写作猫、六只思考猫写作、选题、文章选题、写作思路、写作角度、文章结构、自媒体长文、观点文章、深度写作、帮我写文章、帮我拓展思路、帮我找资料、帮我生成初稿、thinking cats for writing、AI writing partner。 |
| argument-hint | 一句话描述你想写的主题或文章任务，例如：我想写一篇关于 AI 时代普通写作者还值不值得坚持写长文的文章 |
六只写作猫 Six Writing Cats
用六只彼此独立的写作猫，把一个写作主题拆成事实、价值、风险、情绪、创意和组织方式，最终给作者一份可二次创作的文章起点包与草稿骨架。不要把它当“一键成稿器”；它的价值是先把材料、判断和结构摊开。
When to Use
- 自媒体长文、观点文章、博客、Newsletter、专栏、深度评论。
- 用户需要选题拓展、写作角度、文章结构、事实底座或反方压力测试。
- 用户希望先通过追问澄清真实关切，再生成文章起点包。
- 用户明确提到六只写作猫、六只思考猫写作、AI 协作写作、帮我拓展思路、帮我生成初稿。
Reference Map
先读通用规约，再按阶段读单猫文件；所有 reference 都直接从本文件链接，避免深层跳转。
Workflow Checklist
For substantial runs, track this checklist and update it as work progresses:
Writing Cats Progress:
- [ ] Capture the writing topic or article task
- [ ] Think through the writing scene internally
- [ ] Run 🔵 蓝猫启动
- [ ] Collect missing user input through Socratic inquiry
- [ ] Build one shared writing brief
- [ ] Run ⚪ 🟡 ⚫ 🔴 🟢 cats in parallel
- [ ] Run 🔵 蓝猫收束
- [ ] Write the local Markdown archive
- [ ] Reply with a concise summary and archive path
1. Capture the Task
If the user has not stated a writing topic or article task, ask one question to get it, then stop and wait. Do not analyze before the topic exists.
2. Internal Scene Thinking
Before asking follow-ups, identify the article-specific issue, likely readers, publishing context, needed facts, author material, emotional stakes, and expression risks. Use this to avoid template questions.
3. Blue Start
Run 🔵 蓝猫启动 after reading the subagent contract and blue-cat reference. It only defines the writing task, reader/publishing unknowns, article type, and prioritized missing information. It does not ask the user directly.
4. Socratic Inquiry
The parent agent asks all user-facing questions. Ask at most two questions per turn; if the turn contains questions, end after asking and wait for the real reply.
Use the runtime user-input tool when available. If that tool is missing, unavailable in the current mode, or fails, immediately fall back to numbered plain-text questions. If the user answers a choice question with extra free text, absorb the real answer into the brief instead of forcing a rigid option format.
5. Shared Writing Brief
Before starting the five parallel cats, prepare one shared brief containing:
- writing topic and core thesis
- target reader and publishing context
- writing intent
- user-confirmed views, material, emotion, and hidden concerns
- known facts, assumptions, unknowns, and public facts to verify
- expression directions to avoid
- remaining gaps that still matter
Do not invent user experiences, preferences, evidence, examples, or emotional motives. Mark unconfirmed items as unknown or assumed.
6. Five Cats in Parallel
Run ⚪ 白猫, 🟡 黄猫, ⚫ 黑猫, 🔴 红猫, and 🟢 绿猫 as separate subagents using the same shared brief and inherited parent model. They do not ask the user, read each other, or wait for more input.
- ⚪ 白猫 is the only cat allowed to verify public facts at runtime. It must prefer opened, official or primary sources; search snippets are only leads.
- 🟡 黄猫 covers reader value and propagation value.
- ⚫ 黑猫 covers argument risks, fact risks, opposing views, and misreadings.
- 🔴 红猫 covers author concern, emotional tension, personal entry points, and human texture. Parent must collect at least one red-cat input before this stage.
- 🟢 绿猫 covers new angles, structure variants, title directions, analogies, and counterintuitive cuts.
If a subagent mixes another cat's role, asks the user, invents user facts, or uses a different model, discard and rerun that result.
7. Blue Synthesis
Only 🔵 蓝猫收束 may integrate the five cats. It must generate the 文章起点包, recommended structure, 草稿骨架, fact-use guidance, second-draft suggestions, and the first next writing action.
If subagent capacity blocks blue synthesis, close completed subagents only after saving their outputs, then retry. If it still cannot start, stop and report the runtime blocker; do not let the parent agent impersonate blue synthesis.
8. Archive and Reply
Write the complete archive to <WORKDIR>/writing-drafts/YYYY-MM-DD-主题短名.md. <WORKDIR> means the current working directory; do not assume an environment variable exists.
The archive must include:
- topic, target reader, publishing context
- six-cat insights with fixed visual labels
- 文章起点包
- 草稿骨架 clearly marked as not publish-ready
- next first writing action
系统思考总结： in 2-4 concise sentences
The chat reply may be shorter than the archive, but must include the archive path, six-cat conclusion summary, core thesis, and next first writing action. If writing the archive fails, the chat reply must contain the full deliverable.
Non-Negotiables
- Parent agent is the only user-facing speaker.
- No five-cat stage before the shared brief is complete.
- All subagents must use the parent model or inherited model path; never silently downgrade.
- White cat is the only runtime fact-checking cat; other cats use only the shared brief.
- Blue synthesis is the only cross-cat integration point.
- Final output is a 文章起点包 and 草稿骨架, not a guaranteed publish-ready article.
- Final deliverable does not require a separate visual artifact.
