kol-writer

kol-writer: Self-media outline structuring and corpus management

⚠️ Built-in sub-skills loading paths This skill contains three dependent sub-skills, stored locally in the sub-skills/ directory:

• deep-research ──→ located in sub-skills/deep-research/
• markdown-new ──→ located in sub-skills/markdown-new/
• agent-browser ──→ located in sub-skills/agent-browser/

Must and can only read its sub-skills/ from the relative path SKILL.md, strictly prohibited from accessing the global skill root directory to prevent conflicts.

This skill is used to assist creators in organizing scattered ideas, raw materials, and retrieved information into structured outlines, and archiving personal writing assets.

1. Core process: outline structuring and iteration

graph TD
    AI_Ask["1. 询问基本信息<br>(主题/平台/风格/目标)"] --> U_Answer["2. 用户提供信息"]
    U_Answer --> AI_Init["3. 初始化工作区<br>(创建目录并保存原始输入)"]
    AI_Init --> AI_Note["4. 对话式观点速记<br>(仅记录事实，不做推理)"]
    AI_Note --> AI_Eval["5. 结构化大纲梳理<br>(输出初版大纲并标注调查项)"]
    AI_Eval --> AI_Research["6. 补充检索与回填<br>(针对大纲缺失信息执行检索)"]
    AI_Research --> Loop_Iter{"7. 对话修改与调优<br>(反馈微调)"}
    Loop_Iter -->|未完善| Loop_Iter
    Loop_Iter -->|用户确认| AI_Write["8. 终稿归档<br>(写入 03-大纲终稿.md)"]

1.1 Phase 1: Initialize workspace

• 1. Ask for basic information: The Agent proactively confirms five dimensions with the user: topic, publishing platform, writing style, expected goals, and reference samples.
• 2. User provides information: The user provides input in response to the questions.
• 3. Initialize workspace: Create an independent writing working directory and save the input as 01-用户原始输入.md. See details:01_initialize_project.md

1.2 Phase 2: Quick note of viewpoints

• 4. Quick note of viewpoints: Record the user's viewpoints in dialogue form. The Agent must save the user input to 02-用户想法记录.md as material for outline organization and style imitation. See details:02_quick_notes.md

1.3 Phase 3: Outline organization and retrieval

• 5. Structured outline organization: After the quick notes are completed, the Agent retrieves and calls the corresponding platform specifications in rules/platforms.md , combines existing ideas to refine arguments, and outputs a first draft outline. The outline must use quote blocks (>) to mark items to be investigated or supplemented. See details:03_understanding_prompt.md
• 6. Supplementary retrieval and backfilling: Develop a retrieval plan for the items marked in the outline as pending investigation.After waiting for the user to confirm the plan and select an option, run the retrieval tool to fill the results into the outline. See details:04_deep_research.md

1.4 Phase 4: Outline fine-tuning

• 7. Iterative optimization: Continuously revise the outline based on user feedback, update and overwrite 03-大纲草稿.md, until the user confirms satisfaction. See details:05_iteration_optimization.md

1.5 Phase 5: Final draft archiving

• 8. Final draft archiving: Write the final outline into 03-大纲终稿.md, and output a brief summary and suggested next steps. See details:06_archive_final.md

---

2. Interaction specifications

2.1 Prioritize interaction questioning tools (ask_user / ask_question)

When the execution environment supports ask_question or ask_user , at key decision points the Agentmust prioritize calling the corresponding tool, avoiding plain text interaction.

Application scenarios：

1. Initial confirmation (Step 1): Select publishing platform, writing style, and reference samples.
2. Outline focus confirmation (Step 5): Confirm the outline's directional preference or tonal choice.
3. Retrieval plan confirmation (Step 6.1): Confirm the retrieval plan and select a web access option (such as markdown-new or agent-browser）。

---

3. Corpus collection process

For blog posts, long-form image-text posts, and external excellent cases used for writing reference, their collection, transcription, cleaning, and archiving must follow unified process specifications. See details:07_corpus_collection.md