agentleFS
cellinlabcellinlab/cell-skills/skills/deep-writer/SKILL.md

deep-writer

Skill29 starsChanged 7 months agoMITView on GitHub

Adaptive deep-writing workflow for all forms of deep content creation in Chinese, including turning transcripts and note piles into articles, deepening rough drafts, or writing structured long-form content from a topic, thesis, question, or outline. Use when Codex needs to clarify audience and purpose, build a brief, extract or construct a sound argument structure, merge repeated ideas, design an outline before drafting, deepen analysis, preserve source logic when needed, or deliver writing through staged checkpoints instead of a one-shot article.

The file

Show its rules and text

We split this skill into the rules it gives an agent

28 rules. About 0 are common, 26 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈26

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Default to pausing for user confirmation after analysis and structure unless the user explicitly asks for one-pass output.

    line 17
  2. Distinctive:

    Run the final draft against the brief and the quality checklist.

    line 25
  3. Distinctive:

    If the user only wants a brief, outline, or structural diagnosis, stop at the relevant stage instead of forcing a full article.

    line 27
  4. Distinctive:

    Treat these as the default assumptions unless the user says otherwise:

    line 31
  5. Distinctive:

    Default goal: formal publishable writing, not chat-style rewriting

    line 40
  6. Distinctive:

    Default process: wait for confirmation after Stage 1 and Stage 2

    line 45
  7. Distinctive:

    If the user provides an already-approved outline, treat it as Stage 2 input and move to Stage 3 after a lightweight brief alignment.

    line 47
  8. Distinctive:

    do the relevant analysis for the mode:

    line 58
  9. Distinctive:

    Read references/workflow.md when the material is fragmented, the thesis is still fuzzy, or the structure is not yet obvious.

    line 76
  10. Distinctive:

    Use the accepted Stage 1 output to design the writing frame.

    line 80
  11. Distinctive:

    Do all of the following:

    line 82
  12. Very specific:

    Use assets/outline-template.md when the user wants a reusable structure blueprint.

    line 93
  13. Distinctive:

    Write the full article only after the structure is locked.

    line 105
  14. Distinctive:

    follow the accepted structure instead of improvising a new one

    line 109
  15. Distinctive:

    remove transcript residue and spoken fillers when they exist

    line 111
  16. Distinctive:

    keep the author's core meaning intact in source-driven mode

    line 112
  17. Distinctive:

    keep the improved version aligned with the original draft's intended thesis in draft-deepening mode

    line 113
  18. Distinctive:

    keep the thesis coherent and non-repetitive in topic-driven mode

    line 114
  19. Distinctive:

    Read references/quality-bar.md before finalizing the draft.

    line 117
  20. Very specific:

    Use the staged wrapper in assets/stage-output-template.md unless the user requested another format.

    line 121
  21. Distinctive:

    keep obvious duplicate viewpoints in separate sections

    line 151
  22. Distinctive:

    add generic filler that does not deepen the reader's understanding

    line 152
  23. Distinctive:

    keep the brief as the baseline for structure and drafting decisions

    line 161
  24. Distinctive:

    Read for mode selection, the full stage checklist, clustering heuristics, topic-driven argument design, one-shot mode, and failure recovery.

    line 167
  25. Distinctive:

    Read for the final review checklist, spoken-language cleanup rules, topic-writing quality rules, and over-expansion guardrails.

    line 169
  26. Distinctive:

    Use when drafting the stage-one brief.

    line 171
  27. Distinctive:

    Use when presenting or locking a reusable structure blueprint.

    line 173
  28. Distinctive:

    Use when formatting staged outputs and todo-state updates.

    line 175

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Deep Writer

Overview

Handle deep content writing in three common modes:

  • source-driven: turn transcripts, notes, interviews, or messy drafts into strong articles
  • draft-deepening: take an existing article or outline and make it deeper, clearer, and more complete
  • topic-driven: write a deep piece from a topic, thesis, question, angle, or user idea

Default to pausing for user confirmation after analysis and structure unless the user explicitly asks for one-pass output.

Quick Start

  1. Choose the writing mode before doing anything else.
  2. Infer the audience, scenario, purpose, and stance before drafting.
  3. Complete Stage 1 before writing the article body.
  4. Lock Stage 2 before expanding long-form prose.
  5. Run the final draft against the brief and the quality checklist.

If the user only wants a brief, outline, or structural diagnosis, stop at the relevant stage instead of forcing a full article.

Default Contract

Treat these as the default assumptions unless the user says otherwise:

  • Required: either source material, or a topic / thesis / question / outline to write from
  • Optional: supporting materials, target audience, publication scenario, stance, length, depth, tone, and style
  • Default mode selection:
    • source-driven for transcripts, notes, interviews, speeches, and fragmented material
    • draft-deepening for rough drafts, partial articles, and existing outlines
    • topic-driven for requests that start from a theme, thesis, question, or desired article idea
  • Default language: Chinese
  • Default goal: formal publishable writing, not chat-style rewriting
  • Default structure policy:
    • source-driven: preserve the source's macro frame and allow only micro-adjustments
    • draft-deepening: keep the strongest existing frame, then strengthen weak or thin sections
    • topic-driven: design the best-fit structure around the brief and thesis
  • Default process: wait for confirmation after Stage 1 and Stage 2

If the user provides an already-approved outline, treat it as Stage 2 input and move to Stage 3 after a lightweight brief alignment.

Workflow

Stage 1: Analysis and Planning

Perform all of the following before drafting:

  • identify the writing mode
  • identify the likely audience, scenario, purpose, and stance
  • build a brief using assets/brief-template.md
  • do the relevant analysis for the mode:
    • source-driven: preprocess spoken-language noise, repetitions, and fragmentary clauses
    • draft-deepening: identify weak logic, thin sections, repeated claims, and missing support
    • topic-driven: clarify the central question, working thesis, key angles, and likely counterpoints
  • extract viewpoint units, argument units, or section units and cluster overlapping ideas
  • explain the main logic relations or argument chain

Output at least:

  • the brief
  • mode diagnosis
  • preprocessing notes or argument notes
  • viewpoint clusters or argument clusters
  • logic chain or logic graph summary
  • open questions, evidence gaps, or assumptions when they materially affect the piece
  • updated todo status
  • a confirmation prompt for the next stage

Read references/workflow.md when the material is fragmented, the thesis is still fuzzy, or the structure is not yet obvious.

Stage 2: Structure Design

Use the accepted Stage 1 output to design the writing frame.

Do all of the following:

  • choose the right structure rule for the mode:
    • source-driven: preserve the source's core framework unless the user explicitly requests a rewrite
    • draft-deepening: keep the strongest existing structure, but rebuild weak branches when the draft cannot support the thesis
    • topic-driven: build the strongest structure around the thesis, reader, and scenario
  • merge duplicate sections or overlapping arguments
  • decide section order with the minimum necessary reordering in source-driven mode, or the clearest reader path in topic-driven mode
  • assign each section a clear job, not just a title
  • explain which parts were kept and which parts were adjusted

Use assets/outline-template.md when the user wants a reusable structure blueprint.

Output at least:

  • the optimized structure
  • the role of each section
  • notes on preserved versus adjusted structure
  • updated todo status
  • a confirmation prompt for the next stage

Stage 3: Content Writing

Write the full article only after the structure is locked.

Do all of the following:

  • follow the accepted structure instead of improvising a new one
  • deepen claims with relevant background, mechanisms, examples, or implications
  • remove transcript residue and spoken fillers when they exist
  • keep the author's core meaning intact in source-driven mode
  • keep the improved version aligned with the original draft's intended thesis in draft-deepening mode
  • keep the thesis coherent and non-repetitive in topic-driven mode
  • finish with a quality-check summary

Read references/quality-bar.md before finalizing the draft.

Output Format

Use the staged wrapper in assets/stage-output-template.md unless the user requested another format.

Default wrapper:

【阶段X完成】

---
【本阶段输出】
[stage output]

---
【待办事项状态】
- [completed] 第一阶段:分析与规划
- [pending] 第二阶段:结构设计
- [pending] 第三阶段:内容撰写

---
是否继续下一阶段?(输入“继续”进入下一阶段,或提出修改意见)

For the final stage, replace the confirmation prompt with a short quality-check summary.

Hard Rules

Do not:

  • invent unsupported facts, data, case details, or quotations
  • invent a new central thesis that is absent from the source material in source-driven mode
  • radically rebuild the structure after Stage 2 is accepted
  • keep obvious duplicate viewpoints in separate sections
  • add generic filler that does not deepen the reader's understanding
  • distort the source's stance in order to sound smoother or more authoritative
  • pretend a topic-driven article is well evidenced if the user did not supply evidence and none is available

Always:

  • choose the mode explicitly when the request is ambiguous
  • make the logic chain visible
  • distinguish source viewpoints from supplemental interpretation when that boundary matters
  • keep the brief as the baseline for structure and drafting decisions
  • state assumptions when the user did not provide audience, scenario, or stance

Resource Map

  • references/workflow.md
    • Read for mode selection, the full stage checklist, clustering heuristics, topic-driven argument design, one-shot mode, and failure recovery.
  • references/quality-bar.md
    • Read for the final review checklist, spoken-language cleanup rules, topic-writing quality rules, and over-expansion guardrails.
  • assets/brief-template.md
    • Use when drafting the stage-one brief.
  • assets/outline-template.md
    • Use when presenting or locking a reusable structure blueprint.
  • assets/stage-output-template.md
    • Use when formatting staged outputs and todo-state updates.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.