ClawKit Logo
ClawKitReliability Toolkit
Back to Registry
Official Verified

deep-writer

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.

skill-install — Terminal

Install via CLI (Recommended)

clawhub install openclaw/skills/skills/cellinlab/cell-deep-writer
Or

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:

Metadata

Author@cellinlab
Stars4017
Views2
Updated2026-04-11
View Author Profile
AI Skill Finder

Not sure this is the right skill?

Describe what you want to build — we'll match you to the best skill from 16,000+ options.

Find the right skill
Add to Configuration

Paste this into your clawhub.json to enable this plugin.

{
  "plugins": {
    "official-cellinlab-cell-deep-writer": {
      "enabled": true,
      "auto_update": true
    }
  }
}
Safety NoteClawKit audits metadata but not runtime behavior. Use with caution.