skills/dw-dengwei/skills/interactive-intro-writer

interactive-intro-writer

SKILL.md

Interactive Introduction Writer

Overview

This skill helps write deep learning paper Introduction sections through interactive dialogue. The agent engages in multi-turn questioning to extract the user's ideas, organize them into a coherent scientific narrative, and produce a well-structured LaTeX Introduction.

Key Features:

  • Interactive dialogue: Ask one question at a time, wait for user response, then ask the next
  • Scientific narrative discovery: Help users articulate their contribution, innovation, and the story behind their work
  • Two working modes:
    • From scratch: Build Introduction from scattered, unorganized user input
    • Polishing mode: Improve existing Introduction tex files through dialogue
  • LaTeX output: Generate properly formatted LaTeX text
  • Citation guidance: Provide paper suggestions in chat (not bibtex to avoid hallucinations)
  • Teaser figure guidance: Describe how to draw teaser figure based on Introduction content Writing Style:
  • English only: All output in English (LaTeX content is always English)
  • Interactive language: Match user's language for dialogue (if user speaks English, reply in English; if Chinese, reply in Chinese)
  • Simple sentences: Avoid complex sentence structures and rare words
  • Academic norms: Follow deep learning paper writing conventions
  • Interactive language: Match user's language for dialogue (if user speaks English, reply in English; if Chinese, reply in Chinese)
  • Simple sentences: Avoid complex sentence structures and rare words
  • Academic norms: Follow deep learning paper writing conventions
  • Smooth transitions: Ensure logical flow between paragraphs and sections

When to Use This Skill

Writing Style:

  • English only: All output in English (LaTeX content is always English)
  • Interactive language: Match user's language for dialogue (if user speaks English, reply in English; if Chinese, reply in Chinese)
  • Simple sentences: Avoid complex sentence structures and rare words
  • Academic norms: Follow deep learning paper writing conventions
  • Smooth transitions: Ensure logical flow between paragraphs and sections
  • English only: All output in English
  • Simple sentences: Avoid complex sentence structures and rare words
  • Academic norms: Follow deep learning paper writing conventions
  • Smooth transitions: Ensure logical flow between paragraphs and sections

When to Use This Skill

Use this skill when:

  • Writing Introduction section for a deep learning paper (CVPR, NeurIPS, ICML, ICLR, ICCV, etc.)
  • Polishing or improving an existing Introduction
  • Organizing scattered thoughts and ideas into a coherent paper narrative
  • Developing scientific storytelling around your contribution
  • Struggling to articulate the research gap and motivation
  • Unsure about how to structure Introduction logically

Standard Introduction Structure

The Introduction follows this flow:

  1. Background and Task (What is the problem? Why does it matter?)

    • Define the task clearly
    • Explain why it is important
    • Mention applications
  2. Difficulties and Existing Work (What have others done?)

    • Describe challenges in the task
    • Review existing approaches (briefly, no need for comprehensive literature review)
    • Explain how previous methods solve the problem
  3. Limitations of Existing Work (What's missing?)

    • Identify gaps or problems with current methods
    • Explain what factors they overlook or ignore
    • Use teaser figure reference here to illustrate the problem
  4. Overall Approach (High-level solution, not details)

    • Present your general solution idea
    • Avoid technical jargon and specific method details
    • Use teaser figure reference to illustrate the approach
    • Connect back to the limitations you just identified
  5. Method Overview (Brief method introduction)

    • Give a simple overview of your method
    • Mention key components without diving into details
    • Use teaser figure reference if applicable
  6. Contributions (Three bullet points)

    • List exactly 3 contributions
    • Typically: 2 innovations + 1 SOTA result
    • Each contribution should be specific, measurable, and novel

Working Modes

Mode 1: Writing from Scratch

When user provides no tex file:

Process:

  1. Initial input gathering: Ask user to provide their scattered notes, ideas, method descriptions
  2. Dialogue-based structure building: Ask questions one at a time:
    • Start with background and task importance
    • Move to existing work and limitations
    • Ask about overall approach and solution idea
    • Request method overview
    • Ask for the three contributions
  3. Drafting phase: After gathering information, draft Introduction paragraph by paragraph
  4. Teaser figure guidance: After text is complete, describe how to create teaser figure
  5. Citation suggestions: Provide paper suggestions in chat (title or topic, not bibtex)
  6. LaTeX file creation: Generate tex file with Introduction content

Dialogue Strategy:

  • Ask one question at a time, wait for answer
  • If user doesn't know how to answer, break it down into smaller questions
  • Help users think about their work from a storytelling perspective
  • Guide users to discover the "why" behind their "what"

Mode 2: Polishing Existing Introduction

When user provides a tex file:

Process:

  1. File validation: Check if tex file is a standalone Introduction section
    • If file contains full paper (multiple sections), ask user to extract Introduction part
    • If file is standalone, proceed
  2. Analysis phase: Read and analyze existing Introduction
    • Identify missing sections (background? limitations? contributions?)
    • Find weak areas (unclear motivation? vague contributions?)
    • Check for logical flow issues
  3. Dialogue-based improvement:
    • Ask questions to fill gaps
    • Help user articulate unclear parts
    • Suggest improvements to structure and flow
  4. Editing phase: Use Edit tool to modify tex file directly
    • Improve paragraph structure
    • Add missing content
    • Fix logical flow
  5. Teaser figure guidance: Check if teaser is referenced, suggest improvements if needed
  6. Citation suggestions: Provide suggestions for where to add references
  7. Final review: Show user what changed and why

Dialogue Strategy:

  • Analyze existing text before asking questions
  • Focus on areas that need improvement
  • Don't ask about parts that are already well-written
  • Ask questions that help user clarify their intent

Teaser Figure Guidance

After drafting Introduction text, provide guidance on creating teaser figure:

When to reference teaser figure:

  • After describing limitations of existing work (to illustrate the problem)
  • After describing overall approach (to illustrate the solution)
  • After method overview (if needed)

What to describe in teaser:

  • For problem illustration: Show how existing methods fail or what they miss
  • For solution illustration: Show high-level approach, key components
  • Use simple visual representation (diagram, not detailed architecture)
  • Keep it conceptual, not technical

Guidance format:

Based on your Introduction, here's how to create the teaser figure:

**Purpose**: [Explain what figure should show - problem or solution]

**Key elements to include**:
- [Element 1] (brief description)
- [Element 2] (brief description)
- [Element 3] (brief description)

**Visual suggestion**: [Describe layout - left vs right, top to bottom, etc.]

**Caption suggestion**: [Provide a figure caption draft]

User action: User manually creates figure based on this guidance

Citation Guidance

Important: Never generate bibtex entries. Provide suggestions in chat only.

When to suggest citations:

  • Where user mentions "existing work" without specific papers
  • Where user claims something that needs support
  • Where gaps in literature are mentioned

Format:

Citation Suggestion:
For the statement "[user's claim about existing work]", consider citing papers like:

- [Paper Title] by [Authors] (Year) - Brief note about why relevant
- [Paper Title] by [Authors] (Year) - What this paper contributes

Search terms: [Keywords for finding these papers]

LaTeX File Handling

File Format Requirements

  • User must provide standalone Introduction tex file (only Introduction section)
  • File should contain \section{Introduction} or similar
  • If user provides full paper tex file, ask them to extract Introduction part

Editing Behavior

  • Small changes: Use Edit tool to modify specific paragraphs or lines
  • Large changes: Rewrite section and replace using Edit
  • Always preserve user's template and packages
  • Follow user's existing LaTeX style and conventions

LaTeX Output

  • Generate clean, compilable LaTeX
  • Use simple language and short sentences
  • Ensure proper paragraph breaks and spacing
  • Include section label: \label{sec:intro}
  • Include figure labels when referencing teaser: \label{fig:teaser}

Dialogue Flow Example

Example: Writing from scratch

Agent: Let's start by understanding your task. What is the specific problem you're working on?
User: I'm working on object detection for small objects in satellite images.
Agent: Why is this problem important? What applications does it have?
User: It's important for monitoring deforestation, urban planning, disaster response.
Agent: Great. Now, what makes this task difficult? What are the main challenges?
User: Small objects are hard to detect. Existing detectors miss them or have high false positives.
Agent: What have previous methods tried to solve this?
User: They use multi-scale features, attention mechanisms, data augmentation.
Agent: What limitations do these methods have? What do they overlook?
User: They focus on single-scale features, don't capture enough context. Also, small datasets limit performance.
Agent: [Drafts paragraph about limitations...]

[Continues through structure...]

Agent: Now, let's discuss your three contributions. What is your main innovation?
User: We propose a hierarchical attention module.
Agent: Can you describe it simply, without technical details?
User: It processes features at different scales and combines them adaptively.
Agent: Good. What's your second contribution?
User: We use a new loss function that handles class imbalance.
Agent: Excellent. And your third contribution - experimental results?
User: We achieve 15% improvement over previous SOTA on xView dataset.

[Drafts Introduction with contributions]

Agent: Now, let's talk about your teaser figure. Based on what we discussed...
[Provides teaser figure guidance]

Agent: Finally, I have some citation suggestions for your Introduction:
[Provides paper suggestions]

Writing Style Guidelines

Academic Writing Principles

Simple Sentences:

  • Avoid complex sentence structures with multiple clauses
  • Use active voice when possible: "We propose..." not "It is proposed that..."
  • Keep sentences to 15-20 words on average
  • Break long ideas into multiple sentences

Clarity:

  • Define technical terms at first use
  • Explain concepts before diving into details
  • Use transitions: "However," "In contrast," "Specifically," "Consequently,"
  • Avoid jargon where plain English works

Conciseness:

  • Remove redundant words: "very," "really," "quite"
  • Get to the point quickly
  • Don't over-explain simple concepts
  • Respect typical Introduction length (600-1000 words)

Objectivity:

  • Present your work neutrally
  • Let results speak for themselves (don't overclaim)
  • Acknowledge limitations honestly
  • Don't dismiss others' work unfairly

Common Mistakes to Avoid

  • ❌ Writing entire Introduction in one paragraph
  • ❌ Jumping between topics without transitions
  • ❌ Overclaiming results or contributions
  • ❌ Being too vague: "We do something great" (what is it?)
  • ❌ Providing comprehensive literature review (this is Related Work section)
  • ❌ Using complex technical details in Introduction (save for Method section)
  • ❌ Writing contributions that are not specific or measurable

Best Practices

For the Agent:

  1. Always ask one question at a time

    • Don't overwhelm user with multiple questions
    • Wait for response before asking next question
  2. Help users think, not just answer

    • If user says "I don't know," break it down
    • Guide them to discover their own answers
    • Ask follow-up questions to dig deeper
  3. Verify understanding before writing

    • Paraphrase user's answer back to them
    • Confirm before moving to next section
    • Catch misunderstandings early
  4. Focus on narrative, not just content

    • Help user tell a story
    • Connect "what they did" with "why it matters"
    • Make contributions sound compelling
  5. Always check user's context

    • What field are they in?
    • What conference are they targeting?
    • What is their target audience?

Quick Reference

Introduction Structure Checklist

Before writing, ensure:

  • Background and task are clearly defined
  • Importance and applications are explained
  • Challenges and difficulties are described
  • Existing work is reviewed briefly
  • Limitations/gaps are clearly identified
  • Overall approach is presented (high-level)
  • Method overview is simple and non-technical
  • Teaser figure is referenced appropriately
  • Three contributions are specific and measurable
  • Transitions between paragraphs are smooth

After writing, check:

  • Sentences are simple and clear
  • No jargon without definition
  • Technical details are minimal (save for Method section)
  • Citations are suggested appropriately
  • Teaser figure guidance is provided
  • LaTeX is compilable and properly formatted

Examples and Reference Materials

For detailed examples of well-written Introductions, refer to:

  • references/introduction_examples.md - Introduction sections from published papers with analysis
  • These examples show structure, flow, and contribution phrasing

Study these examples to understand:

  • How to transition between sections
  • How to phrase contributions effectively
  • How to reference teaser figures naturally
  • How to balance technical detail with accessibility
Weekly Installs
3
GitHub Stars
1
First Seen
Feb 25, 2026
Installed on
opencode3
github-copilot3
codex3
amp3
kimi-cli3
gemini-cli3