interactive-intro-writer

# 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:** ```markdown 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