--- name: ai-writing-assistant description: Leverage AI tools effectively for professional writing tasks. Provides prompt patterns, refinement workflows, voice preservation techniques, and quality checkpoints for AI-augmented drafting. allowed-tools: Read, Glob, Grep --- # AI Writing Assistant Skill Use AI tools strategically to enhance your professional writing while maintaining authenticity and quality. ## Keywords ai, chatgpt, copilot, grammarly, prompt, draft, writing assistant, voice, tone, refinement, editing, automation ## When to Use This Skill This skill provides guidance when developers need to: - Draft professional content with AI assistance - Refine and improve existing drafts using AI - Maintain authentic voice while using AI tools - Choose appropriate AI workflows for different writing tasks - Verify and polish AI-generated content - Collaborate between AI drafts and human editing ## Core Philosophy **AI as collaborator, not replacement.** AI writing tools work best when you: 1. **Provide clear direction** - Good prompts produce good drafts 2. **Iterate on output** - First draft is never final 3. **Inject your voice** - Add personality and expertise 4. **Verify everything** - AI makes confident mistakes 5. **Know when NOT to use AI** - Some tasks need pure human judgment ## The Two Workflows ### Workflow 1: AI-First Drafting **Use when:** Starting from blank, need structure, time-pressured ```text ┌─────────────────────────────────────────┐ │ 1. PROMPT │ │ Provide context, constraints, examples │ ├─────────────────────────────────────────┤ │ 2. GENERATE │ │ Let AI produce initial draft │ ├─────────────────────────────────────────┤ │ 3. EVALUATE │ │ Check accuracy, tone, completeness │ ├─────────────────────────────────────────┤ │ 4. REFINE │ │ Add voice, fix errors, enhance │ ├─────────────────────────────────────────┤ │ 5. VERIFY │ │ Final quality check before sending │ └─────────────────────────────────────────┘ ``` **Best for:** - First drafts of emails - Outline generation - Template creation - Brainstorming options - Summarizing content ### Workflow 2: Human-First Refinement **Use when:** You have ideas, need polish or alternatives ```text ┌─────────────────────────────────────────┐ │ 1. DRAFT │ │ Write your rough version first │ ├─────────────────────────────────────────┤ │ 2. ASK │ │ Request specific improvements │ ├─────────────────────────────────────────┤ │ 3. COMPARE │ │ Evaluate AI suggestions vs original │ ├─────────────────────────────────────────┤ │ 4. MERGE │ │ Combine best of both versions │ ├─────────────────────────────────────────┤ │ 5. VERIFY │ │ Final quality check │ └─────────────────────────────────────────┘ ``` **Best for:** - Improving clarity - Finding better phrasing - Expanding or condensing - Tone adjustment - Grammar and style fixes ## Prompt Patterns by Task ### Email Prompts **Basic email draft:** ```text Write a professional email to [recipient role] about [topic]. Context: - [Relationship/history] - [Goal of this email] - [Any constraints] Tone: [professional/friendly/urgent/casual] Length: [brief/detailed] ``` **Email refinement:** ```text Improve this email for clarity and professional tone: [Paste your draft] Keep the core message but: - Make it more concise - Fix any awkward phrasing - Ensure call-to-action is clear ``` ### Feedback Prompts **Structuring feedback:** ```text Help me structure feedback for a team member using the SBI model. Situation: [When and where] Behavior: [What I observed - facts only] Impact: [Effect on team/project/outcomes] Make it constructive and specific. Avoid vague language. ``` **Preparing for difficult conversation:** ```text I need to have a difficult conversation about [topic]. Context: [Background] My concern: [What I want to address] Desired outcome: [What I hope to achieve] Help me: 1. Plan my opening statement 2. Anticipate possible responses 3. Prepare redirects if conversation derails ``` ### Presentation Prompts **Outline generation:** ```text Create an outline for a [duration] technical presentation on [topic]. Audience: [Who they are, their level] Goal: [What should they learn/do after] Key message: [Main takeaway] Use the What-Why-How structure: - What: Hook and problem statement - Why: Why this matters to them - How: The solution/approach - Close: Key takeaways and CTA ``` **Slide content:** ```text Create slide content for: [Section topic] Requirements: - Title: Action-oriented, 5-8 words - Bullets: 3-5 points, 5-7 words each - One main idea only This slide is about: [Specific point] ``` ### LinkedIn/Content Prompts **Post creation:** ```text Create a LinkedIn post about [topic]. Frame: [learning/win/behind-the-build/contrarian/quick-tip] Hook: Start with something that makes people click "see more" Length: 150-250 words End with: Question to drive engagement My angle/experience: [What makes my perspective unique] ``` **Content repurposing:** ```text Turn this [blog post/documentation/talk] into a LinkedIn post: [Paste content] Extract the most interesting insight and structure as: 1. Hook (attention-grabbing opening) 2. Story/context (brief setup) 3. Insight (the key learning) 4. CTA (question for engagement) ``` ## Voice Preservation Techniques ### The Voice Sample Before using AI for your content, create a voice sample: ```text Analyze my writing style from these samples: [Paste 3-5 paragraphs of your writing] Identify: 1. Sentence structure patterns 2. Word choice tendencies 3. Tone and formality level 4. Unique phrases or expressions 5. How I open and close sections ``` Then use in subsequent prompts: ```text Write in a style that matches this voice profile: [Paste analyzed voice profile] Now write: [Your request] ``` ### The Injection Method Add personal touches after AI drafting: 1. **Add your examples** - Replace generic examples with real ones 2. **Insert your phrases** - Add expressions you actually use 3. **Include specifics** - Add names, dates, details AI couldn't know 4. **Add opinions** - AI stays neutral; add your actual take 5. **Include humor** - If that's your style, add it yourself ### The Hybrid Paragraph Write key paragraphs yourself, let AI fill in the rest: ```text Complete this draft, matching the style of paragraphs I've written: My paragraphs (keep exactly as-is): [Your key paragraphs] Fill in sections marked [AI DRAFT]: 1. [AI DRAFT - Introduction] 2. [Your paragraph] 3. [AI DRAFT - Transition] 4. [Your paragraph] 5. [AI DRAFT - Conclusion] ``` ## Quality Checkpoints ### Checkpoint 1: Accuracy Check **Before sending anything AI-assisted, verify:** - [ ] All facts are correct (AI makes up statistics) - [ ] Names and titles are accurate - [ ] Technical details are right - [ ] Links work (AI invents URLs) - [ ] Quotes are real (AI fabricates quotes) - [ ] Dates and numbers are correct ### Checkpoint 2: Tone Check **Read aloud and ask:** - [ ] Does this sound like me? - [ ] Would I actually say this in person? - [ ] Is the formality level appropriate? - [ ] Are there phrases I'd never use? - [ ] Does it feel authentic? ### Checkpoint 3: Content Check - [ ] Is the main point clear? - [ ] Is it the right length? - [ ] Does it answer the recipient's actual question? - [ ] Is the call-to-action obvious? - [ ] Would I be proud to have my name on this? ### Checkpoint 4: Risk Check **For sensitive content:** - [ ] Could this be misunderstood? - [ ] Is anything potentially offensive? - [ ] Am I sharing anything I shouldn't? - [ ] What if this were forwarded/screenshot? - [ ] Should a human review this? ## When NOT to Use AI ### Tasks Requiring Human Judgment - **Layoff or termination communications** - Too sensitive for AI - **Legal or compliance content** - Needs expert review - **Highly personal feedback** - Requires genuine human care - **Crisis communications** - Stakes too high - **Salary or promotion discussions** - Requires human touch - **Content about specific people** - Risk of inaccuracy ### Tasks AI Struggles With - **Nuanced cultural context** - AI may miss subtlety - **Company-specific jargon** - AI doesn't know your abbreviations - **Recent events** - Training cutoffs limit knowledge - **Highly technical accuracy** - AI confidently makes mistakes - **Sarcasm or subtle humor** - Often misses the mark ## Common AI Writing Mistakes ### Mistake 1: Over-Formality **AI writes:** "I am writing to inform you that..." **Fix:** Replace with natural phrasing: "Quick update on..." ### Mistake 2: Hedge Words **AI writes:** "It might be potentially beneficial to perhaps consider..." **Fix:** Be direct: "I recommend..." ### Mistake 3: Hollow Phrases **AI writes:** "In today's fast-paced world..." **Fix:** Delete or replace with specific context ### Mistake 4: Excessive Structure **AI writes:** "Firstly... Secondly... Thirdly..." **Fix:** Vary sentence structure, remove unnecessary markers ### Mistake 5: Generic Examples **AI writes:** "For example, Company XYZ..." **Fix:** Replace with real, specific examples ## Tool-Specific Tips ### ChatGPT/Claude - Use system prompts to set context - Iterate with follow-up requests - Ask for multiple versions - Request specific changes, not vague "improve" ### Grammarly - Good for grammar, punctuation, clarity - Be skeptical of tone suggestions - Disable suggestions that change your voice - Use as final polish, not primary drafting ### GitHub Copilot (for Documentation) - Good for code comments and docstrings - Verify technical accuracy carefully - Adjust for your team's style - Use for boilerplate, edit the substance ## References For detailed guidance, see: - `references/ai-prompt-patterns.md` - Comprehensive prompt templates by use case - `references/voice-preservation.md` - Techniques for maintaining authenticity ## Related Commands - `/soft-skills:draft-email` - Generate professional emails - `/soft-skills:craft-linkedin-post` - Create engaging posts - `/soft-skills:prepare-feedback` - Structure feedback using SBI ## Anti-Patterns to Avoid - Sending AI output without reading it - Using AI for highly sensitive content - Trusting AI facts without verification - Letting AI override your authentic voice - Over-relying on AI for tasks you should learn - Copy-pasting prompts without customization ## Version History - **v1.0.0** (2025-12-26): Initial release --- ## Last Updated **Date:** 2025-12-26 **Model:** claude-opus-4-5-20251101