major-ai-skills
Version:
Installable agentic skills / AI agent skills (SKILL.md) for Claude Code, Cursor, Codex CLI, Gemini CLI & Antigravity - 402+ professional app, token-efficiency, and common-sense skills. SEO/GEO ready.
101 lines (75 loc) • 4.71 kB
Markdown
name: outline-before-writing
description: "Agree on a document outline and logical progression before drafting a long piece."
category: common-sense
risk: safe
source: self
source_type: self
date_added: "2026-08-26"
tags: ["outlining", "writing-workflows", "document-structure", "planning", "productivity", "prompt-engineering"]
tools: ["claude", "cursor", "gemini", "codex", "chatgpt"]
# Outline Documents Before Writing (The Structural Gate) (AI Skill)
## Overview
When asked to write a 1,500-word article, technical guide, or report in a single prompt, AI models generate a flat, unstructured wall of text where subtopics overlap, key arguments are buried, and transitions feel disjointed.
The **Structural Outline Gate Protocol** requires the AI to generate a strict, hierarchical **5-to-7 point outline** first. Once you review, reorder, or trim headings, each section is expanded with surgical precision.
## The Structural Gate Workflow
```
┌─────────────────────────────────────────────────────────────┐
│ The Structural Outline Gate │
│ │
│ User Topic ──► [ STEP 1: Generate 5-Point Hierarchical Outline ] │
│ │ │
│ Human Reviews & Adjusts ◄────────┘ │
│ (Swap sections, delete filler, add unique case study) │
│ │ │
│ ▼ │
│ [ STEP 2: Lock Headings & Expand Section by Section ] │
│ ↳ Cohesive, High-Depth, Zero-Rambling Final Document │
└─────────────────────────────────────────────────────────────┘
```
## Master Outlining Prompt Templates
### Pattern 1: The Hierarchical Outline Gate
Use for long-form essays, whitepapers, guides, and proposals:
```markdown
I want to write a comprehensive guide on [TOPIC].
Target Audience: [AUDIENCE: e.g. Staff Engineers / Marketing Directors].
Primary Thesis / Message: [CORE THESIS].
Do NOT start writing the full text yet.
Provide a **Hierarchical Outline** with:
1. Exactly 5 to 6 major numbered H2 sections.
2. Under each H2, list 2-3 specific bullet points detailing the exact arguments, examples, or metrics to be covered.
3. Suggest a 1-sentence opening hook for the introduction.
Conclude by asking: "Do you want to adjust any section before we draft?"
```
### Pattern 2: The Section-by-Section Expansion Directive
Use after you approve or modify the outline:
```markdown
The outline is approved.
Now, draft **Section 1 and Section 2 ONLY**:
- Maintain an authoritative, conversational tone.
- Include concrete examples and code/data blocks where specified.
- Stop at the end of Section 2 so I can review tone before you continue.
```
## Real-World Case Study
### Scenario: Writing an Architecture Guide on Event-Driven Microservices
#### The 1-Shot Writing Failure
> **Prompt**: *"Write a 2,000-word guide on event-driven architecture."*
> ❌ *Result: A rambling, repetitive essay that mentions Kafka 40 times, repeats the definition of pub/sub in 3 different sections, and skips critical dead-letter queue (DLQ) failure handling.*
#### The Outline Gate Success
> **AI Outline Generated**:
> 1. **The Distributed Monolith Trap**: Why REST API chains fail under load.
> 2. **Core Pub/Sub Mechanics**: Producers, Brokers, Consumers, and Topic Partitioning.
> 3. **The 3 Golden Rules of Event Design**: Immutability, Schema Evolution (Protobuf/Avro), and Idempotent Consumers.
> 4. **Handling Partial Failures**: Dead Letter Queues (DLQ) and Exponential Backoff Retries.
> 5. **Migration Blueprint**: From synchronous HTTP calls to asynchronous messaging in 4 steps.
**Human Edit**: *"Add a section on Outbox Pattern between 3 and 4, then proceed with Section 1."*
**Outcome**: The resulting guide was rigorous, structured logically, and required zero rewriting.
## Summary Rules
- **Never skip the outline gate for content $> 500$ words**: It is 10x faster to delete an outline bullet than to rewrite a 400-word generated paragraph.
- **Lock the headings first**: Once headings are locked, generation tokens are channeled with maximum depth into each subtopic.