Context Engineering Guide - Document Specification

Context Engineering Guide - Document Specification

Version: 1.0

Date: 2025-11-28


1. Audience

Primary Audience

  • Who: Developers, technical professionals using AI tools
  • Knowledge Level: Familiar with AI basics, use ChatGPT/Claude occasionally
  • Pain Points: Inconsistent AI results, frustration with outputs, "AI output is unusable"
  • Goals: Get better results, save time, work more effectively with AI

Secondary Audience

  • Who: Managers, team leads, non-technical professionals, content creators
  • Knowledge Level: Basic AI awareness, limited hands-on experience
  • Pain Points: Don't know where to start, team adoption challenges
  • Goals: Understand the concept, implement with team, improve personal productivity

Audience Balance

  • Technical depth: 60% accessible to everyone, 40% technical details
  • Examples split: Every concept has both technical AND non-technical example
  • Jargon: Define terms on first use, avoid unnecessary complexity

2. Format

File Format

  • Primary: Markdown (.md)
  • Export targets: PDF (primary), HTML/web (secondary)

Document Structure

guide_draft/
├── 00_introduction.md
├── 01_part1_ch1_why_prompts_fail.md
├── 02_part1_ch2_good_context.md
├── 03_part2_ch3_before_prompt.md
├── 04_part2_ch4_during_after.md
├── 05_part2_ch5_tools.md
├── 06_part3_ch6_examples.md
├── 07_part3_ch7_teams.md
├── 08_part4_ch8_vibe_coding.md
├── 09_part4_ch9_wont_replace.md
├── 10_appendix.md
└── GUIDE_FINAL.md (assembled)

Length Guidelines

| Component | Target Length | |-----------|---------------| | Introduction | 600-800 words | | Chapter (average) | 2,500-4,000 words | | Section (average) | 400-800 words | | Chapter summary | 100-150 words | | Appendix | 1,500-2,000 words | | Total guide | ~24,000 words (~48 pages) |


3. Style Guide

Voice & Tone

  • Voice: First person, author as practitioner ("I discovered...", "In my experience...")
  • Tone: Professional but conversational, like explaining to a smart colleague
  • Formality: Semi-formal, contractions okay, no slang
  • Personality: Direct, practical, occasionally humorous (Borovička joke), always honest about limitations

Language Rules

  1. Active voice preferred: "AI processes the context" not "The context is processed by AI"
  2. Short sentences: Average 15-20 words per sentence
  3. Short paragraphs: 3-5 sentences max
  4. Define jargon: First use of technical term includes brief definition
  5. No fluff: Every sentence must add value
  6. Concrete over abstract: "Fix in 2 minutes" not "dramatically improved efficiency"

Formatting Rules

  1. Headers: Use # for parts, ## for chapters, ### for sections, #### for subsections
  2. Lists: Use - for unordered, 1. for sequential steps
  3. Code blocks: Use ``` for any code, prompts, or structured text
  4. Emphasis: Bold for key terms first time, italic for titles/emphasis
  5. Quotes: Use > blockquote for external citations

Example: Good vs Bad Writing

Bad: "The utilization of contextual information in the facilitation of AI-based task completion is fundamentally important for achieving optimal outcomes in various scenarios."

Good: "Good context leads to good AI output. Here's why: AI doesn't know your project, your goals, or your constraints. You have to tell it."


4. Visual Elements

Diagrams Needed

| Diagram | Chapter | Description | |---------|---------|-------------| | Prompt vs Context | Ch 1 | Input divided: Prompt (10%) vs Context (90%) | | Five Components | Ch 2 | Task, Constraints, Background, Examples, Criteria | | Task Template | Ch 2 | Visual template layout | | Research Workflow | Ch 5 | Perplexity → Extract → Claude/ChatGPT | | Fix vs Fresh Decision | Ch 4 | Decision tree for when to restart | | Tool Division | Ch 5 | Perplexity/Claude/ChatGPT by task type | | Team Failure Pattern | Ch 7 | Licenses → Training → Chaos → Abandonment | | 6-Week Timeline | Ch 7 | Implementation roadmap |

Tables Needed

| Table | Chapter | Content | |-------|---------|---------| | Tool comparison | Ch 5 | Perplexity vs Claude vs ChatGPT | | Common mistakes | Ch 3 | Mistake + Fix | | Role responsibilities | Ch 7 | PO, Dev, QA, PM duties | | Success metrics | Ch 7 | What to track vs NOT track | | 10 Advices | Appendix | Quick reference checklist |

Example Box Format

Standard format for before/after comparisons:

> ❌ **Without Context:**
> "Fix the bug"
>
> **Result:** 200 lines of generic checks, none solving the problem

> ✅ **With Context:**
> "Fix permission error in ProfileView.tsx:156. User has admin role but can't edit.
> Check: Role loading, UI vs API permission logic"
>
> **Result:** Found exact issue in 2 minutes - UI checks wrong role

Code/Prompt Blocks

  • Use syntax highlighting where applicable (markdown, sql, typescript)
  • Include comments for complex prompts
  • Show realistic examples from actual work (anonymized)

5. Citations & Attribution

Internal Citations (Author's Content)

  • No formal citation needed for own articles
  • Reference naturally: "As I mentioned earlier..." or "This relates to the 5-component framework"

External Citations Format

> "Quote text here"
> — **Author Name**, Platform/Source, Year

Citation Index (End of Guide)

## References

1. Karpathy, A. (2025). Twitter/X post on context engineering.
   "Context engineering is the delicate art and science..."

2. Arrowsmith, G. (2025). LinkedIn post on QA bottleneck.
   "QA is about to become a huge bottleneck..."

3. Martiniak, M. (2025). Survey on programmer time allocation.

Attribution Rules

  1. Always attribute direct quotes
  2. Paraphrased ideas from specific sources need attribution
  3. General knowledge doesn't need attribution
  4. When in doubt, attribute

6. Consistency Requirements

Terminology Table

| Term | Use | Don't Use | |------|-----|-----------| | Context engineering | ✓ | Context crafting, context design | | AI | ✓ | Artificial intelligence (spell out once) | | Prompt | ✓ (the instruction) | Query, request, command | | Context | ✓ (background info) | Data, information (when meaning context) | | Output | ✓ (AI response) | Result, answer (when technical) | | Session | ✓ (conversation) | Chat, thread | | Claude | ✓ | Claude AI, Anthropic's Claude | | ChatGPT | ✓ | GPT, OpenAI's ChatGPT | | Perplexity | ✓ | Perplexity AI |

Formatting Consistency Checklist

  • [ ] All chapters use same header levels
  • [ ] All code blocks use same style (```language)
  • [ ] All examples follow ❌/✅ format
  • [ ] All external quotes use > blockquote with attribution
  • [ ] All lists use consistent punctuation (periods or none)
  • [ ] All chapter summaries follow same structure

Naming Consistency

  • "Part" for major divisions (Part 1, Part 2...)
  • "Chapter" for chapters within parts (Chapter 1, Chapter 2...)
  • "Section" for sections within chapters (3.1, 3.2...)
  • Numbering: 1.1, 1.2 (not 1a, 1b)

7. Technical Specifications

Markdown Flavor

  • Use: GitHub Flavored Markdown (GFM)
  • Tables: Pipe tables supported
  • Task lists: - [ ] supported
  • Footnotes: Avoid (use inline attribution instead)

Special Characters Used

  • ✓ (checkmark)
  • ✗ (x mark)
  • → (arrow)
  • ❌ (red X for bad example)
  • ✅ (green check for good example)
  • ⚠️ (warning)
  • — (em dash for attribution)

Line Length

  • Soft wrap at 80-100 characters for editor readability
  • No hard line breaks within paragraphs (let renderer wrap)

File Naming

00_introduction.md
01_part1_ch1_[short_name].md
02_part1_ch2_[short_name].md
...
10_appendix.md

8. Chapter Template

Every chapter follows this structure:

# Chapter [N]: [Title]

[Opening hook - 2-3 sentences that grab attention]

In this chapter, you'll learn:
- [Outcome 1]
- [Outcome 2]
- [Outcome 3]

---

## [N.1] [Section Title]

[Section content with examples]

---

## [N.2] [Section Title]

[Section content with examples]

---

## Chapter Summary

**Key Takeaways:**
1. [Most important point]
2. [Second point]
3. [Third point]

**Try This:**
[One specific action reader can take immediately]

---

*Next: [Brief teaser for next chapter]*

9. Quality Checklist for Writers

Before Writing Each Chapter:

  • [ ] Read topic analysis file completely
  • [ ] Review this document spec (voice, tone, format)
  • [ ] Check previous chapter ending (for continuity)
  • [ ] Note required visuals/examples from topic analysis
  • [ ] Set target word count

While Writing:

  • [ ] Follow voice/tone guidelines
  • [ ] Use correct formatting (headers, code blocks, quotes)
  • [ ] Include both technical and non-technical examples
  • [ ] Add citations/attributions as you write
  • [ ] Create visuals or mark placeholder locations

After Writing:

  • [ ] Run consistency check against this spec
  • [ ] Verify all planned visuals included
  • [ ] Check word count is in range
  • [ ] Read aloud for flow
  • [ ] Verify chapter summary matches content

10. Quick Reference Card

Voice: First person, practical, direct

Tone: Professional but conversational

Examples: Always bad/good pairs with ❌/✅

Headers: # Part, ## Chapter, ### Section

Code: ```language for all structured text

Quotes: > blockquote with — attribution

Terms: Use terminology table consistently


Next Step

→ Proceed to 05_introduction_writing.md

Article Details

Category
context engineering new
Published
November 28, 2025
Length
1,493 words
9,534 characters
~6 pages
Status
Draft Preview

More from context engineering new

Why You Give Much and Get Little (And How to Change It)

# Why You Give Much and Get Little (And How to Change It) "AI output is unusable" - I hear this constantly. Many people don't use AI precisely because of this. But the problem isn't AI. The problem i...

Read article