Prompt Engineering & Technical Writing
Create effective AI prompts and technical content. Covers prompt structure, anti-patterns, cross-platform compatibility, and WordPress publishing.
Prompt Structure
Core Prompts (< 2000 tokens)
Self-contained, portable across all platforms:
markdown
# [Prompt Name] > **Category**: [category] > **Platforms**: All [Prompt content — complete instructions in a single block] ## Usage [How to use] ## Variables - [VARIABLE]: [description]
Extended Prompts
Unlimited length, may reference skills or tools:
markdown
# [Prompt Name] > **Category**: [category] > **Platforms**: Claude Code | Cursor <prompt> [Full prompt content with tool references] </prompt> ## Usage [Platform-specific instructions] ## Customization [What can be modified]
Writing Rules
Anti-Patterns (Never Use)
Banned words: delve, tapestry, myriad, leverage, utilize, seamless, robust, paradigm, synergy, cutting-edge, game-changer
Banned transitions: furthermore, moreover, additionally, subsequently, consequently
Banned openings: "In today's...", "Have you ever...", "Let's dive into..."
Good Practices
- •Use active voice and present tense
- •Use second person ("you")
- •Use contractions (it's, don't, won't)
- •Vary sentence and paragraph lengths
- •Be direct — state things without hedging
- •No unnecessary qualifiers (very, quite, rather)
Technical Blog Posts
When drafting from a codebase:
- •Research: Read recent commits, review changed files, note design decisions
- •Plan: Opening hook, overview, problem/value, technical details, challenges, future
- •Write: 500-1200 words, code snippets 5-15 lines, save to
.blog/YYYY-MM-DD-slug.md - •Publish (optional): Create WordPress draft via REST API
Prompt Quality Dimensions
Score each 1-5:
| Dimension | What to Check |
|---|---|
| Structure | Title, metadata, proper tags, usage section |
| Clarity | Unambiguous instructions, documented variables |
| Effectiveness | Enough context, specific constraints, defined output |
| Compatibility | Works without tools, under 2000 tokens (core), no hard-coded paths |
| Style | No AI indicator words, active voice, current WP version |
Cross-Platform Tips
- •ChatGPT/Gemini: No tool access — prompts must be self-contained
- •Claude Code: Can reference files, use skills, run commands
- •Cursor: Can read project files, uses rules files
- •Core prompts must work on the lowest common denominator
Token Optimization
For core prompts under 2000 tokens:
- •Use concise instructions, not verbose explanations
- •Prefer lists over paragraphs
- •Include only essential code examples
- •Reference skills instead of duplicating knowledge
- •Remove redundant phrasing