| name | description |
|---|---|
rule-of-five |
Iterative refinement methodology for high-quality outputs. Based on Jeffrey Emanuel's discovery that LLM agents produce best work through 4-5 refinement passes. Use when creating substantial content that needs to be polished, professional, or production-ready—documents, code, designs, analyses, presentations. Triggers include requests for "high quality", "polished", "production-ready", or when the user explicitly asks for iterative refinement. |
LLM agents produce their best work through iterative refinement rather than attempting perfection in a single pass. This skill implements a structured 5-pass approach: draft first (breadth), then four editorial passes focused on distinct quality dimensions.
Apply this methodology for:
- Documents, reports, or presentations that will be shared externally
- Code that needs to be production-ready
- Creative work where quality matters
- Any output where the user signals they want polished results
Skip for: quick answers, exploratory discussions, or when the user explicitly wants speed over polish.
Focus: SHAPE
Get the structure right. Breadth over depth. Don't aim for perfection—aim for coverage.
- Establish the overall architecture
- Hit all the major points
- Leave placeholders for details you'll fill later
- Bias toward inclusion (easier to cut than add)
Focus: TRUTH
Fix errors, bugs, and mistakes. Is the logic sound?
- Verify facts and claims
- Check calculations and code logic
- Ensure internal consistency
- Fix technical errors
Focus: UNDERSTANDING
Can someone else understand this? Simplify. Remove jargon.
- Restructure confusing sections
- Replace jargon with plain language (or define terms)
- Add transitions between ideas
- Ensure the reader can follow without your context
Focus: ROBUSTNESS
What could go wrong? What's missing? Handle the unusual.
- Consider failure modes
- Add error handling (for code)
- Address counterarguments (for writing)
- Fill gaps in coverage
Focus: POLISH
This is the last pass. Make it shine.
- Tighten prose, remove redundancy
- Improve naming and formatting
- Add finishing touches
- Ask: "Would I be proud to ship this?"
- Each pass should be a distinct mental mode—don't try to do everything at once
- The passes are sequential; each builds on the previous
- For shorter tasks, passes can be lighter; for substantial work, each deserves full attention
- Signal progress to the user between passes when working on longer content