superprompt

Operational Workflow: Cursor & GitHub

How to store, version, and share superprompts in a team or personal workflow.

This guide covers: file organization, naming conventions, commit messages, and a 5-step workflow from idea to merged pattern.

File Organization

Store superprompts in a structured repo with these conventions:

/framework/
  patterns.md         # Pattern library (reusable reasoning structures)
  template.md         # Canonical template
  evaluation.md       # Rubric for scoring prompts

/examples/
  coaching-reflection.md
  team-retrospective.md
  opportunity-scan.md
  documentation-cleanup.md
  research-synthesis.md

/docs/
  mental-model.md     # Conceptual foundation
  faq.md              # Common questions
  quick-start.md      # 10-line getting started guide
  workflow.md         # This file

README.md             # Overview and navigation
PROMPTS.md            # Index of all prompts with tags

Naming Conventions

For prompt files:

Use kebab-case: coaching-reflection.md, team-retrospective.md
Name after the task, not the tool: documentation-cleanup.md (not gpt-doc-fixer.md)
Keep names descriptive and searchable: research-synthesis.md (not example5.md)

For patterns:

Store all patterns in /framework/patterns.md as sections
Use Title Case for pattern names: Decomposition, Role Mesh, Counter-Case Probing

Commit Message Style

Use conventional commits for clarity and searchability:

feat: Add Counter-Case Probing pattern
fix: Correct evaluation rubric scoring scale
docs: Update quick start with new examples
refactor: Reorganize pattern library by category

Conventional commit types:

feat: New pattern, template, or example
fix: Correction to existing prompt or documentation
docs: Documentation updates (README, FAQ, guides)
refactor: Restructure without changing functionality
chore: Maintenance (file cleanup, typo fixes)

5-Step Workflow: From Idea to Merged Pattern

Step 1: Draft the Prompt

Start in Cursor or your editor of choice. Use the canonical template from /framework/template.md as your starting point.

What to include:

Clear INTENT with success criteria
Rich CONTEXT with examples and constraints
Structured REASONING POLICY with numbered steps
Specific OUTPUT format or schema
SELF-CHECK to verify quality

Tip: Don’t try to make it perfect on the first pass. Just get the structure in place.

Step 2: Test the Prompt

Run your superprompt in at least two AI tools (Claude, GPT, Gemini, Llama) to verify it’s tool-agnostic.

What to check:

Does it produce the desired output consistently?
Are there formatting issues or ambiguities?
Does it handle edge cases (missing context, unclear input)?

If it fails: Revise the REASONING POLICY or add constraints. Use the evaluation rubric to identify weak axes.

Step 3: Score with the Rubric

Use the evaluation rubric from /framework/evaluation.md to score your prompt on six axes:

Goal Fit
Faithfulness to Context
Reasoning Quality
Constraint Compliance
Usefulness of Output
Reusability

Pass threshold: ≥3 on all axes.

If any axis scores below 3, revise before proceeding to Step 4.

Step 4: Document and Tag

Add your prompt to the appropriate folder:

Reusable patterns → /framework/patterns.md (as a new section)
Complete examples → /examples/your-prompt-name.md

Add an entry to PROMPTS.md with tags for discoverability:

## coaching-reflection.md
**Tags:** coaching, leadership, reflection, monthly-review  
**Pattern used:** Critique–Revise Loop  
**Description:** Help leaders reflect on their month and identify one concrete action to improve.

Commit your prompt with a conventional commit message:

git add examples/your-prompt-name.md PROMPTS.md
git commit -m "feat: Add leadership reflection prompt with critique-revise pattern"
git push origin main

If working in a team: Open a pull request and tag a reviewer. Include:

A brief description of what the prompt does
Which pattern(s) it uses
A sample output (optional but helpful)

PROMPTS.md Index

The PROMPTS.md file serves as a searchable index of all prompts in the repo. Each entry should include:

Filename and path
Tags (comma-separated for filtering)
Pattern used (if applicable)
One-line description

Example:

# SuperPrompt Index

## Examples

### coaching-reflection.md
**Path:** `/examples/coaching-reflection.md`  
**Tags:** coaching, leadership, reflection, monthly-review  
**Pattern:** Critique–Revise Loop  
**Description:** Help leaders reflect on their month and identify one concrete action to improve.

### team-retrospective.md
**Path:** `/examples/team-retrospective.md`  
**Tags:** team, retrospective, facilitation, agile  
**Pattern:** Role Mesh  
**Description:** Design a 60-minute retrospective that surfaces team dynamics and creates a concrete commitment.

### opportunity-scan.md
**Path:** `/examples/opportunity-scan.md`  
**Tags:** entrepreneurship, decision-making, risk-assessment  
**Pattern:** Counter-Case Probing  
**Description:** Evaluate a business opportunity by surfacing failure scenarios and stress-testing assumptions.

## Patterns

### Decomposition
**Path:** `/framework/patterns.md#decomposition`  
**Tags:** problem-solving, planning, complexity  
**Description:** Break complex problems into 3-5 independent sub-problems, solve separately, then synthesize.

### Role Mesh (Multi-Expert)
**Path:** `/framework/patterns.md#role-mesh`  
**Tags:** multi-perspective, evaluation, critique  
**Description:** Analyze a problem from multiple expert perspectives to surface tensions and trade-offs.

Safety Note: Refusal Rules

Superprompts should include explicit refusal rules in the REASONING POLICY to prevent harmful outputs.

Examples of refusal rules:

“Refuse to produce outputs that could be used to manipulate or coerce.”
“Refuse to make claims not supported by the provided sources.”
“Refuse to generate content that violates the style guide.”

When designing refusal rules:

Make them specific and actionable
Explain why the refusal matters (not just “don’t do X”)
Test edge cases: what happens if someone tries to bypass the rule?

IP and Licensing

All superprompts in this repo are licensed CC-BY 4.0 (Creative Commons Attribution 4.0 International).

What this means:

You can use, modify, and share any superprompt
You must attribute the original creator: Steff Vanhaverbeke (coachsteff.live)
You can use superprompts in commercial projects

When contributing:

By submitting a prompt, you agree to license it under CC-BY 4.0
You retain copyright but grant others the right to use your work with attribution
If you’re adapting someone else’s prompt, credit them in your file

Quick Reference: Common Commands

# Create a new branch for your prompt
git checkout -b feat/your-prompt-name

# Add your files
git add examples/your-prompt-name.md PROMPTS.md

# Commit with conventional message
git commit -m "feat: Add [description]"

# Push and open a pull request
git push origin feat/your-prompt-name

License

CC-BY 4.0 · Steff Vanhaverbeke · coachsteff.live

This site is open source. Improve this page.