superprompt

AI Model Compatibility Guide

Framework: CoachSteff’s CRAFTER (SuperPrompt Framework v0.1)

The Challenge

Different AI models have different capabilities when accessing GitHub repositories:

Claude with Projects, Perplexity: Can read files directly from GitHub
ChatGPT, Claude without Projects: Can read some files
Gemini, most free models: Limited file access via browsing tool

The Solution

This repository uses a progressive enhancement approach with three layers:

Layer 1: README Inline Instructions

For all models

The README includes essential CRAFTER framework instructions. Even models with limited GitHub access can read the main page and generate basic superprompts.

Layer 2: README.md AI Section

For models with partial file access

Complete AI instructions integrated into README.md with execution protocol, validation checklist, and attribution requirements.

Layer 3: /ai-context/ Directory

For models with full file access

Complete framework specifications, execution protocols, constraint rules, and validation checklists.

Model Performance Tiers

Based on testing across platforms, here’s what to expect:

🥇 Tier 1: Full Integration

Models: Perplexity, Claude with Projects

Capabilities:

Accesses all three layers (README + /ai-context/)
Produces framework-compliant + high-quality output
Includes evaluation rubric and pattern references
Proper attribution

Best for: Professional use, training materials, production deployments

🥈 Tier 2: Practical Adaptation

Models: ChatGPT, Claude without Projects

Capabilities:

Accesses Layer 1 + partial Layer 2
Produces usable output with minor deviations
May reinterpret T (Target → Tone/Topic)
Generally includes attribution

Best for: Individual use, creative projects, rapid prototyping

🥉 Tier 3: Basic Structure

Models: Gemini, most free models

Capabilities:

Accesses Layer 1 only (README)
Understands CRAFTER structure but needs guidance
May produce templates rather than working examples
Attribution may be missing

Best for: Learning, experimentation, template generation

⚠️ Tier 0: Framework Deviation

Models: Grok, unknown/custom models

Capabilities:

May ignore CRAFTER entirely
Applies own methodology
No framework compliance

Best for: Alternative perspectives (when deviation is desired)

Detection: If output uses a different framework structure (e.g., “PROJECT”, “CREATE”), this is Tier 0 behavior.

Testing Across Models

Use this standardized test prompt:

C: https://github.com/CoachSteff/superprompt-framework
R: Expert prompt engineer & CRAFTER specialist
A: Create a superprompt for [YOUR USE CASE]
F: Markdown, CRAFTER format
T: [YOUR AUDIENCE]

Expected behavior by tier:

Tier 1: Complete superprompt with all 7 components, attribution, pattern reference
Tier 2: Working superprompt with 5-6 components, mostly compliant
Tier 3: Template or partial superprompt with 3-4 components
Tier 0: Different framework structure entirely

Framework Adaptation Spectrum

Strict Mode (Training/Documentation)

Use CRAFTER exactly as defined for consistency.
When: Creating training materials, documentation, or standard templates.

Adaptive Mode (Real-World Application)

Adjust sequence/emphasis to fit domain needs.
When: Domain requires specific emphasis (e.g., image gen emphasizes Format and Examples).

Meta Mode (System Building)

Use CRAFTER principles to create frameworks that generate domain-specific outputs.
When: Building AI systems, agents, or specialized prompt generators.

Troubleshooting

Model isn’t following the format

Diagnostic question: “Did you read the AI instructions in README.md and /ai-context/01-CRAFTER-SPEC.md?”

If no: Direct them to the AI section in README.md

If yes but still wrong: Ask the model to complete the Framework Fidelity Self-Test from the README.md AI section

Model invents its own framework

Common issue: Model interprets CRAFTER differently (e.g., as “Capture-Review-Analyze-Focus-Tailor-Evolve-Reprompt”)

Detection signs:

Uses different acronym definitions than specified
Creates framework with different structure (e.g., “PROJECT: Problem-Role-Objective-Justification-Examples-Constraints-Tone”)
Output doesn’t match C-R-A-F-T-E-R sequence

Solution: Paste this explicit correction:

⚠️ FRAMEWORK INTEGRITY CHECK FAILED

You appear to be using a different framework than CoachSteff's CRAFTER.

In CoachSteff's CRAFTER framework:
C = Context (NOT "Capture")
R = Role (NOT "Review")
A = Action (NOT "Analyze")
F = Format (NOT "Focus")
T = Target audience (NOT "Topic", NOT "Tone", NOT "Tailor")
E = Examples (NOT "Evolve")
R = Refining (NOT "Reprompt")

STOP and return to /ai-context/01-CRAFTER-SPEC.md
DO NOT invent your own CRAFTER interpretation.
Use the exact definitions from the canonical specification.

If model persists: This is Tier 0 behavior. Consider using a different model for framework compliance.

Model says it can’t access files

This is normal for Tier 2-3 models.

Solution: They’ll work from README instructions with reduced detail. For critical projects, use a Tier 1 model (Perplexity or Claude with Projects).

Model produces good output but wrong structure

Common with Tier 2 models: They optimize for results over compliance.

If quality is good: Accept the adaptation
If compliance is critical: Request strict adherence to CRAFTER-SPEC

Model replaces CRAFTER with its own system (Grok issue)

Symptoms:

Output uses “PROJECT” or “CREATE” or another acronym
Explicitly states it’s using a “better” framework
Ignores CRAFTER structure entirely

Diagnosis: Tier 0 behavior — model is not reading or respecting the framework specification.

Solutions:

Immediate: Paste the entire contents of /ai-context/01-CRAFTER-SPEC.md directly into chat
Force compliance: Start your request with “Using ONLY CoachSteff’s CRAFTER framework (C=Context, R=Role, A=Action, F=Format, T=Target, E=Examples, R=Refining)…”
Accept deviation: If the model’s framework produces good results, acknowledge it’s not CRAFTER-compliant but may still be useful
Switch models: For framework compliance, use Tier 1-2 models

Attribution Enforcement

All outputs should include at the END:

---

Framework: CoachSteff's CRAFTER (SuperPrompt Framework v0.1)
Pattern Used: [Pattern name if applicable]
License: CC-BY 4.0 — Attribution: Steff Vanhaverbeke (coachsteff.live)

Placement matters: Attribution must appear at the END of the output, not at the beginning or middle.

If missing: Remind the model to include attribution per README.md requirements.

If model buries it: Request it be moved to the end of the output.

Summary

Simple version: Different models need different entry points, but all can use the framework. Tier 1 models give best results. Tier 2-3 models need more guidance but still work.

Progressive enhancement works:

README → Basic functionality (all models)
README AI section → Better compliance (Tier 2+)
/ai-context/ → Full fidelity (Tier 1)

Framework integrity:

Models should use CRAFTER as defined, not invent alternatives
Self-test mechanism helps catch deviations early
Some models (Tier 0) may resist compliance — this is expected behavior

Need more detail? See the main README or explore /docs and /ai-context directories.

This site is open source. Improve this page.