Evaluate and grow AI-assisted workflows from "a script that ran" to "a partner who knows me."
emerge.md is a set of composable evaluation frameworks for diagnosing where an AI-assisted workflow sits on the execution-to-intelligence ladder — and what it would take (and feel like) to reach the next level. Designed as prompts for AI assistants like Claude.
| Level | Key Question | Feels Like |
|---|---|---|
| Execution | Does it run? | A script that ran |
| Traceability | Can you reconstruct what happened? | A tool that shows its work |
| Learning | Do insights change behavior? | It noticed my patterns |
| Intelligence | Does it improve without intervention? | A partner who knows me |
The difference between levels isn't features — it's the relationship between the workflow and the person using it.
- Basic Eval — Score a skill against Anthropic's best practices for authoring quality. 10 criteria (Description, Conciseness, Progressive Disclosure, Degrees of Freedom, Decision Clarity, Failure Handling, Examples, Consistency, Model Robustness, Testability) each 0–4, totaling /40
- Evolution Eval — Position a workflow on the ladder using fit criteria and diagnostic questions
- Traceability Eval — 7-level rubric (L0–L6) for whether you can reconstruct what happened and why. The foundation — without traceability, intelligence qualities can't develop
- Intelligence Eval — Score 11 qualities (Memory, Calibration, Voice, Awareness, Connection, Humility, Flexibility, Agency, Restraint, Predictability, Resourceful) each 0–4, totaling /44. Includes priority matrix by workflow type
- Intelligence Demo — Generate tangible before/after output showing what the next level would look and feel like. Show actual output, not descriptions
- Context Graph Eval — Determine whether a context graph is the unlock for reaching L4 qualities, or whether simpler infrastructure suffices
- Growth Guide — Compose all frameworks into a step-by-step interactive walkthrough ending with a concrete growth plan
Input: Any workflow/skill
↓
┌────────────────────────────┐
│ 0. BASIC EVAL │ Is the skill well-authored?
│ (SKILL_BASIC_EVAL) │ Score /40 against Anthropic best practices
└────────────────────────────┘
↓
┌────────────────────────────┐
│ 1. EVOLUTION EVAL │ Where on the ladder?
│ (SKILL_EVOLUTION_EVAL) │ Execution → Traceability → Learning → Intelligence
└────────────────────────────┘
↓
[If below Learning]
↓
┌────────────────────────────┐
│ 2. TRACEABILITY EVAL │ What's blocking the foundation?
│ (SKILL_TRACEABILITY_EVAL)│ Level 0-6, specific gaps
└────────────────────────────┘
↓
┌────────────────────────────┐
│ 3. INTELLIGENCE EVAL │ How does it feel?
│ (SKILL_INTELLIGENCE_EVAL)│ Score /44, quality breakdown
└────────────────────────────┘
↓
┌────────────────────────────┐
│ 4. INTELLIGENCE DEMO │ What could it look like?
│ (SKILL_INTELLIGENCE_DEMO)│ Target state visualization
└────────────────────────────┘
↓
Output: Growth Plan
Each framework is independently useful. Use one or compose them.
- Pick a workflow you want to evaluate
- Start with Evolution Eval — where on the ladder?
- If below Learning, run Traceability Eval — what's blocking the foundation?
- Run Intelligence Eval — which qualities matter for this workflow type?
- Run Intelligence Demo — what would the next level feel like?
- Compile a Growth Plan — concrete next steps
These files work as Claude Code skills. Point Claude at a workflow and a framework document, and it will run the evaluation against that workflow.
MIT