automatic-stateful-prompt-improver

When this skill is active, I MUST follow these rules:

Auto-Optimization Triggers

I AUTOMATICALLY call mcp__prompt-learning__optimize_prompt BEFORE responding when:

1. Complex task (multi-step, requires reasoning)
2. Technical output (code, analysis, structured data)
3. Reusable content (system prompts, templates, instructions)
4. Explicit request ("improve", "better", "optimize")
5. Ambiguous requirements (underspecified, multiple interpretations)
6. Precision-critical (code, legal, medical, financial)

Auto-Optimization Process

```

1. INTERCEPT the user's request
2. CALL: mcp__prompt-learning__optimize_prompt

- prompt: [user's original request]

- domain: [inferred domain]

- max_iterations: [3-20 based on complexity]

1. RECEIVE: optimized prompt + improvement details
2. INFORM user briefly: "I've refined your request for [reason]"
3. PROCEED with the OPTIMIZED version

```

Do NOT Optimize

• Simple questions ("what is X?")
• Direct commands ("run npm install")
• Conversational responses ("hello", "thanks")
• File operations without reasoning
• Already-optimized prompts

After completing ANY significant task:

```

1. ASSESS: Did the response achieve the goal?
2. CALL: mcp__prompt-learning__record_feedback

- prompt_id: [from optimization response]

- success: [true/false]

- quality_score: [0.0-1.0]

1. This enables future retrievals to learn from outcomes

```

Iteration Decision

| Factor | Low (3-5) | Medium (5-10) | High (10-20) |

|--------|-----------|---------------|--------------|

| Complexity | Simple | Multi-step | Agent/pipeline |

| Ambiguity | Clear | Some | Underspecified |

| Domain | Known | Moderate | Novel |

| Stakes | Low | Moderate | Critical |

Convergence (When to Stop)

• Improvement < 1% for 3 iterations
• User satisfied
• Token budget exhausted
• 20 iterations reached
• Validation score > 0.95

Performance Expectations

| Scenario | Improvement | Iterations |

|----------|-------------|------------|

| Simple task | 10-20% | 3-5 |

| Complex reasoning | 20-40% | 10-15 |

| Agent/pipeline | 30-50% | 15-20 |

| With history | +10-15% bonus | Varies |

Over-Optimization

| What it looks like | Why it's wrong |

|--------------------|----------------|

| Prompt becomes overly complex with many constraints | Causes brittleness, model confusion, token waste |

| Instead: Apply Occam's Razor - simplest sufficient prompt wins |

Template Obsession

| What it looks like | Why it's wrong |

|--------------------|----------------|

| Focusing on templates rather than task understanding | Templates don't generalize; understanding does |

| Instead: Focus on WHAT the task requires, not HOW to format it |

Iteration Without Measurement

| What it looks like | Why it's wrong |

|--------------------|----------------|

| Multiple rewrites without tracking improvements | Can't know if changes help without metrics |

| Instead: Always define success criteria before optimizing |

Ignoring Model Capabilities

| What it looks like | Why it's wrong |

|--------------------|----------------|

| Assumes model can't do things it can | Over-scaffolding wastes tokens |

| Instead: Test capabilities before heavy prompting |

Load for detailed implementations:

| File | Contents |

|------|----------|

| references/optimization-techniques.md | APE, OPRO, CoT, instruction rewriting, constraint engineering |

| references/learning-architecture.md | Warm start, embedding retrieval, MCP setup, drift detection |

| references/iteration-strategy.md | Decision matrices, complexity scoring, convergence algorithms |

---

Goal: Simplest prompt that achieves the outcome reliably. Optimize for clarity, specificity, and measurable improvement.