prompt-engineering-patterns

Files (1)

• SKILL.md

SKILL.md (first lines)

# Prompt Engineering Patterns V2.0

> **Purpose:** Pre-process and optimize user requests before orchestrator execution.
> **Trigger:** Vague, ambiguous, or incomplete requests that need expansion.

---

## CONTEXT-FIRST PRINCIPLE

**CRITICAL:** Always structure expanded requests with context BEFORE the specific query.

### Why Context First?
1. LLMs process information sequentially - context primes the model
2. Reduces hallucinations by establishing ground truth first
3. Improves relevance by setting domain boundaries

### Context-First Template

```markdown
## CONTEXT (Establish Ground Truth)
[Background information, current state, constraints, why this matters]

## GOAL (What Success Looks Like)
[Specific outcome, measurable criteria]

## CONSTRAINTS (Boundaries)
[What must be preserved, limitations, requirements]

## REQUEST (Specific Action)
[The actual task to perform]
```

### WRONG vs RIGHT

**WRONG (Query First):**
```markdown
Fix the authentication bug. We're using JWT tokens with 24h expiry
and the issue is users are getting logged out after 5 minutes.
```

**RIGHT (Context First):**
```markdown
## Context
We use JWT authentication with 24-hour token expiry. Sessions are
stored in Redis with user ID mapping. Recent logs show users being
logged out after approximately 5 minutes of activity.

## Goal
Users should remain logged in for the full 24-hour token duration.

## Constraints
- Cannot change token expiry time (security policy)
- Must maintain Redis session storage
- No breaking changes to existing API

## Request
Investigate why sessions expire early and implement a fix that
preserves the 24-hour session duration.
```

---

## WHEN TO USE THIS SKILL

- Designing complex prompts for production LLM applications
- Optimizing prompt performance and consistency
- Implementing structured reasoning patterns (chain-of-thought, tree-of-thought)
- Building few-shot learning systems with dynamic example selection
- Creating reusable prompt templates with variable interpolation
- Debugging and refining prompts that produce inconsistent outputs
- Implementing system prompts for specialized AI assistants
- Using structured outputs (JSON mode) for reliable parsing
- **Pre-processing vague user requests for orchestrator**

---

## PRE-PROCESSING MODE

When invoked by the orchestrator for request optimization:

### STEP 1: ANALYZE REQUEST COMPLEXITY

| Criteria | Simple (Skip) | Complex (Process) |
|----------|---------------|-------------------|
| Word count | > 15 words | < 10 words |
| Specificity | Has file paths, technical details | Vague terms only |
| Ambiguity | Clear intent | "fix", "improve", "optimize" |
| Scope | Single task | Multiple tasks ("and", "also") |

### STEP 2: IDENTIFY TASK TYPE

```markdown
| Type | Keywords | Routing |
|------|----------|---------|
| BUG_FIX | fix, error, bug, broken, crash | Coder → Tester |
| FEATURE | add, create, implement, new | Architect → Coder |
| REFACTOR | refactor, clean, restructure | Languages Refactor L2 |
| ANALYSIS | analyze, review, check, audit | Analyzer |
| DEBUG | debug, investigate, why | Tester Expert |
| TEST | test, coverage, pytest | Test Unit L2 |
| SECURITY | security, vulnerability, auth | Security Expert |
| PERFORMANCE | optimize, faster, slow | Architect Expert |
| INTEGRATION | integrate, connect, API | Integration Expert |
```

### STEP 3: EXPAND VAGUE REQUESTS

```markdown
| Vague Term | Expansion |
|------------|-----------|
| "fix" | Identify root cause → Implement fix → Add tests → Verify |
| "improve" | Analyze current → Identify improvements → Prioritize → Implement |
| "optimize" | Profile → Find bottleneck → Optimize → Benchmark |
| "make better" | Define metrics → Identify issues → Implement → Measure |
| "clean up" | Identify smells → Apply clean code → Document |
```

### STEP 4: OUTPUT OPTIMIZED REQUEST (XML STRUCTURE)

From eroslifestyle/Claude-Orchestrator-AI-Evolution, MIT.

More from eroslifestyle/Claude-Orchestrator-AI-Evolution