@rfxlamia/skillkit 1.1.0 → 1.2.0

package/skills/skills/prompt-engineering/references/advanced-combinations.md

@@ -0,0 +1,444 @@
+# Advanced Techniques: Combining Methods
+
+## Overview
+
+Most real-world tasks benefit from combining multiple prompt engineering techniques. This guide shows proven combinations and when to use them.
+
+## Common Combinations
+
+### 1. Few-Shot + Chain of Thought (Few-Shot CoT)
+
+**When to use:** Need both format consistency AND transparent reasoning
+
+**Structure:**
+```
+Show examples with reasoning steps included in each example.
+```
+
+**Template:**
+```
+Task: Solve complex math word problems
+
+Example 1 (with reasoning):
+Problem: If 3 oranges cost $2, how much do 7 oranges cost?
+
+Reasoning:
+Step 1: Find cost per orange = $2 ÷ 3 = $0.67
+Step 2: Calculate for 7 oranges = $0.67 × 7 = $4.69
+Answer: $4.69
+
+Example 2 (with reasoning):
+Problem: A car travels 120 miles in 2 hours. At the same speed, 
+how long for 300 miles?
+
+Reasoning:
+Step 1: Find speed = 120 miles ÷ 2 hours = 60 mph
+Step 2: Calculate time = 300 miles ÷ 60 mph = 5 hours
+Answer: 5 hours
+
+Now solve (show your reasoning):
+Problem: If 5 apples cost $3, how much do 8 apples cost?
+Reasoning:
+```
+
+**Benefits:**
+- Consistent output format (from Few-Shot)
+- Transparent reasoning (from CoT)
+- Higher accuracy than either alone
+
+**Token cost:** Medium-High (200-800 tokens)
+
+---
+
+### 2. ReAct + Self-Consistency
+
+**When to use:** High-stakes decisions requiring both tool use AND verification
+
+**Structure:**
+```
+Use ReAct framework multiple times with different tool sequences,
+then compare results for consistency.
+```
+
+**Template:**
+```
+Problem: [Complex problem requiring both tools and high confidence]
+
+Attempt 1 (Tool Sequence A):
+Thought 1: [Reasoning]
+Action 1: [Tool/API call]
+Observation 1: [Result]
+...
+Final Answer 1: [Result]
+
+Attempt 2 (Tool Sequence B):
+Thought 1: [Different approach]
+Action 1: [Different tool or order]
+Observation 1: [Result]
+...
+Final Answer 2: [Result]
+
+Attempt 3 (Tool Sequence C):
+[Third approach]
+Final Answer 3: [Result]
+
+Consistency Check:
+Answers: [Answer 1, Answer 2, Answer 3]
+Agreement: [X/3 agree]
+Final Answer: [Most consistent result]
+```
+
+**Benefits:**
+- Tool use capability (from ReAct)
+- Verified answers (from Self-Consistency)
+- Higher reliability for critical tasks
+
+**Token cost:** Very High (1000-3000+ tokens)
+
+---
+
+### 3. Tree of Thoughts + XML (Claude-Specific)
+
+**When to use:** Complex planning for Claude with clear structure needed
+
+**Template:**
+```xml
+<tree_of_thoughts>
+  <problem>[Complex planning problem]</problem>
+  
+  <exploration_config>
+    <max_depth>3</max_depth>
+    <branches_per_level>3</branches_per_level>
+    <evaluation_criteria>
+      <criterion weight="0.4">Feasibility</criterion>
+      <criterion weight="0.3">Cost</criterion>
+      <criterion weight="0.3">Time</criterion>
+    </evaluation_criteria>
+  </exploration_config>
+  
+  <level_1>
+    <thought id="A">
+      <description>[Approach A]</description>
+      <evaluation>
+        <feasibility>0.9</feasibility>
+        <cost>0.6</cost>
+        <time>0.7</time>
+        <total_score>0.75</total_score>
+      </evaluation>
+    </thought>
+    <!-- More thoughts -->
+    <selected>A</selected>
+  </level_1>
+  
+  <!-- Continue levels -->
+</tree_of_thoughts>
+```
+
+**Benefits:**
+- Systematic exploration (from ToT)
+- Clear structure (from XML)
+- Claude-optimized format
+
+**Token cost:** High (800-2000 tokens)
+
+---
+
+### 4. Few-Shot + JSON (API Integration)
+
+**When to use:** Need format consistency + machine parsing
+
+**Template:**
+```json
+{
+  "task": "Extract structured data from text",
+  "examples": [
+    {
+      "input": "John Doe, 30 years old, works at Google",
+      "output": {
+        "name": "John Doe",
+        "age": 30,
+        "company": "Google"
+      }
+    },
+    {
+      "input": "Jane Smith, age 25, employed by Microsoft",
+      "output": {
+        "name": "Jane Smith",
+        "age": 25,
+        "company": "Microsoft"
+      }
+    }
+  ],
+  "new_input": "[Text to process]",
+  "output_schema": {
+    "type": "object",
+    "properties": {
+      "name": {"type": "string"},
+      "age": {"type": "integer"},
+      "company": {"type": "string"}
+    },
+    "required": ["name", "age", "company"]
+  }
+}
+```
+
+**Benefits:**
+- Format learning (from Few-Shot)
+- Machine-parseable (from JSON)
+- Schema validation possible
+
+**Token cost:** Medium (200-600 tokens)
+
+---
+
+### 5. Zero-Shot CoT + XML (Claude Simple Reasoning)
+
+**When to use:** Simple reasoning tasks for Claude without examples
+
+**Template:**
+```xml
+<reasoning_task>
+  <problem>
+    [State problem clearly]
+  </problem>
+  
+  <instruction>
+    Let's solve this step by step.
+  </instruction>
+  
+  <reasoning_steps>
+    <!-- LLM fills this in -->
+  </reasoning_steps>
+  
+  <answer>
+    <!-- LLM provides final answer -->
+  </answer>
+</reasoning_task>
+```
+
+**Benefits:**
+- Step-by-step reasoning (from CoT)
+- Clean structure (from XML)
+- No examples needed (from Zero-Shot)
+
+**Token cost:** Low-Medium (100-300 tokens)
+
+---
+
+## Decision Matrix: Which Combination?
+
+| Need | Combination | Token Cost |
+|------|-------------|------------|
+| Format + Reasoning | Few-Shot CoT | Medium-High |
+| Tools + Verification | ReAct + Self-Consistency | Very High |
+| Planning + Structure (Claude) | ToT + XML | High |
+| Format + Parsing | Few-Shot + JSON | Medium |
+| Simple Reasoning (Claude) | Zero-Shot CoT + XML | Low-Medium |
+
+---
+
+## Layering Techniques
+
+### Layer 1: Base Method
+Choose primary technique:
+- Zero-Shot / Few-Shot / CoT / ReAct / ToT / Self-Consistency
+
+### Layer 2: Format
+Choose output format:
+- Natural Language / JSON / XML / YAML
+
+### Layer 3: Enhancements (Optional)
+Add refinements:
+- Examples (if not Few-Shot)
+- Verification steps
+- Edge case handling
+- Output schema
+
+**Example Layering:**
+```
+Base: ReAct (tool use needed)
+Format: JSON (API integration)
+Enhancement: Add schema validation
+
+Result: ReAct + JSON + Schema
+```
+
+---
+
+## Anti-Patterns to Avoid
+
+### ❌ Don't Mix Contradictory Techniques
+
+**Bad:**
+```
+Use Zero-Shot (no examples)
+Here are 5 examples... (Few-Shot)
+But ignore the examples and reason from scratch (Contradiction)
+```
+
+### ❌ Don't Over-Engineer Simple Tasks
+
+**Bad:**
+```
+Use Tree of Thoughts + Self-Consistency + Few-Shot CoT 
+to answer: "What is 2 + 2?"
+
+Token cost: 2000+
+Correct approach: Zero-Shot, 10 tokens
+```
+
+### ❌ Don't Combine Incompatible Formats
+
+**Bad:**
+```xml
+<!-- XML structure -->
+<prompt>
+  {
+    "json": "inside xml"  <!-- Confusing mix -->
+  }
+</prompt>
+```
+
+**Good:**
+```xml
+<prompt>
+  <data_format>JSON</data_format>
+  <expected_output>
+    <!-- Describe JSON, don't embed it -->
+  </expected_output>
+</prompt>
+```
+
+---
+
+## Progressive Enhancement
+
+Start simple, add complexity only if needed:
+
+### Level 1: Start Simple
+```
+Zero-Shot with Natural Language
+```
+
+### Level 2: Add Examples (if output quality poor)
+```
+Few-Shot with Natural Language
+```
+
+### Level 3: Add Reasoning (if logic unclear)
+```
+Few-Shot CoT with Natural Language
+```
+
+### Level 4: Add Structure (if Claude or complex)
+```
+Few-Shot CoT with XML
+```
+
+### Level 5: Add Verification (if high-stakes)
+```
+Few-Shot CoT + Self-Consistency with XML
+```
+
+**Rule:** Only progress to next level if previous level insufficient.
+
+---
+
+## Real-World Combination Examples
+
+### Example 1: Medical Diagnosis Support
+
+**Combination:** Few-Shot CoT + Self-Consistency + XML
+
+```xml
+<medical_diagnosis>
+  <examples>
+    <example>
+      <symptoms>[Symptom list]</symptoms>
+      <reasoning>
+        <step>Step 1: Consider symptom cluster</step>
+        <step>Step 2: Differential diagnosis</step>
+        <step>Step 3: Most likely condition</step>
+      </reasoning>
+      <diagnosis>[Diagnosis]</diagnosis>
+    </example>
+  </examples>
+  
+  <new_case>
+    <symptoms>[Patient symptoms]</symptoms>
+  </new_case>
+  
+  <instruction>
+    Provide 3 independent diagnostic approaches,
+    then select most consistent diagnosis.
+  </instruction>
+</medical_diagnosis>
+```
+
+**Why this combination:**
+- Few-Shot CoT: Show diagnostic reasoning pattern
+- Self-Consistency: Critical decision needs verification
+- XML: Claude-optimized, clear structure
+
+---
+
+### Example 2: Code Generation with Validation
+
+**Combination:** Few-Shot + ReAct + JSON
+
+```json
+{
+  "task": "Generate and validate Python function",
+  "examples": [
+    {
+      "request": "Function to calculate factorial",
+      "code": "def factorial(n): ...",
+      "tests": ["assert factorial(5) == 120"]
+    }
+  ],
+  "new_request": "[User request]",
+  "workflow": {
+    "step_1": {
+      "action": "GENERATE_CODE",
+      "thought": "Based on examples, create function"
+    },
+    "step_2": {
+      "action": "VALIDATE_SYNTAX",
+      "thought": "Check for syntax errors"
+    },
+    "step_3": {
+      "action": "RUN_TESTS",
+      "thought": "Execute test cases"
+    },
+    "step_4": {
+      "action": "FINISH",
+      "output": "Validated code"
+    }
+  }
+}
+```
+
+**Why this combination:**
+- Few-Shot: Show code style and pattern
+- ReAct: Multi-step with validation actions
+- JSON: Machine-parseable output
+
+---
+
+## Quick Reference
+
+| Scenario | Best Combination | Cost |
+|----------|-----------------|------|
+| Math word problems | Few-Shot CoT + Natural Language | Medium |
+| Medical diagnosis | Few-Shot CoT + Self-Consistency + XML | Very High |
+| API data extraction | Few-Shot + JSON | Medium |
+| Strategic planning (Claude) | ToT + XML | High |
+| Tool use + verification | ReAct + Self-Consistency | Very High |
+| Simple Claude reasoning | Zero-Shot CoT + XML | Low |
+
+---
+
+**Related:**
+- Individual technique guides for details
+- [Decision Matrix](decision_matrix.md) - Choosing techniques
+- [Pitfalls](pitfalls.md) - Avoid anti-patterns

package/skills/skills/prompt-engineering/references/chain-of-thought.md

@@ -0,0 +1,140 @@
+# Chain of Thought (CoT) Prompting
+
+## Overview
+
+Chain of Thought prompting encourages the LLM to break down complex problems into intermediate reasoning steps, making the thought process explicit and transparent.
+
+## When to Use
+
+✅ **Use CoT When:**
+- Multi-step reasoning required
+- Mathematical or logical problems
+- Need transparency in reasoning
+- Complex decision-making tasks
+- Problem involves calculations or deductions
+
+❌ **Don't Use CoT When:**
+- Simple, single-step tasks
+- Just need final answer quickly
+- Token budget is extremely tight
+- Task doesn't benefit from step-by-step breakdown
+
+## Variants
+
+### 1. Zero-Shot CoT
+Simply add "Let's think step by step" to prompt.
+```
+Problem: [Your problem]
+Let's think step by step:
+```
+
+### 2. Few-Shot CoT
+Provide examples with reasoning steps shown.
+```
+Example:
+Problem: [Example problem]
+Step 1: [First reasoning step]
+Step 2: [Second reasoning step]
+Answer: [Result]
+
+Now solve:
+Problem: [Your problem]
+```
+
+### 3. Auto-CoT
+Let model generate reasoning automatically with minimal guidance.
+
+## Template: Zero-Shot CoT
+
+```
+Problem: [State the problem clearly]
+
+Let's solve this step by step:
+
+1. First, let's identify what we know:
+   [List known information]
+
+2. Next, let's determine what we need to find:
+   [List unknowns]
+
+3. Now, let's work through the solution:
+   Step 1: [First reasoning step]
+   Step 2: [Second reasoning step]
+   Step 3: [Third reasoning step]
+
+4. Finally, let's verify our answer:
+   [Verification step]
+
+Answer: [Final answer]
+```
+
+## Template: Few-Shot CoT (XML for Claude)
+
+```xml
+<chain_of_thought>
+  <examples>
+    <example>
+      <problem>[Problem 1]</problem>
+      <reasoning>
+        <step number="1">[First step]</step>
+        <step number="2">[Second step]</step>
+        <step number="3">[Third step]</step>
+      </reasoning>
+      <answer>[Answer 1]</answer>
+    </example>
+  </examples>
+
+  <new_problem>
+    [Your actual problem]
+  </new_problem>
+
+  <instruction>
+    Please solve using the same step-by-step reasoning approach.
+  </instruction>
+</chain_of_thought>
+```
+
+## Real-World Example
+
+### Math Problem
+
+```
+Problem: A store offers 20% discount on items over $100, and additional 10% member discount. 
+If an item costs $150 and I'm a member, what's the final price?
+
+Let's think step by step:
+
+1. Original price: $150
+2. First, apply 20% discount (item over $100): 
+   $150 × 0.80 = $120
+3. Then, apply 10% member discount on discounted price: 
+   $120 × 0.90 = $108
+4. Verification: Total discount = $150 - $108 = $42 (28%)
+   This makes sense: 20% + 10% of remaining ≈ 28%
+
+Answer: $108
+```
+
+## Best Practices
+
+✓ Break down into clear, numbered steps
+✓ Show all intermediate calculations
+✓ Make assumptions explicit
+✓ Verify answer makes sense
+✓ Use consistent step format
+
+✗ Don't skip critical steps
+✗ Avoid jumping to conclusions
+✗ Don't mix multiple calculations in one step
+
+## Token Cost
+
+- Typical: 100-500 tokens
+- Trade-off: More tokens for better accuracy and transparency
+
+---
+
+**Related:**
+- [Few-Shot](few-shot.md) - Combine with CoT for Few-Shot CoT
+- [Self-Consistency](self-consistency.md) - Multiple CoT paths
+- [Decision Matrix](decision_matrix.md) - When to use CoT