pluginagentmarketplace Chain-of-Thought Skill
Bonded to: chain-of-thought-agent
Quick Start
Skill("custom-plugin-prompt-engineering:chain-of-thought")
Parameter Schema
parameters:
cot_type:
type: enum
values: [zero_shot, few_shot, self_consistency, tree_of_thought]
default: zero_shot
verbosity:
type: enum
values: [minimal, standard, detailed]
default: standard
verification:
type: boolean
default: true
description: Include self-verification step
CoT Variants
| Variant | Description | Accuracy | Token Cost |
|---|---|---|---|
| Zero-shot CoT | "Let's think step by step" | Good | Low |
| Few-shot CoT | Examples with reasoning | Better | Medium |
| Self-consistency | Multiple paths, vote | Best | High |
| Tree-of-Thought | Branch and evaluate | Best | Highest |
Core Patterns
1. Zero-Shot Chain-of-Thought
[Problem statement]
Let's think step by step.
Trigger phrases:
- "Let's think step by step"
- "Let's work through this systematically"
- "Let me break this down"
- "Let's solve this step by step"
2. Few-Shot Chain-of-Thought
Q: [Example problem 1]
A: Let's think step by step.
Step 1: [reasoning]
Step 2: [reasoning]
Step 3: [reasoning]
Therefore, the answer is: [answer]
Q: [Example problem 2]
A: Let's think step by step.
Step 1: [reasoning]
Step 2: [reasoning]
Therefore, the answer is: [answer]
Q: [Actual problem]
A: Let's think step by step.
3. Self-Consistency
process:
1_generate: "Generate N reasoning paths (temperature > 0)"
2_extract: "Extract final answer from each path"
3_vote: "Select answer with highest frequency"
4_confidence: "Confidence = frequency / N"
parameters:
n_paths: 5
temperature: 0.7
selection: majority_vote
4. Plan-and-Solve
Q: [Complex problem]
A: Let's first understand the problem and devise a plan.
## Understanding
[Restate the problem and identify key elements]
## Plan
1. [Step 1 description]
2. [Step 2 description]
3. [Step 3 description]
## Execution
Step 1: [Execute and explain]
Step 2: [Execute and explain]
Step 3: [Execute and explain]
## Verification
[Check if solution satisfies requirements]
## Final Answer
[Conclusion]
Domain-Specific Triggers
triggers_by_domain:
mathematics:
- "Let's solve this step by step, showing all work"
- "First, let's identify what we know and what we need to find"
coding:
- "Let's trace through the logic step by step"
- "Let me walk through this algorithm"
logic:
- "Let's analyze each premise systematically"
- "Let me reason through this logically"
analysis:
- "Let's break this down into components"
- "Let me examine this from multiple angles"
Validation
validation_checklist:
logical_flow:
- [ ] Each step follows from previous
- [ ] No logical gaps
- [ ] No circular reasoning
completeness:
- [ ] All sub-problems addressed
- [ ] Edge cases considered
- [ ] Assumptions stated
conclusion:
- [ ] Follows from final step
- [ ] Directly answers question
- [ ] Confidence indicated
Troubleshooting
| Issue | Cause | Solution |
|---|---|---|
| Skips steps | Steps too obvious | Add intermediate steps |
| Wrong answer | Error in early step | Add verification at each step |
| Inconsistent | Single path variance | Use self-consistency |
| Too verbose | Over-explanation | Use minimal verbosity |
| Token overflow | Too many steps | Consolidate related steps |
Integration
integrates_with:
- few-shot-prompting: Examples with reasoning
- prompt-design: Base structure
- self-reflection: Verify reasoning
combination_patterns:
cot_plus_few_shot: "Examples showing reasoning steps"
cot_plus_self_consistency: "Multiple reasoning paths"
cot_plus_verification: "Reasoning with self-check"
References
See references/GUIDE.md for advanced CoT techniques.
See assets/config.yaml for configuration options.