prompt-engineering

Prompt Engineering

Comprehensive prompt engineering guidance covering Claude (Anthropic), Gemini (Google), and Grok (xAI). Each model has unique strengths and optimal prompting patterns.

When to Invoke This Skill

Use this skill when:

Crafting prompts for Claude, Gemini, or Grok models
Leveraging model-specific capabilities (XML tags, system instructions, real-time knowledge)
Working with long-context tasks (up to 1M+ tokens)
Implementing extended thinking or reasoning features
Building agentic workflows with parallel tool use
Designing prompts for multimodal inputs (text, images, audio, video)

Quick Model Selection

Model	Best For	Key Strength	Prompt Style
Claude Sonnet 4.5	Balanced tasks, long-context, instruction following	XML-style structure, extended thinking, parallel tools	Structured with XML tags
Gemini 3 Flash/Pro	Multimodal, ultra-long context, system instructions	Native multimodal, 1M+ context, JSON schema output	System instruction based
Grok	Current events, controversial topics, creative work	Real-time X (Twitter) access, witty personality	Conversational, direct

Universal Prompting Techniques

These techniques work across all models with adaptations noted below.

1. Zero-Shot Prompting

Claude (XML-style):

<task>
Extract the key dates and events from the following text:
</task>

<text>
[paste text]
</text>

<output_format>
JSON with keys "date", "event", "description"
</output_format>

Gemini (System Instructions):

{
  "system_instruction": {
    "parts": [{"text": "You are a data extraction specialist. Output JSON only."}]
  },
  "contents": [{"parts": [{"text": "Extract dates from: [text]"}]}]
}

Grok (Conversational):

Hey, can you pull out all the important dates from this text and give me a JSON summary?

[text]

2. Few-Shot Prompting (Multishot)

Claude:

<examples>
<example>
<input>The conference is scheduled for March 15, 2025 in San Francisco.</input>
<output>{"date": "2025-03-15", "event": "conference", "location": "San Francisco"}</output>
</example>
<example>
<input>Our next board meeting is on June 22nd.</input>
<output>{"date": "2025-06-22", "event": "board meeting"}</output>
</example>
</examples>

<input>
The product launches on September 1st in New York.
</input>

<output>

Gemini: Similar structure with system instruction for classification task.

Grok: Provide natural language examples matching conversational tone.

3. Chain-of-Thought Prompting

Claude (Extended Thinking - API):

response = client.messages.create(
    model="claude-sonnet-4-5-20250929",
    max_tokens=4096,
    thinking={"type": "enabled", "budget_tokens": 8192},
    messages=[{
        "role": "user",
        "content": "<task>[complex reasoning task]</task>\n\nPlease show your reasoning step by step."
    }]
)
# Response includes <thinking> block followed by <answer>

Gemini (Thinking Config):

response = client.models.generate_content(
    model="gemini-2.5-flash",
    contents="Solve step-by-step: [problem]",
    config=types.GenerateContentConfig(
        thinking_config=types.ThinkingConfig(thinking_budget=8192)
    )
)

Grok (Zero-Shot CoT):

Let's work through this step by step:

[complex problem]

Walk me through your reasoning.

4. Prompt Chaining

Break complex tasks into sequential steps.

Claude (XML structure):

<!-- Step 1 -->
<task>Extract relevant quotes about [topic].</task>
<document>[text]</document>

<!-- Step 2 -->
<task>Summarize extracted quotes and synthesize key insights.</task>
<quotes>[from step 1]</quotes>

Gemini: Use long context (1M+ tokens) to maintain conversation history.

Grok: Natural conversation flow with follow-up questions.

5. ReAct Prompting

Thought-Action-Observation cycles for reasoning with tools.

Claude (XML):

<question>[research question]</question>

<thought_1>[what needs to be done first]</thought_1>
<action_1>[tool use]</action_1>
<observation_1>[result]</observation_1>

<thought_2>[next step based on observation]</thought_2>
<final_answer>[conclusion]</final_answer>

Gemini: Similar structure with system instruction for reasoning behavior.

Grok: Conversational "Let's approach this methodically" style.

6. Tree of Thoughts

Explore multiple reasoning paths before concluding.

Claude:

<thought_paths>
<path_1>
<assumption>[approach 1]</assumption>
<reasoning>[step-by-step]</reasoning>
<conclusion>[result]</conclusion>
</path_1>
<path_2>...</path_2>
</thought_paths>

<synthesis>[best path and why]</synthesis>

Grok (Creative exploration):

Let's imagine three different perspectives on [topic]:
1. The Optimist's View: ...
2. The Skeptic's View: ...
3. The Pragmatist's View: ...

Where do these perspectives intersect?

Model-Specific Best Practices

Claude (Anthropic)

Key Features:

XML-style tags for structure (official Anthropic recommendation)
Extended thinking with tool use support
Parallel tool execution
Memory files for persistent knowledge
Up to 1M tokens context (Sonnet 4.5 beta)

Prompt Structure:

[task_context]
Setting the stage and overall context

[tone_context]
How Claude should approach the task

[input_data]
The actual data to work with

[examples]
Few-shot examples

[task_description]
Specific task details

[immediate_task]
The immediate action to take

[output_formatting]
Expected output structure

Best Practices:

Use XML tags liberally for structure
Enable extended thinking for complex reasoning
Leverage parallel tool use
Use cache control for repeated prompts
Prefill responses to guide format

Gemini (Google)

Key Features:

System instructions (primary behavior control)
Native multimodal (text, images, audio, video, code)
Ultra-long context (1M+ tokens)
JSON schema validation for structured output
Function calling for agent building

System Instruction Template:

<role>
You are Gemini, a specialized assistant for [Domain].
You are precise, analytical, and persistent.
</role>

<instructions>
1. Plan: Analyze the task and create a step-by-step plan.
2. Execute: Carry out the plan.
3. Validate: Review your output against the user's task.
4. Format: Present the final answer in the requested structure.
</instructions>

<constraints>
- Verbosity: [Low/Medium/High]
- Tone: [Formal/Casual/Technical]
</constraints>

<output_format>
Structure your response as follows:
1. Executive Summary: [Short overview]
2. Detailed Response: [The main content]
</output_format>

Best Practices:

Always use system instructions for behavior
Leverage multimodal inputs (combine text + images)
Use JSON schema for structured outputs
Process large documents with 1M+ context
Enable thinking_budget for complex tasks

Grok (xAI)

Key Features:

Witty, rebellious personality
Real-time X (Twitter) knowledge access
Relaxed content filtering
Natural, conversational style

Prompt Style:

Conversational, direct, and flexible. Grok responds best to natural language rather than over-structured prompts.

Best Practices:

Use conversational language (not overly formal)
Leverage real-time knowledge for current events
Embrace the wit (work with it, not against it)
Be direct—no need for delicate prompting
Explore controversial or filtered topics

Model Capabilities Comparison

Feature	Claude Sonnet 4.5	Gemini 3 Flash/Pro	Grok
Context Window	200K / 1M (beta)	1M+ tokens	1M+ tokens
Extended Thinking	✅ (with tools)	✅ (thinking config)	⚠️ Limited
Multimodal	✅ Yes	✅ Native	✅ Yes
Real-Time Knowledge	❌ No	❌ No	✅ Yes (X/Twitter)
Structured Output	✅ (prefill)	✅ (JSON schema)	⚠️ Manual
Parallel Tools	✅ Yes	✅ Yes	✅ Yes
Content Filtering	Strict	Moderate	Relaxed
Best For	Long-context, instruction following	Multimodal, system instructions	Current events, creative work

Anti-Patterns to Avoid

Anti-Pattern	Why It Fails	Fix
Claude: Ambiguous instructions	Follows literally	Be explicit with XML tags
Claude: Missing output format	Unpredictable formatting	Always specify format
Gemini: Ignoring system instructions	Wastes key feature	Always set system_instruction
Gemini: Not using multimodal	Underutilizes strength	Combine text, images, audio
Grok: Overly formal prompts	Against personality	Use conversational language
Grok: Ignoring real-time knowledge	Wastes unique capability	Ask about current events
All: Single-shot for complex tasks	Misses reasoning depth	Use chaining or multi-turn

Quick Reference Templates

Document Analysis (Claude)

<task>
[specific analysis task]
</task>

<document>
[paste document]
</document>

<output_format>
[expected structure]
</output_format>

Multimodal Input (Gemini)

{
  "contents": [{
    "parts": [
      {"text": "Describe what's in this image."},
      {"inline_data": {"mime_type": "image/jpeg", "data": "[base64]"}}
    ]
  }]
}

Current Events Query (Grok)

What's the latest on [topic] according to posts on X? Give me a summary of the current sentiment.

Model Selection Guide

Use Case	Recommended Model	Why
Long document analysis	Claude Sonnet 4.5	1M context, extended thinking
Multimodal understanding	Gemini 3 Pro	Native multimodal architecture
Current events/news	Grok	Real-time X (Twitter) access
Code generation	Claude Sonnet 4.5 or Opus 4.5	SOTA coding capabilities
Structured JSON output	Gemini 3 Flash	JSON schema validation
Creative writing	Grok	Witty, less constrained
Agentic workflows	Claude Sonnet 4.5	Parallel tool use
System instruction needs	Gemini 3 Flash/Pro	Primary feature
Instruction following	Claude Sonnet 4.5	Best in class

About this skill

When to Invoke This Skill

Quick Model Selection

Universal Prompting Techniques

1. Zero-Shot Prompting

2. Few-Shot Prompting (Multishot)

3. Chain-of-Thought Prompting

4. Prompt Chaining

5. ReAct Prompting

6. Tree of Thoughts

Model-Specific Best Practices

Claude (Anthropic)

Gemini (Google)

Grok (xAI)

Model Capabilities Comparison

Anti-Patterns to Avoid

Quick Reference Templates

Document Analysis (Claude)

Multimodal Input (Gemini)

Current Events Query (Grok)

Model Selection Guide

See Also

Reference Documentation

Related Skills