AI Prompting Cheatsheet

You are a prompt engineering expert. I need a prompt for
the following use case:

Use case: [describe what you need the AI to do]
Target model: [Claude / GPT-4 / Llama]
Input format: [what the user will provide]
Output format: [what the AI should return]
Quality bar: [production / prototype / exploration]

Generate an optimized prompt that includes:
1. Clear role definition
2. Specific instructions
3. Output format specification
4. 2-3 few-shot examples
5. Edge case handling
6. Guardrails

Also provide:
- A test case I can use to validate the prompt
- Common failure modes to watch for
- Suggested temperature setting

# Chain 1: Research
Prompt 1: "List the top 5 challenges of [topic].
Return as a numbered list."

# Chain 2: Deep Dive (uses output of Chain 1)
Prompt 2: "For each challenge below, provide:
- Root cause
- Impact (1-10)
- Potential solution
Challenges: {output_of_prompt_1}"

# Chain 3: Synthesis (uses output of Chain 2)
Prompt 3: "Based on this analysis, write a 500-word
executive recommendation. Prioritize by impact score.
Analysis: {output_of_prompt_2}"

# Chain 4: Polish
Prompt 4: "Review and improve this recommendation.
Check for: logical consistency, actionability,
missing stakeholder concerns.
Draft: {output_of_prompt_3}"

# For documents exceeding context window:

Step 1: Split document into chunks of ~2000 tokens
Step 2: Summarize each chunk:
  "Summarize this section in 3-5 bullet points,
   preserving key facts and numbers:
   [chunk]"
Step 3: Combine chunk summaries:
  "Synthesize these section summaries into a
   coherent overall summary. Resolve contradictions.
   Maintain chronological order:
   [all chunk summaries]"
Step 4: Final polish:
  "Refine this summary. Ensure it reads as a
   standalone document, not a collection of fragments.
   Add transitions between topics: [combined summary]"

[First, generate a response to the task]

Now critically review your own response:

1. ACCURACY: Are all facts and claims correct?
   Flag anything you're less than 90% confident about.

2. COMPLETENESS: Did you miss any important aspects?
   List what a domain expert might add.

3. CLARITY: Is the response easy to understand?
   Identify any jargon that needs explanation.

4. BIAS: Are there any assumptions or biases?
   Consider alternative perspectives.

5. ACTIONABILITY: Can the reader act on this immediately?
   What additional context would they need?

Based on this self-review, provide an improved version
of your response.

# Strategies to maximize context window efficiency:

1. COMPRESS: "Summarize the following into the minimum
   tokens needed to preserve all actionable information:"

2. PRIORITIZE: "I'm sending you context in priority order.
   The most important information is first. If you need
   to discard context, start from the end."

3. STRUCTURED REFERENCES: Instead of pasting full docs:
   "Reference these definitions:
   - USER = authenticated user with role X
   - ORDER = {id, items[], status, created_at}
   - VALID_STATES = [pending, confirmed, shipped, delivered]
   Use these references in your response."

4. DELTA UPDATES: "Previous context: [summary of prior turns].
   New information: [only what changed].
   Updated task: [current request]"