Common Pitfalls

Take a vague prompt and make it specific. Notice how adding details transforms the quality of results.

I have a vague prompt that needs improvement.

Original vague prompt: "${vaguePrompt}"

Make this prompt specific by adding:
1. **Audience**: Who will read/use this?
2. **Format**: What structure should it have?
3. **Length**: How long should it be?
4. **Tone**: What voice or style?
5. **Context**: What's the situation or purpose?
6. **Constraints**: Any must-haves or must-avoids?

Rewrite the prompt with all these details included.

Use this to verify your prompt contains all necessary context before sending.

Review this prompt for missing context:

"${promptToCheck}"

Check for:
1. **Referenced but not included**: Does it mention "the code," "the document," "earlier," or "above" without including the actual content?

2. **Assumed knowledge**: Does it assume knowledge about a specific project, company, or situation?

3. **Implicit requirements**: Are there unstated expectations about format, length, or style?

4. **Missing background**: Would a smart stranger understand what's being asked?

List what's missing and suggest how to add it.

Check your prompts for hidden biases and leading language.

Analyze this prompt for bias and leading language:

"${promptToAnalyze}"

Check for:
1. **Embedded assumptions**: Does the question assume something is true?
2. **Leading phrasing**: Does "Why is X good?" assume X is good?
3. **Missing alternatives**: Does it ignore other possibilities?
4. **Confirmation seeking**: Is it asking for validation rather than analysis?

Rewrite the prompt to be neutral and open-ended.

Use this to get the AI to flag its own uncertainties and potential errors.

I need you to provide information about: ${topic}

IMPORTANT: After your response, add a section called "Verification Notes" that includes:

1. **Confidence Level**: How certain are you about this information? (High/Medium/Low)

2. **Potential Errors**: What parts of this response are most likely to be wrong or outdated?

3. **What to Verify**: What specific claims should the user fact-check independently?

4. **Sources to Check**: Where could the user verify this information?

Be honest about limitations. It's better to flag uncertainty than to sound confident about something wrong.

When your first result isn't right, use this to systematically improve it.

My original prompt was:
"${originalPrompt}"

The output I got was:
"${outputReceived}"

What's wrong with it:
"${whatIsWrong}"

Help me iterate:

1. **Diagnosis**: Why did the original prompt produce this result?

2. **Missing Elements**: What was I not explicit about that I should have been?

3. **Revised Prompt**: Rewrite my prompt to address these issues.

4. **What to Watch For**: What should I check in the new output?

Generate clear format specifications for any output type you need.

I need AI output in a specific format.

**What I'm asking for**: ${taskDescription}
**How I'll use the output**: ${intendedUse}
**Preferred format**: ${formatType} (JSON, Markdown, CSV, bullet points, etc.)

Generate a format specification I can add to my prompt, including:

1. **Exact structure** with field names and types
2. **Example output** showing the format
3. **Constraints** (e.g., "Return ONLY the JSON, no explanation")
4. **Edge cases** (what to output if data is missing)

Get a strategy for processing documents that exceed context limits.

I have a large document to analyze:

**Document type**: ${documentType}
**Approximate length**: ${documentLength}
**What I need to extract/analyze**: ${analysisGoal}
**Model I'm using**: ${modelName}

Create a chunking strategy:

1. **How to divide**: Logical break points for this document type
2. **What to include in each chunk**: Context needed for standalone analysis
3. **How to synthesize**: Combining results from multiple chunks
4. **What to watch for**: Information that might span chunks

Check your prompt for security issues before sending.

Review this prompt for security concerns:

"${promptToReview}"

Check for:

1. **Exposed Secrets**: API keys, passwords, tokens, credentials
2. **Personal Data**: Names, emails, addresses, phone numbers, SSNs
3. **Proprietary Info**: Trade secrets, internal strategies, confidential data
4. **Injection Risks**: User input that could manipulate the prompt

For each issue found:
- Explain the risk
- Suggest how to redact or protect the information
- Recommend safer alternatives

Structure your prompt to minimize hallucination risk and flag uncertainties.

I need information about: ${topic}

Please follow these guidelines to minimize errors:

1. **Stick to well-established facts**. Avoid obscure claims that are hard to verify.

2. **Flag uncertainty**. If you're not confident about something, say "I believe..." or "This may need verification..."

3. **No invented sources**. Don't cite specific papers, books, or URLs unless you're certain they exist. Instead, describe where to find this type of information.

4. **Acknowledge knowledge limits**. If my question is about events after your training data, say so.

5. **Separate fact from inference**. Clearly distinguish between "X is true" and "Based on Y, X is likely true."

Now, with these guidelines in mind: ${actualQuestion}