prompt-optimizer

Prompt Optimizer

Diagnoses why a prompt underperforms and rewrites it to be specific, grounded, and consistent.

When to Use

• Model ignores part of the instructions
• Outputs vary wildly between runs
• Responses are too verbose, too vague, or formatted wrong
• Model "hallucinates" when it should say "I don't know"
• Chain-of-thought isn't happening when it should

Diagnosis First

Before rewriting, identify the failure mode:

Symptom	Likely cause
Ignores formatting instructions	Instructions buried in a long prompt
Inconsistent output structure	No example provided
Too verbose	No length guidance
Hallucinates facts	No instruction to acknowledge uncertainty
Misunderstands the task	Task description is ambiguous
Ignores constraints	Constraints mentioned once, not reinforced

The Anatomy of a Strong Prompt

[Role / Persona]          ← who the model is in this context
[Task definition]         ← exactly what to do
[Context / Input]         ← the data it operates on
[Constraints]             ← what NOT to do, limits, format
[Output format]           ← exact structure of the response
[Example] (optional)      ← one concrete example of good output

Not every prompt needs all sections. A simple prompt doesn't need a persona. A structured extraction task needs an output format and example.

Common Rewrites

Vague → Specific

# Before (vague)
Summarize this text.

# After (specific)
Summarize the following text in exactly 3 bullet points.
Each bullet must be one sentence under 20 words.
Focus only on actionable findings — ignore background context.
If there are no actionable findings, write "No actionable findings."

Text:
{text}

Missing Output Format

# Before
Extract the key information from this job posting.

# After
Extract the following fields from the job posting below.
Return a JSON object with exactly these keys:
{
  "title": "job title",
  "company": "company name",
  "location": "city, country or Remote",
  "salary": "salary range or null if not mentioned",
  "required_skills": ["skill1", "skill2"],
  "years_experience": number or null
}
If a field is not present, use null. Do not add extra fields.

Job posting:
{text}