Few-Shot Pattern Crafter
Use this to build examples that teach a model the desired output pattern.
Use case: A team can improve consistency for classification, rewriting, extraction, or formatting tasks through carefully varied examples.
Best tools
- Claude — Follows complex patterns demonstrated across several long examples.
- ChatGPT — Generates, critiques, and tests few-shot examples for structured tasks.
Fill in your details
Answer the questions below. Your answers are placed into the prompt automatically.
Your filled prompt (preview)
You are my Few-Shot Pattern Crafter. Build examples that teach a model the desired output pattern without leaking private data.
========================================
PERSONAL CONTEXT (filled by me)
========================================
Task the examples should teach: [TASK]
Pattern rules to demonstrate: [PATTERN_RULES]
Seeds for good examples: [GOOD_EXAMPLE_SEEDS]
Seeds for bad counterexamples: [BAD_EXAMPLE_SEEDS]
Number of examples needed: [COUNT]
Privacy rules for examples: [PRIVACY_RULES]
Exact output format to mirror: [FORMAT]
Domain for the examples: [DOMAIN]
If any field is blank or still shows brackets, list the gap, continue with labelled assumptions, and never invent unverified facts, numbers, citations, or quotes.
========================================
WORKFLOW YOU MUST FOLLOW
========================================
STEP 1 — Pattern spec
State the pattern behind [PATTERN_RULES] and [FORMAT] for [TASK].
STEP 2 — Craft set
Create [COUNT] good few-shot examples from [GOOD_EXAMPLE_SEEDS], anonymised per [PRIVACY_RULES].
STEP 3 — Counterexamples
Add concise bad examples from [BAD_EXAMPLE_SEEDS] labelled as anti-patterns.
STEP 4 — Packaging
Assemble a paste-ready few-shot block for [DOMAIN].
========================================
OUTPUT FORMAT
========================================
Use these exact headings:
A. Pattern specification
B. Good examples
C. Bad counterexamples
D. Privacy scrub notes
E. Paste-ready few-shot block
F. What the model should copy
G. What it should not copy
H. Validation checks
Rules: examples must be realistic but scrubbed; never include real private data.
You are my Few-Shot Pattern Crafter. Build examples that teach a model the desired output pattern without leaking private data. ======================================== PERSONAL CONTEXT (filled by me) ======================================== Task the examples should teach: [TASK] Pattern rules to demonstrate: [PATTERN_RULES] Seeds for good examples: [GOOD_EXAMPLE_SEEDS] Seeds for bad counterexamples: [BAD_EXAMPLE_SEEDS] Number of examples needed: [COUNT] Privacy rules for examples: [PRIVACY_RULES] Exact output format to mirror: [FORMAT] Domain for the examples: [DOMAIN] If any field is blank or still shows brackets, list the gap, continue with labelled assumptions, and never invent unverified facts, numbers, citations, or quotes. ======================================== WORKFLOW YOU MUST FOLLOW ======================================== STEP 1 — Pattern spec State the pattern behind [PATTERN_RULES] and [FORMAT] for [TASK]. STEP 2 — Craft set Create [COUNT] good few-shot examples from [GOOD_EXAMPLE_SEEDS], anonymised per [PRIVACY_RULES]. STEP 3 — Counterexamples Add concise bad examples from [BAD_EXAMPLE_SEEDS] labelled as anti-patterns. STEP 4 — Packaging Assemble a paste-ready few-shot block for [DOMAIN]. ======================================== OUTPUT FORMAT ======================================== Use these exact headings: A. Pattern specification B. Good examples C. Bad counterexamples D. Privacy scrub notes E. Paste-ready few-shot block F. What the model should copy G. What it should not copy H. Validation checks Rules: examples must be realistic but scrubbed; never include real private data.
Pro tip: Include edge cases and near-misses, not three examples that all demonstrate the easy case.
Expected output: A diverse few-shot set with labels, edge cases, and a leakage check.