$ cat latest.md ▊

Prompt patterns I actually reuse

Rubric-first grading, negative few-shot examples, and chain-of-verification — the only three patterns I trust in production.

Read the post → 1 min read · 3 views
prompt-patterns.md

 

Most prompt advice is folklore. These three survive contact with production.

 

 

```python

RUBRIC = """

Score 1-5 on: correctness, clarity, completeness.

Respond ONLY with JSON: {"scores": [...], "reason": "..."}

"""

 

def grade(answer: str) -> dict:

# judge != generator: use a separate call for scoring

return llm.chat([

{"role": "system", "content": RUBRIC},

— end of preview —

All posts

(3)