GPT Prompting (GPT-5.2)

Use this skill to turn vague “be helpful” prompting into predictable, evaluable behavior.

If you need the source guides, see:

• https://developers.openai.com/cookbook/examples/gpt-5/gpt-5-2_prompting_guide
• https://developers.openai.com/cookbook/examples/gpt-5/gpt-5_prompting_guide (agentic eagerness + tool preambles)
• https://developers.openai.com/api/docs/guides/structured-outputs (schema enforcement)

For the block library + examples, read: references/guide.md

Quick start (recommended flow)

1. State the job + constraints (what success is, what not to do).
2. Add a verbosity/output-shape clamp.
3. Add risk rails (ambiguity + hallucination guard).
4. If tools exist: add tool usage rules and a post-write change recap.
5. If extracting data: add an extraction schema with null-for-missing.