Kenhuang Academy
  • English

      • Resources

      Prompt engineering tracks models and APIs. Prefer vendor docs. This course keeps the stable contract habits (role / task / constraints / format, delimiters, examples, schema, eval).

      OpenAI now treats production prompts as application code: git, review with the feature, eval as a gate. Hosted reusable prompt objects are in a deprecation window (see their deprecations page). Own your gold questions and a runner; do not assume a hosted eval dashboard will last forever.

      Official

      ResourceURL
      OpenAI Promptinghttps://developers.openai.com/api/docs/guides/prompting
      OpenAI Prompt engineeringhttps://developers.openai.com/api/docs/guides/prompt-engineering
      OpenAI Structured Outputshttps://developers.openai.com/api/docs/guides/structured-outputs
      OpenAI migrate from prompt objectshttps://developers.openai.com/api/docs/guides/prompting/migrate-from-prompt-object
      Anthropic Prompting best practiceshttps://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices
      Anthropic docs homehttps://platform.claude.com/docs

      Follow-on courses on this site (suggested order)

      GoalCourse
      Turn the contract into editor rules / SkillsCursor
      Terminal agent + CLAUDE.mdClaude Code
      system_prompt + tool loop in codeLangChain
      Agent / workflow prompts on a canvasDify
      Tool schemas and safe defaultsMCP · MCP security
      Stop stuffing a corpus into the promptRAG · LangChain RAG

      Read this primer once, then pick one product course. Do not collect five copies of the same four blocks in five UIs.

      What to keep in the repo

      ArtifactTypical pathRole
      Durable policyAGENTS.md, .cursor/rules, CLAUDE.mdSystem contract
      Task templatesprompts/*.md or prompts/*.tsBuild user messages with typed holes
      Gold questionsevals/gold.jsonlRegression
      SchemasPydantic / JSON Schema next to the promptOutput contract

      If a sentence cannot show up as a gold-question expectation, it probably does not belong in the prompt.

      Vendor vocabulary (so doc searches line up):

      This courseOpenAI often saysAnthropic often says
      System contractinstructions / system messagesystem prompt
      Four blocksRole in system; task and examples in userBe clear and direct; XML sections
      Structured outputStructured Outputs / schemaTools or JSON mode (check current docs)
      EvaluationPrompts as code + your own evalsIterate on real tasks, not empty boilerplate

      Tutorial index

      1. Prompt Engineering tutorial
      2. Introduction
      3. Four building blocks
      4. Structure and delimiters
      5. Few-shot examples
      6. Structured output
      7. Tools and agents
      8. Evaluation
      9. Pitfalls
      10. Practical examples

      Contribute

      Part of Kenhuang Academy. When vendor docs change, open an issue or PR and update docs/zh/prompt-engineering/ and docs/en/prompt-engineering/ together. Do not add jailbreaks, injection exploits, or social-engineering material.

      If an outside article conflicts with this course (for example by teaching attack steps), follow this course’s boundary and the vendor’s safety docs.

      Next: Introduction · Four building blocks · Practical examples

      评论