AI

Author:

Last updated: | View on GitHub

categories: ai, prompting, automation

What a good prompt actually looks like (in practice)

  1. Start with the goal
    Clearly state what you want to achieve in one or two sentences.
    If there are strict requirements or success criteria, mention them right away.

  2. Assign a role only if essential
    Only specify a role (e.g. “Act as a senior SRE”) if it truly changes how the assistant should respond.
    Always clarify who the response is for: beginner, expert, manager, etc.

  3. Give relevant context
    Share the necessary background or sources for the task (text, links, files, images).
    Highlight assumptions to treat as true and note anything that’s unknown.

  4. Clarify boundaries
    Define what the assistant should and shouldn’t do.
    For example: “No web browsing,” “Don’t suggest new tools,” “Use a formal tone,” etc.

  5. State important rules upfront
    Include guidelines, style, legal, or safety constraints early.
    If some rules are more important than others, mention how to handle conflicts between them.

  6. Describe the approach you want
    Instead of just “think step-by-step,” ask for:
    • Key assumptions
    • Remaining uncertainties
    • Basic checks or validations
    • Brief rationale for trade-offs
      Specify when to ask questions versus when to make assumptions.
  7. Define the output format
    State clearly:
    • Required format (section headings, bulleted lists, code, table, JSON, etc.)
    • Length, tone, and language preferences
    • If you need multiple versions, say how they should differ.

  8. Provide examples when helpful
    Include one or two positive examples if you can.
    Show a quick “don’t do this” example for clarity if certain mistakes should be avoided.

  9. Clarify next steps
    Say what should happen after the draft:
    • Should the assistant ask questions?
    • Suggest improvements?
    • Wait for user edits? Only include prior conversation/history if it’s important for interpretation.
  10. Optionally, add a head start
    If you have a rough draft or template, share it.
    This helps the assistant build on your structure instead of starting from scratch.

Ralph wiggum

Ralph is a Bash loop:

while :; do cat PROMPT.md | claude-code ; done

References