Prompt Structure

# Prompt Structure Reference

Use this reference to build the common structure for every generated prompt.

## Principles

- Start outcome-first: define the result, success criteria, constraints, and evidence required.
- Use clear sections so the receiving agent can locate instructions quickly.
- Put dynamic task context near the top, but tell the receiving agent to discover repo truth before making claims.
- Embed context the prompt author already resolved (issue/PR/URL/file digests, conversation facts, configured memory facts) instead of delegating rediscovery, but instruct the receiving agent to re-verify digests against the live source.
- Use examples or templates only when they reduce ambiguity.
- Keep reasoning instructions practical: ask the agent to plan and verify, but do not require hidden reasoning in the final answer.
- Prefer concise prompts that preserve judgment. Do not overload the prompt with every possible edge case.
- Include uncertainty handling: ask clarifying questions only when required, otherwise make a conservative assumption and report it.
- Make validation explicit. Good prompts state what must be checked before the agent calls the task done.
- Wrap the final answer in exactly one fenced `markdown` code block so the user can copy the generated prompt without extra cleanup.

## Required Sections

Use these sections inside the fenced output code block, in order unless the user's task clearly needs a different order.