A good prompt for an AI coding agent defines the desired outcome, supplies relevant evidence, states constraints, and describes how success will be tested. It gives the agent enough freedom to solve the task without leaving important decisions ambiguous.
Use this structure:
- Context: what the system does and where the issue appears.
- Goal: the exact user-visible or technical result.
- Scope: files or services that may change.
- Constraints: compatibility, dependencies, style, security, and performance.
- Evidence: issue text, logs, screenshots, examples, or failing tests.
- Acceptance: behaviors and commands that must pass.
- Non-goals: related work that should remain untouched.
For large tasks, ask for a plan first. Review dependencies, migrations, and assumptions before allowing implementation. Store durable coding conventions in project rules rather than bloating every prompt.
Verdent's Plan Mode can turn an incomplete request into structured requirements through clarification and refinement. That helps when the user knows the outcome but not the implementation. A good prompt does not need to be long; it needs to remove the expensive ambiguities. If the result is wrong, improve the requirement or evidence instead of merely asking the agent to “try again.”
