Turn an Idea into a Working Feature
Give your coding agent a real user problem, a sensible scope, and a finish line you can check.
- Small working scope
- Existing project patterns
- Visible proof it works
When to use it
You know what someone needs to do, but need help turning that into working software.
Bring this context
Your repository, the user workflow, a relevant existing screen, and observable acceptance criteria.
build-a-feature.md
copy · customize · try itHelp me build something people can actually use.
The person using it: [who they are]
The job they need to finish: [one concrete task]
What happens today: [current workflow and friction]
What should happen instead: [desired workflow]
Project or repository: [path]
A good existing example to follow: [screen, module, or file]
Done means: [3 observable things a user can do]
Outside this task: [explicit exclusions]
Read the project instructions and the code around this workflow. Tell me where the change belongs. If a missing answer would change the implementation materially, ask that question; otherwise state a reasonable assumption and keep moving.
Implement the smallest complete version of the requested workflow. Reuse the project's components, data access, and permission patterns. Make loading, empty, error, and success states useful. Preserve unrelated work.
Check the acceptance criteria against the implementation. Run relevant project checks, and exercise the user flow if a browser is available. Fix failures caused by this change. Don't describe an unrun check as passing.
Finish with what the user can now do, what you verified, and anything still unresolved. Explain it like a builder handing working software to a teammate.Make this prompt yours
- 1. Download or copyCopy the prompt into your AI tool, or download a copy to keep.
- 2. Customize itReplace bracketed details with your context. Include a real example when you can.
- 3. Test it on real workTry a real task, inspect the output, and refine the prompt where the result misses the mark.
Research behind this prompt
An original DevLaunch adaptation, informed by these sources. Reviewed September 23, 2026; source publication dates vary. Test the result against your own work.
- OpenAI · Codex best practices ↗
Task context, clear constraints, and a concrete definition of done.
- Anthropic · Claude Code best practices ↗
Inspect the project first and give the agent a way to verify its work.