Turn a vague request into a prompt your coding agent can actually execute
One bounded prompt with goal, inputs, constraints, done-criteria, and out-of-scope; the questions you still have to answer yourself; and what changed from your draft and why.
Paste what you were about to type into your coding agent. The Agent rewrites it into a prompt with a clear goal, explicit inputs, constraints, and a definition of done, and tells you what it could not decide for you.
Good for
Someone who keeps getting plausible-but-wrong results from a coding agent and suspects the ask, not the agent.
Not for
Prompts for image or video models, system prompts for a product, or requests with no concrete task behind them.
What you give
- What you were about to ask
- Target coding agent
- What the agent will have access to (repo, files, tools) (optional)
What you get back
- 1. The prompt
- 2. Questions only you can answer
- 3. What changed and why
Example
See an example
## The prompt Goal: make `pnpm test:integration` pass again after the taxonomy migration without changing any test expectation. Inputs: the failing output pasted below; the migration in `drizzle/0018_*.sql`; the schema in `lib/db/schema/`. Constraints: do not edit files under `tests/`; do not add a new migration; keep `db:check` clean. Done when: the full integration suite passes locally and `git diff --stat` touches only `lib/` and `drizzle/meta`. Out of scope: performance, unrelated lint warnings. ## Questions only you can answer Is the failing test expecting the old label strings or the new keys? If the former, the test is wrong and this prompt's constraint blocks the fix. ## What changed and why - Replaced "fix the tests" with a checkable done-criterion, so the agent cannot declare success early. - Named the files the agent may touch, because the draft let it edit tests. - Added out-of-scope, because the draft invited drive-by refactors.
Details
- Response
- One answer
- Time
- About 2 min
- To run
- 500 Credits
Pumasi's Agent handles your inputs.