Free prompt builder

Generate the brief the agent should have received.

Turn a vague request into a scoped task with source context, permission boundaries, acceptance checks, and a way back when work fails. Start from a complete example or replace each field with your own.

Choose an exampleOpen the builder

Free, public, and available without sign-in.

Worked examples

Start with a complete brief

Each example fills every field. Load the nearest shape, then replace its details.

Your brief

Define the work before the agent starts

12/12required fields filled
01Purpose

Name the person, the repeated job, and a result someone can point at.

The shape changes what evidence, scope, and permissions the work needs.

Inspect source material, compare evidence, and return a decision-ready recommendation.

Use one concrete verb and one bounded object.

A real role makes workflow and review choices easier to judge.

Describe what exists or works when the task is done.

02Grounding

Point to the real source and separate facts from questions that still need proof.

Name files, links, systems, schemas, current behavior, and anything already decided.

Expose guesses early so the agent checks them before building on them.

03Scope

Required work says what enters version one. Non-goals keep it from growing mid-task.

List capabilities, steps, states, or decisions. Give each line one job.

List at least three plausible additions that stay outside this task.

Include the stack, formats, compatibility rules, policy, timing, or dependency limits.

04Access and boundaries

A prompt can request restraint. Named permissions and environment limits control consequences.

Pick the narrowest mode that can produce the result.

The agent may change only the paths listed in the brief. Commands and outside access still need approval.

Name what can be read, changed, run, or contacted, plus the actions that always need approval.

05Acceptance

Observable checks give “done” a fixed meaning before the work can negotiate with it.

Write checks a person can see from outside the implementation.

Include an empty, missing, duplicate, invalid, denied, or partial case.

Match the evidence to the type of result instead of defaulting to one test style.

Use known inputs, one difficult case, and expected outputs before any live or repeated run.

06Handoff

Name the evidence and files a reviewer receives, plus the point where the agent stops.

List files, reports, commands, screenshots, limitations, and the expected final summary.

Decide the retreat condition before a failed attempt turns into a chain of fixes.

After two failed verification passes, stop and report the evidence, hypothesis, and smallest next step.

12 required fields shape the brief.
Why it is shaped this way

The brief becomes the review checklist

Purpose and grounding tell the agent what problem exists and which sources deserve trust. Scope and boundaries limit the work and its consequences. Acceptance and handoff turn “done” into evidence another person can inspect.

A useful brief stays small enough to review. If the required work cannot fit into clear stages with a check at the end of each one, split the task before work begins.