Public training path

Complete User Guide

Prompt Architect · Beginner-friendly walkthrough

Prompt Architect is the right place to start when the task matters enough that you do not want to improvise. Use it when the output needs a clear role, scope, constraints, format, and a version of the prompt that another person can run later without extra...

Next best action

Preview the guidance here, then create an account to save workspaces, unlock guided execution, and continue inside the platform.

Start with the first section Create account to continue

Sections

2 guided blocks

Read Time

7 min focused read

Coverage

206 searchable doc sections

prompt architecttutorialwalkthroughbeginneradvancedcontrolsquality gatefields

Section 1 of 2

How to Use Prompt Architect from First Draft to Ready-to-Run Prompt

prompt architecttutorialwalkthroughbeginneradvanced

Use it for: new prompt creation, standardizing team prompts, translating messy ideas into a reusable spec, and preparing prompts that will later be benchmarked in PromptForge.

Do not use it for: quick one-off chatting, long experimentation with many model variants, or document-grounded audits. Those belong in Lab, Vault, or ComplyIQ after the prompt spec is stable.

Start mode

Choose scratch when you are designing from zero, or template mode when a prepared use-case launch already matches the task.

Role and language

Set the working role and final output language before writing fields so the whole prompt compiles in the right voice and structure.

Detail level

Lean draft is faster for rough work. Enterprise rigor is better when another team, client, or reviewer will depend on the output.

Quality gate

The quality panel tells you what is missing, what is already strong, and whether the draft is ready for execution or handoff.

Step 1: Set the frame before typing

Pick the start mode, role, output language, and detail level. This prevents the session from drifting into the wrong audience, tone, or rigor level.

Step 2: Write the discovery input in plain language

Describe the business task, the audience, and the desired output. Write naturally first. The system is built to refine rough input, not punish it.

Step 3: Fill the generated fields methodically

Work through Persona, Goal, Objectives, Scope, Tone, Negative Constraints, Edge Cases, and Output Rules. Each field should answer one practical question a reviewer would ask.

Step 4: Use AI Rewrite deliberately

Rewrite fields that are vague, conversational, or incomplete. Keep the useful parts, then manually restore any nuance the rewrite trimmed away.

Step 5: Read the quality gate before executing

If the panel says something essential is missing, fix that before you run the prompt. A fast weak run usually costs more time than one careful edit.

Step 6: Decide the next destination

Execute when you need a quick proof, save when the prompt should be reused, and move into PromptForge when the prompt is strong enough to benchmark or compare across models.

Do This

Name the task in business language before you optimize wording.

Treat Negative Constraints and Output Rules as the last defense against bad outputs.

Keep one realistic example input nearby so you can sanity-check the prompt immediately.

Save the session once the structure is good, even if the wording still needs later polishing.

Avoid

Do not leave generated fields blank just because the first draft feels obvious to you.

Do not ask the system for a perfect prompt if you have not defined what success looks like.

Do not benchmark unstable prompts in the Lab before the quality gate looks healthy.

Do not hand off a prompt without sample inputs and clear pass-fail signals.

Pro Tip: The fastest good workflow

Use Prompt Architect in two passes. Pass one gets the structure right. Pass two tightens the language. Most weak prompts fail because users try to do both at once and skip the review step between them.

Section 2 of 2

Session States and Control Reference

prompt architectcontrolsquality gatefieldssession

Prompt Architect is built around a persistent session, not a one-shot text box. That session stores the entry mode, detail level, field values, target model information, template lineage, and the quality report, so the prompt can be reopened and improved without rebuilding the whole brief from memory.

Entry mode

Use scratch mode when the use case is still undefined, and template mode when you want to inherit a known structure and keep the source template relationship visible.

Phase flow

The workspace moves through Discovery, Fields, Refine, and Output so you always know whether you are framing the task, filling controls, polishing wording, or reviewing compiled output.

Role, language, and model target

These settings shape how the session is authored and compiled. Pick them early so every field is written for the same audience and execution environment.

Quality report

Every meaningful edit updates the quality report, which exposes score, status, strengths, missing fields, warnings, and specific recommendations.

Scratch: create a session from zero when the task, audience, or operating pattern is still unclear.

Template: preload a known field structure and keep the session tied to a source template so later reviewers understand what was inherited versus customized.

Lean detail: use lighter scaffolding for fast internal drafting and early exploration.

Enterprise detail: use more explicit control language when another team, approver, or client will depend on the result.

Update fields: persist the current field values and refresh the session state without throwing away prior work.

Refine field: improve one field in context instead of asking the system to re-author the entire prompt blindly.

Create the session correctly

Decide scratch versus template, then set the role, language, and detail level before filling fields.

Update fields in batches

Save groups of related edits together so the quality report reflects a meaningful state, not half-finished fragments.

Use field refinement only on weak spots

If one field is vague, refine that field. Do not rewrite stable controls just because one paragraph is weak.

Compile only after reading the report

If the session is blocked, stop and fix the missing context instead of wasting time and credits on a weak compile.

Pro Tip: Use the quality report as a reviewer

The fastest teams do not ask whether the prompt feels good. They ask whether the quality report still shows unmanaged risk and whether a new operator could understand the prompt without the author present.

Prompt QA & Team Handoff

Academy v4.0 · Interactive Documentation · Beginner Mode