Prompt Adviser.
PROMPT ADVISER / A BETTER STARTING POINT

Stop Masking Your AI Guess.

Turn a vague request into a precise, ready-to-run prompt. Prompt Adviser challenges your brief, catches missing details and asks the questions that matter before you spend time on the wrong output.

Download Prompt AdviserSee it in action

Free and open source. Editions for Codex, Claude and Google Antigravity.

Explore on GitHub A winding signal path: a vague request becomes a clearer brief, then a prompt ready for your decision.01 / A VAGUE REQUEST02 / BETTER QUESTIONS03 / A CLEARER PROMPT
ResearchImagesVideoCodingComputer use
01 / HOW IT WORKS

Get the brief right
before the work starts.

Prompt Adviser is an installable skill: a set of instructions that guides your AI assistant through a better briefing process. You bring the idea. It helps define the goal, context, output, constraints and checks. You leave with a copyable prompt and decide when to run it.

  1. 01

    Draft

    Bring your starting idea.

  2. 02

    Completeness check

    Find gaps in the brief.

  3. 03

    Three useful questions

    Clarify what matters.

    ↳ Loop back for more questions
  4. 04

    First 80% checkpoint

    Check sources and optional roles.

  5. 05

    Finished prompt + model suggestion

    Recommends a model that fits the task. Saves tokens and speeds up your reply. You choose.

  6. You choose “Run this prompt”

    Direct instruction, host permissions.

↳ Loop back for more questions

A quick and easy way to build your prompt

Three useful numbered questions per round, then “Shall I ask more?” You may finalise at any score.

80% Prompt Review and Assistance

Check your prompt against open source prompt directories to enrich your prompt.

Recommended Models

The best and most economical model is recommended before you run your prompt, including Jev where probabilistic decision making works better.

02 / SEE THE DIFFERENCE

From a rough idea
to a ready-to-run prompt.

One reusable bottle. A short video.
A clearer brief before anything runs.

Scripted illustration · no live AI calls

Your assistant reviews.
You decide what runs.

REUSABLE BOTTLE / VIDEO BRIEF 01 OF 05
THE STARTING POINT

One idea.
A few missing details.

“Make a short video about a reusable bottle.”

Before making anything, the adviser checks which parts of the brief are covered: goal, audience, context, output, constraints and checks.

Coverage, not qualityILLUSTRATIVE PROGRESSION
01 / 05
03 / WHERE IT FITS

Put a clearer brief to work.

Research

“Research this market.”

Clarify: Decision, geography, date range, source standards, comparison.

Decision-focused research

Images

“Create a product image.”

Clarify: Reference, composition, aspect ratio, text, preservation.

Explicit visual requirements

Video

“Make a promotional video.”

Clarify: Duration, shots, motion, continuity, sound.

Coherent sequence and delivery format

Coding

“Build a website.”

Clarify: Audience, routes, functions, constraints, acceptance.

Defined build scope

Computer use

“Clean up my CRM.”

Clarify: Records, matching rules, allowed changes, approvals, verification.

Controlled actions and observable completion

Practical intended benefits, not measured performance claims for Prompt Adviser.

04 / WHY THE DETAILS MATTER

The details you leave out become decisions AI makes for you.

Relevant detail removes ambiguity. Longer isn’t automatically better.
Built around research-supported prompting techniques.

CLAM / 2023

Clarification before an answer.

Table 4 reports raw question-answering accuracy of 34.25% by default versus 54.75% with CLAM — a 20.5 percentage-point difference on constructed Ambiguous TriviaQA, using text-davinci-002 and simulated clarification. An older, narrow study; not a Prompt Adviser test.

Read the CLAM study
ProTeGi / EMNLP 2023

Feedback, then refinement.

Reports up to 31% improvement in classification performance measured by F1, through automated feedback and repeated testing. This concerns the study’s optimisation method, not a product improvement percentage for Prompt Adviser.

Read the ProTeGi study

Study results depend on the model, task and method. They support these techniques, not measured performance claims for Prompt Adviser.

Fuller research context
05 / MAKE IT PART OF YOUR WORKFLOW

Same skill.
Your assistant of choice.

Download your edition, then follow the platform-specific installation steps.

Free skill. MIT licence. Your AI subscription or optional API usage is separate. No automatic upgrades.

06 / WHO MADE THIS
AEAndy Evans · UK
  • Entrepreneur
  • Investor & adviser
  • CEO & co-owner, Eleven Content

Andy Evans,
aka Digital Andy.

Andy Evans, also known as Digital Andy, is a UK entrepreneur, investor and adviser, and CEO and co-owner of Eleven Content, a B2B content agency.

He created Prompt Adviser to help himself and other people get better results from AI by asking better questions first. It brings structure to the brief, challenges assumptions and turns a rough idea into a clear prompt you control.

Built for anyone who wants to put AI to better use.

07 / A FEW GOOD QUESTIONS

Before you
get started.

Is Prompt Adviser really free?

Yes. The skill is free and open source under the MIT licence. Your AI assistant subscription or optional API usage may still cost money.

Can I install the Codex ZIP in ChatGPT?

Not directly in ordinary ChatGPT. The Codex edition is for skills-enabled Codex desktop and CLI hosts. See the platform instructions for the correct install path.

Do I have to choose a role?

No. Roles are optional, and No role is always a valid choice. A role is framing, not a credential: it does not give a model new capabilities.

Do I need Jev?

Generally not for a single prompt. Jev is an optional tool for structured, repeated routing, classification or scoring. It is not called during ordinary prompt review.

What happens to private details?

Source searches use generic task terms, not your private draft. Your assistant host and provider policies still apply. This site does not collect visitor prompts or provide a live AI service.

Do I have to reach 80%?

No. You may finalise at any score. Task execution begins only when you directly say “Run this prompt”, subject to your host’s permissions.

Help make the questions better.

Found a gap? Share an issue or contribute a pull request. Please use fictional or anonymised examples only.

Open an issuePull requests