The Prompts screen: project-aware prompts for any chat assistant

The Prompts screen sits under Prompts in each project's sidebar. It is a library of prompts already filled with your project's context, to paste into a chat assistant you already use: ChatGPT, Claude or Gemini. It is free on every plan, spends no credit and makes no AI call of its own. The work happens in your assistant; VarynForge writes the prompt.

What the screen shows#

A grid of cards, one per template. Each card has a name, a category badge (Discovery, Planning or Audit) and a one-line description.

Template	Category	Needs	You type in
Expand my niche	Discovery	niche	nothing
Brainstorm seed keywords	Discovery	niche	nothing
Decode search intent	Discovery	niche	a query
Audit my content gaps	Audit	niche, website, pages	nothing
Tear down a competitor	Audit	niche, website	a competitor page URL
Outline a single article	Planning	niche	a topic, a keyword
Generate article titles	Planning	niche	a topic
Build a client-ready pitch report	Planning	niche	the client's name

The target keyword for an outline and the client name for a pitch report are optional. A note at the top of the screen says how old the project's research data is.

Where the data comes from#

Each prompt is filled from four parts of your project: the confirmed niche, the connected website and its profile, the pages mapped from your site, and your competitors, minus any you marked ignored. The context is read fresh every time a prompt is generated, so a change to the project shows up in the next prompt. Nothing is cached from an earlier render.

What you can do here#

• View prompt opens a template that needs no input. The filled prompt appears right away.
• Customize & view opens a template that takes input. Fill in the fields, then press Generate prompt.
• Copy prompt copies the text to your clipboard. If the copy fails, the prompt sits in a text box you can select by hand.
• Edit inputs returns to the form to change what you typed.
• Close shuts the dialog.

A card whose context is missing is locked and says what it needs, for example "Needs a connected website." or "Needs pages discovered for your website." Set that part of the project up and the card unlocks.

The prompts are a starting point, not research data. A keyword list from "Brainstorm seed keywords" is a guess from your assistant until you check it against search data, which is what a research run does.

The same thing over MCP#

The templates on this screen are not MCP tools. An agent connected over MCP does not need them: it can read the niche, the pages and the competitors directly and reason over them itself.

One template is shared. "Build a client-ready pitch report" is the same template that get_pitch_report_payload returns as renderGuide. Over MCP it is filled from a completed run's competitors, next to the run's data. See client reports.

get_writer_system_prompt is a different thing. It returns the writer's methodology for drafting one content type from a brief, with a channel of article, x, linkedin, reels or youtube. It is the same text for every project and is not filled from project context. The brief carries the project specifics. It does not appear on this screen. See drafting and verification.