Ecosystem Features

Customization & Automation

Learn what Scripts, Rules, Projects, Skills, and Plugins are—explained with simple analogies and exact usage steps.

1. Scripts

🧸 The Analogy (For an 8-year-old) Imagine you want to paint a castle, but you don't know how to paint. A Script is like a pre-drawn coloring book page that we give you. You just load the page into your AI, and it knows exactly where the lines are and how to paint the castle perfectly!

In our classroom, a Script is a pre-written document of instructions created by our agency. You don't need to write code or download complex programming files. You simply load these pre-made scripts directly into your AI chat account to instantly teach it a specific writing job.

How to Load Across Platforms

  • Claude: Paste the script into the System Prompt console or the Custom Instructions box inside a Claude Project.
  • ChatGPT: Paste it into your account's Custom Instructions or the System Instructions field when creating a Custom GPT.
  • Gemini: Paste the script into the System Instructions panel inside Google AI Studio.
  • Groq & OpenRouter: Pass the script text programmatically under the "role": "system" parameter inside your API developer playground or local tools.

What Scripts Do Best

  • Setting Up Roles: Instantly turning standard Claude or ChatGPT into specialized developmental editors, pacing auditors, or subgenre experts.
  • Formatting Output: Telling the AI exactly how to layout character cards, structural matrices, or critique worksheets.
  • Saving Effort: Eliminating the need to type out long instruction lists manually every time you start a new conversation.

2. Rules

🧸 The Analogy (For an 8-year-old) Imagine playing a game of tag, but before starting, you say: "House rules! Nobody can touch the safe zone, and everyone must hop on one leg!" The AI will remember these rules for every single game you play from now on.

Rules (often called "Custom Instructions" or "System Prompts") are permanent style guidelines. They tell the AI how to act, what tone to write in, and what terms to avoid—so you don't have to repeat your instructions in every single message.

Setup on Consumer Platforms

  • ChatGPT: Click your profile picture (bottom left) → click "Customize ChatGPT" → paste rules under "How would you like ChatGPT to respond?" → click Save.
  • Claude: Click your profile picture → click "System Prompts" → paste rules → toggle "Apply to new chats" → click Save.
  • Gemini (Advanced): Open Gemini Advanced → click "Gems" → click "Create Gem" → paste your writing rules and save.

Setup on API Platforms

  • Groq: Open the Groq Cloud Console → go to the Playground → paste your rules into the "System Message" box on the left.
  • OpenRouter: Open the OpenRouter Playground → select your model → paste your style rules into the System Message prompt panel.

3. Projects

🧸 The Analogy (For an 8-year-old) A Project is like a school binder. Inside the binder, you clip your character drawings, your manuscript pages, and your rules sheet. The AI helper sits inside the binder and only reads what you clipped in. It won't look at anything outside that binder!

Projects (sandboxes/workspaces) allow you to bundle reference files (manuscripts, lore books, character summaries) and rules that only apply to chats created within that specific project workspace.

Setup on Claude, ChatGPT, & Gemini

  • Claude Projects (Claude.ai): Click "Projects" on the left sidebar → click "Create Project" → click "Add Content" to upload files → click "Set Custom Instructions" to lock in your writing/tone rules.
  • ChatGPT Projects (Teams/Enterprise): Open your Workspace → click "Create Project" → upload your manuscript chapters and prompt instructions directly to the project folder.
  • Gemini Library & Gems: Go to Google AI Studio → use the "Library" and "Saved Prompts" to group related prompts, or use "Gemini Gems" to combine uploaded files and rules.

API Projects (Groq & OpenRouter)

  • No Native Portal: Because Groq and OpenRouter are developer APIs, they don't have consumer websites with project folders.
  • Local Projects: Instead, you manage projects using local developer tools and IDE agents (like Claude Code, Antigravity, or Cline) that link local folder paths on your computer to the API backend.

4. Skills

🧸 The Analogy (For an 8-year-old) Imagine you have a magical puppy. You teach him a trick called "Sniff Out Repeating Words." Whenever you say that trick name and show him a paragraph, he runs over and barks at the words you used too many times!

A Skill is a packaged, task-oriented workflow folder. Unlike basic chats, in agentic environments like Claude Cowork, Claude Code, and ChatGPT Codex, a Skill is saved directly as a tool that the AI can choose to activate based on your phrasing.

How Skills Work in Codex & Cowork

  1. Folder Packaging: A Skill is stored as a directory containing a SKILL.md file and a references/ folder containing bibles, voice rules, and passes.
  2. Metadata Registration: At the very top of SKILL.md, there is a YAML frontmatter block with a name and a description: 
    ---
name: zenith-hall-editing
description: Trigger on any request that involves editing, revising, or cleaning Zenith Hall manuscripts...
---
  3. Automatic Triggers: Place the folder in your agent's directory. The agent reads the description. When you type trigger phrases (e.g. "audit chapter 3" or "fix the yeses"), the agent automatically loads the skill's reference rules and executes the editing workflow.

Skills on API (Groq/OpenRouter)

  • Programmatic Triggers: To use skills with API backends, local agent scripts or wrappers monitor your input chat. When they detect trigger keywords, they programmatically append the skill's reference files and instructions to the API payload before sending it.

5. Plugins & Custom GPTs

🧸 The Analogy (For an 8-year-old) Imagine your AI is a superhero. A Plugin is like giving the superhero a gadget, like a magnifying glass to search the internet. A Custom GPT is like inviting a clone of the superhero who is wearing a chef hat and only knows how to bake cookies!

Plugins (extensions) allow the AI to connect to other apps (like Google Drive, Spotify, or the web). Custom GPTs are customized versions of ChatGPT created by users to do one specific job.

Usage on All Platforms

  • ChatGPT (Custom GPTs): Click "Explore GPTs" on the sidebar → search for a custom writing bot → click "Start Chat". The bot is pre-loaded with specific instructions.
  • Gemini (Extensions): Open Gemini → type "@" in the message bar → select an extension (e.g. @Google Drive or @Google Docs) → type your query (e.g. "@Google Drive read my draft.txt file").
  • Claude (Tools & MCP): Supports Model Context Protocol (MCP) and API tools, allowing Claude to read local files, search the web, and execute terminal scripts when requested.
  • Groq & OpenRouter (JSON Function Calling): Pass JSON schemas detailing available external tools in the API payload, allowing the model to return structured tool calls to fetch data or run terminal scripts.

6. Core Controls: Roles, Memory, & Temperature

🧸 The Analogy (For an 8-year-old) Imagine your AI is an actor in a play. The System Prompt is the stage director whispering instructions from the curtains ("You are a quiet, spooky vampire"). The User Prompt is you in the audience asking a question ("What's your name?"). The Context Window is how many script pages the actor can remember before he starts forgetting the beginning of the play. And the Temperature is a dial: set it to 0 to make him say his lines exactly as written, or set it to 1 to let him run around and make up wild stories!

To control your writing environment, you must understand how to manage roles, memory boundaries, and creativity levels across all platforms.

System vs. User Roles

  • Claude & ChatGPT: Keep instructions (how to write, styling sheets, character sheets) in the System Instructions / Custom Instructions / Projects layer. Use the active chat box only for your immediate commands (e.g. "Edit this paragraph").
  • Gemini: In Google AI Studio, write your rules in the left-hand column labeled "System Instructions". Do not type them directly in the chat box!
  • Groq & OpenRouter Playground: Write your rules in the System Message block. This keeps your writing guidelines clean and distinct from the raw text you ask it to analyze.

Context Windows (Memory Limits)

  • Claude: Claude Projects show a visual percentage bar indicating how full your project's memory is. If it approaches 100%, delete old chat sessions to free up space.
  • ChatGPT: Has a memory limit of about 32,000 to 128,000 tokens. If a chat session grows too long, the AI will silently forget the earliest chapters of your book. Start a fresh chat periodically.
  • Gemini: Has a massive 2,000,000 token context window. It can hold entire multi-book series in memory simultaneously without forgetting anything.
  • Groq & OpenRouter: Playground tools show an active token counter. Be mindful of limits when pasting large manuscripts.

Temperature (The Creativity Dial)

  • Standard Models (Claude 3.5 Sonnet, GPT-4o, Gemini 1.5):
    • 0.0 (Cold/Strict): Best for line editing, grammar, and formatting. The model outputs predictable, highly logical text. Set this in the side panel of AI Studio, Groq Playground, or OpenAI Playground.
    • 0.7 - 1.0 (Balanced/Creative): Best for brainstorming plot ideas, character voices, and drafting.
    • Groq & OpenRouter Note: Do not set above 1.2, or the model will produce gibberish.
  • Modern Thinking Models (OpenAI o1/o3-mini, Gemini Thinking, DeepSeek-R1):
    • Why Temperature is Obsolete: Reasoning models think behind the scenes—drafting, testing, and correcting their own logic before writing their final answer. If you mess with the temperature dial, you break their logical chain-of-thought!
    • What to Do: For these models, temperature is locked or disabled. Leave it on default (or 1.0) and let the model manage its own internal creativity and accuracy.