What Is This?

The file prompt.md in this directory is a ready-made prompt that you feed to a large language model (LLM) to have it generate a complete quest pack for your BBS. The LLM acts as a collaborative quest designer: it interviews you about your BBS's identity, builds a coherent world around your answers, and then produces every source file the devkit tools need to compile into a running campaign.

This document explains how to get the best results out of that process. It does not cover the devkit tools themselves -- see clandev.txt for that.

What Does It Produce?

The prompt instructs the LLM to generate eleven output files:

1. quests.ini — Quest registry
2. quests.hlp — In-game quest descriptions (one block per quest)
3. NPC Info .txt — NPC metadata and topic list
4. Monster .txt — Combat stats for monsters and NPCs
5. Chat .txt — NPC dialogue and topic blocks
6. Event .evt — Quest encounter logic
7. Listing .lst — Two-column file mapping filenames to PAK aliases
8. clans.ini — Complete configuration file with NpcFile entries
9. readme.txt — Compilation and installation instructions
10. build.bat — Windows batch file for compilation
11. Makefile — Unix makefile for compilation

Files 1-8 are everything needed to build and install the pack. Files 9-11 exist so you don't have to figure out the compilation order yourself.

Context Window Size

Context window size is the single most important variable. A full campaign with 30+ quests, 40+ NPCs, and interlinked scripts will be tens of thousands of tokens of source material. Models with small context windows will truncate their own output, lose track of flag assignments halfway through, and produce NPCs that reference events from a quest the model has already forgotten.

As a rough guide:

Use the largest context window available to you. If a model offers both a standard and an extended context option, use the extended one.

Reasoning Mode

Most frontier LLMs offer a "reasoning" or "extended thinking" mode that causes the model to work through the problem at length internally before writing its response. This makes a meaningful difference for quest generation.

The prompt asks the model to design consistent cross-references between dozens of files, track flag assignments so no two flags serve the same purpose, and ensure every ACS condition in every chat file is actually tested by at least one command in an event script. Reasoning mode makes these kinds of errors far less frequent.

Recommended Models

Based on testing as of early 2026 (capabilities change; treat this as directional):

Longer outputs and more complex campaigns consistently benefit from larger context and stronger reasoning ability. If the budget allows, go bigger.

Loading the Prompt

The entire contents of prompt.md must reach the LLM before any conversation begins. How you do this depends on your interface:

Claude.ai

Open a new conversation. Paste the entire contents of prompt.md as your first message, then send it. Wait for the model to acknowledge before continuing.

API-based Interfaces (OpenAI, Anthropic API, etc.)

Place the contents of prompt.md in the system prompt field. Leave the first user message empty or use it to begin the interview.

Web Interfaces with File Upload

Upload prompt.md as an attachment and tell the model to read it before responding. Some models require explicit instruction to read attached files; say so clearly.

What files will you generate?

It should list all eleven output files without hesitation. If it lists fewer, re-paste the prompt or try a different interface.

The World-Building Interview

The prompt instructs the model to interview you before generating anything. It will ask six questions, one at a time:

1. The village — what it is, what it feels like
2. History and hidden things — what players can discover
3. Power and conflict — factions in tension
4. The world beyond — surrounding geography
5. Tone — the register of dialogue and consequence
6. The sysop's vision — what you most want players to feel

Answer as specifically as you can. Generic answers produce generic quests. The model needs concrete names, specific conflicts, and particular locations to produce content that feels like it belongs to your BBS rather than any BBS.

Skipping the Interview

If you already have a clear world in mind, you can skip the interview by writing a world description up front. Begin your first message (after loading the prompt) with something like:

I have prepared the following world description in place of the
interview.  Please take careful notes on every specific detail --
names, factions, locations, events -- and use it as the foundation
for the entire pack.  Do not proceed to generation until you have
confirmed your understanding.

[Your description here]

Include all six interview topics in your description: village character, history and hidden things, factions in tension, surrounding geography, tone, and what you want players to feel by the end of the campaign.

Generation Takes Time

A full 60-day campaign with 30+ quests and 40+ ambient NPCs is a large body of work. Expect the LLM to take anywhere from 20 minutes to over an hour, depending on the model, output size, and server load. This is normal.

Most interfaces will show a spinner or progress indicator while the model is working. If the spinner stops and the model has not finished, see section 3.2.

Prompting the Model to Continue

Chat interfaces typically cap individual response length. The LLM will stop mid-generation and wait for input. When it does, send a short message:

Continue.

The model should resume from where it stopped without regenerating what it already produced. If it seems confused about where it left off, tell it:

Continue with the chat file.

or whatever file it was in the middle of. You may need to do this 3-5 times for a large campaign. After each continuation, glance at the output to confirm it picked up at the right point rather than jumping backward.

Output Order

The prompt specifies this generation order, and the model should follow it:

1. Monster/NPC definition file (.txt)
2. NPC Info file (.txt)
3. Chat file (.txt)
4. Event script (.evt)
5. quests.ini
6. quests.hlp
7. PAK listing file (.lst)
8. clans.ini
9. readme.txt
10. build.bat
11. Makefile

Later files depend on decisions made in earlier ones -- clans.ini references compiled .npc filenames; the listing file references all compiled output; the readme and build scripts reference all of the above. If the model skips a file, prompt it to produce the missing one before continuing.

What to Check Before Compiling

Read through the generated source files before handing them to the devkit tools. The ecomp compiler catches structural errors but not semantic mistakes. Here is what to look for:

Text Line Length

The engine does not word-wrap. Each Text command's visible content must be 78 characters or fewer. Color codes (|0C, |02, etc.) take up source characters but zero display columns -- only count the characters that will actually appear on screen.

Chained ACS Conditions

{Q1}{Q2}Command is a compile error. The correct form is {Q1&Q2}Command. Conditions are combined inside a single pair of braces using & (AND), | (OR), ! (NOT), and parentheses for grouping.

Condition on Its Own Line

A {condition} written on a line by itself is silently discarded. The condition and the command it gates must be on the same line.

Invalid Jump Targets

Jump must point to a real Event or Result block label. STOP and NextLine are not valid Jump targets -- they are only recognized by Fight, Option, and Input. A Jump to STOP will cause a runtime crash.

Missing End

Every Event, Result, and Topic block must close with End. Without it, execution falls through into the next block. ecomp will not catch this.

Quest Logic in Topic Blocks

TellQuest and DoneQuest are fine in Topics. Fights, major flag changes, and story-critical rewards must live in Event or Result blocks only. Quest logic in a Topic block bypasses the one-quest-per-day limit and breaks campaign pacing.

Flag Number Conflicts

Each flag number within a scope (P, D, G, T) should have exactly one purpose. If P3 is used as a story-progress flag in one file and a daily cooldown in another, one of them will silently stomp the other.

String and Path Lengths

Text/Prompt/AddNews arguments: 254 character maximum. File path aliases: 29 character maximum.

Asking the LLM to Fix Mistakes

When ecomp reports an error, paste the error message back to the model:

ecomp reported this error when compiling the event file:

fallstatt.evt line 47: Error: unknown condition 'Q1Q2'

Please fix it and show me the corrected block.

Models can usually diagnose and correct individual ecomp errors from the output alone. Fix one issue at a time rather than pasting a long error log and asking for a bulk fix -- bulk fixes introduce new errors.

Regenerating Individual Files

If a specific file needs significant revision -- a poorly designed quest, a chat file that does not match the established tone, flags that are inconsistent across files -- ask the model to regenerate just that file:

The chat file for the blacksmith NPC does not match the grim tone we
established in the interview.  Please regenerate just that NPC's Topic
blocks.  Keep all flag numbers and labels identical to what you already
generated.

If you need to change the world design significantly after generation (rename an NPC, move a quest to a different story beat, change a faction's role), make the change in the chat file and event script together, and ask the model to verify that all cross-references are still consistent before you compile.

The Readme the LLM Generates

The generated readme.txt lists the exact compilation commands for your pack in dependency order. Follow it sequentially. All devkit tools are in your game directory; the readme will prefix commands with ./ for Unix systems.

A typical sequence for a pack named fallstatt looks like:

./mcomp    fallstatt.mon.txt  fallstatt.mon
./makenpc  fallstatt.npc.txt  fallstatt.npc
./ecomp    fallstatt.q.txt    fallstatt.q
./ecomp    fallstatt.evt.txt  fallstatt.evt
./makepak  fallstatt.pak      fallstatt.lst

Run these in order. If any step fails, the error output will identify the problem. See section 4.2 for how to ask the LLM to fix it.

Build Scripts

In addition to the readme, the LLM generates:

Both run the same compilation sequence as the readme, in the correct order. Use whichever matches your server's operating system. If a step fails, the error output will name the tool and line that caused the failure.

Packaging for Distribution

Once the pack is compiled and tested, distribute the following files:

packname.pak    (compiled archive -- all scripts and NPC data)
quests.ini      (replaces existing -- back this up first)
quests.hlp      (replaces existing -- back this up first)
clans.ini       (replaces existing -- back this up first)
readme.txt      (installation instructions for the recipient)

Frequently Asked Questions