It was 3 AM on day two of a 72-hour jam. My game was, on paper, completely done: a combat platformer with three levels and a boss, all the systems working, no crashes. It also had the emotional depth of a tax form. There was no story, no world, nobody to talk to — just the player avatar punching things in a pretty void. I had six hours of clock left and one increasingly warm energy drink, and I made a bet: the highest-value thing I could do with those six hours wasn't another mechanic, it was giving the place a memory.
Different angle, same tool
My first write-up of this prompt was a marathon — 40 NPCs over 12 days. This was the exact opposite: a dead sprint where I needed three characters to land instantly and survive with essentially zero revision, because there was no time for a second pass. I ran the Branching NPC Dialogue Tree With Personality, Lore Gates, and Dead Ends prompt on Claude Opus 4.8 three times, back to back, and committed to using what came out.
Character one: a wounded soldier at the mouth of level 1 — gruff, matter-of-fact, drops a real hint about the boss's weakness if you choose the right branch. Character two: a strange child in level 2 who talks in riddles and, only if you keep pushing past the point of comfort, reveals why the world broke. Character three: the boss's old mentor at the level 3 gate — sorrowful, conflicted, optional context that recolors the entire final fight if you bother to find it. Three people, three voices, three secrets.
Because I was racing the clock, I leaned hard on the lore-gate mechanic instead of on branch width. I set player_options_count to 2 to keep each tree tight and fast to implement, and I let the [LORE REVEAL] tag carry the emotional weight that I didn't have time to build any other way.
- Three complete dialogue trees generated and wired into the game in under 3 hours
- Judges' written notes mentioned 'surprisingly rich world-building for a 72-hour project'
- The riddling child was the single most-commented-on element in all player feedback
- I slept for two whole hours before the submission deadline, a personal jam record I am still proud of
Why the gated lore did the heavy lifting
The reason these three felt intentional rather than bolted on at 3 AM is the reveal mechanic. The information isn't free. You only learn why the world collapsed if you actively choose to keep pushing the child past the point where a polite player would stop. Players who found the full path felt like they'd discovered something nobody handed them, and that specific feeling — earned discovery — is exactly what you cannot manufacture by hand under a deadline. The prompt turns it into a checkbox: name the lore in one field, and it gets gated behind a real choice automatically, with the [LORE REVEAL] tag telling me precisely which branch to wire the trigger to.
The advanced tweak I'd pass on to anyone trying this: write the secret as a single concrete sentence in lore_to_reveal, never a paragraph. 'The boss is the child's father' beats three sentences of backstory every single time, because Claude then earns its way to that one devastating line through the dialogue instead of dumping a wiki entry on the player. Constraint forces craft. The smaller the secret, the bigger the reveal lands.
Why two options beat four under deadline
I want to defend the player_options_count of 2 specifically, because the reflex under pressure is to crank everything up — more branches, more choices, more content. That reflex is wrong at 3 AM. Every extra branch is another set of lines I have to read, implement, and connect to a game state, and a tree with four top-level options and sub-branches off each is a small writing project on its own. Two options keep each tree shallow enough to wire in minutes while still giving the player a real choice — talk straight, or push too far — which is all the gated reveal actually needs. The riddling child has exactly two paths and is the most-loved character in the build. Width is not depth. A narrow tree with a real secret at the bottom beats a sprawling one with nothing hidden in it, and it beats it on implementation time too, which under a jam clock is the metric that actually decides what ships.
If your jam game is mechanically finished but emotionally hollow at 3 AM, grab the Branching NPC Dialogue Tree With Personality, Lore Gates, and Dead Ends prompt on Prompt Dock, give it two minutes per character, and let three good NPCs do what an hour of cutscene writing can't. When the jam's over and you have a quest-giver who needs an actual quest, I hand them straight to my Full Questline Generator — Objectives, Twist, Branching Endings, and Rewards prompt. It's the cheapest sense-of-history that money can't buy and an energy drink can.