v0.23.2 Autonomous npc QoL
What's Changed
This is mostly a bunch of fixes for autonomous npcs. I noticed that they didn't follow the prompt very well, and have attache the event history to the end to give them something to 'attach' to.
There are also a bunch of other smaller changes. I got rid of the 'tell' cache, replacing it with the existing 'event' cache, since I only wanted one 'history' in the context.
- Autonomous agents follow a very simple 'chain of thought' logic, where they state a goal, express their thoughts about it, and then generate the action. The 'thoughts' are sometimes more interesting than what they say (and indeed sometimes they don't say anything), so I added a bool (only applies to those with 'autonomous' true) that will tell their thoughts as well. This is 'output_thoughts' in the character json.
They still require work, but have become better.
-
Fixed prompt context when an npc 'arrives' after being loaded.
-
Turn-based (
server_tick_method = TickMethod.COMMAND) stories now work better. Previously npc actions were delayed until the next prompt until they became visible. -
One small new feature is a wizard command:
create_item. This will, non-surprisingly, add an item to the current location. It will first search among 'catalogue' items for the story to see if it can instantiate an existing type. Failing that, it will try to parse the type among existing item classes (Food, Health, Wearable, Weapon, etc), or just create a generic Item, if it fails that.
!create_item type name description
"What happened to v0.23.1?" the observant might ask. It was just a hotfix for the save story dialogue not being clickable. No release made.
Full Changelog: v0.23.0...v0.23.2
