BoT: Balaur of Thought

A Silly Tavern script meant to force the square peg of common sense into the round hole of LLM-driven RP.

⭐Changes⭐ (It's gonna be a big one):

The User manual has been updated.
(almost) full rewrite.
Group chat support (finally)
/sendas and /trigger are handled properly.
/comment, /note, /sd and their variants are handled properly.
Databank is back in the form of long term memory for character and chat. A new [🗃️] menu was added and will be improved in tje future.
Optional delay between generations for backends that complain about too many requests (In the [🧠] menu).
Rethinking char message on character greeting prompts the LLM to generate a new greeting taking user persona into account.
You can now ask the LLM (not the character) arbitrary questions about the chat.
Rephrasing can now take arbitrary user-input as rephrasing criteria.
Injection strings can now be viewed and edited.

4.01 bugfixes:

BOTMKzINJECT BUG.
/; TYPO BUG
Typo in databank entry autogeneration prompt led to use an entirely different prompt.
Added new initial analysis delay option under the [🧠] menu, it is off by default.

Features

Prompts LLM to make inferences about the scene, spatial-awareness, and dialog. It also provides up to three courses of action, similar to tree-of-thought.
Only the last batch of analyzes are injected into the context for token economy on limited context sizes.
Analyses are available between sessions.
Allows user to see every past analysis and prompt, the latter in a neat, color-coded format.
Allows re-generating analyses in batch or individually, and add further one-time instructions in case the LLM is being an ass.
For cases of extreme LLM assery analyses can be edited manually.
In case I am being an ass, prompts can be edited bit by bit.
Databank support for long-term memory. Duplicate and contradiction control are based on memory topic.
Additional tools: Rephrasing, interrogation, translation capabilities testing, synching persona if it's switched during chat.

Every analysis is basically an extra generation, so with all analysis enabled, every char reply will take 4 generations. That is 4x the time and 4x the cost for pay-per-token plans.
Every LLM has it's own quirks, so your mileage may vary.
A few measures have been placed to harden script's logic and keeping it from falling apart on unexpected situations, however, ancelling stuff is still not recommended.
Since default syntax for injections and memories is XML-based you might need to enable show tags in chat in order to see them from the view menu [👀] or you can change the syntax manually.
Better results are achieved with good descriptions of user's persona. This affects some LLMs more than others though.

Catbox link: BoT 4.01
Mediafire link: BoT 4.01

Step 1:
%:50

Step 2:
%:50

Step 3:
%:50

Step 4:
%:50

Step 5:
%:50

Step 6:
%:50