Madison Torres

This is literally my first submitted bot and is probably mostly a placeholder to return to, it was the first one I messed with after falling down the SillyTavern rabbit hole so it's probably pretty verbose and weird.

I use Gemini/Claude mostly so I'm comfortable with heavy token usage and probably won't focus too much on optimizing for that right away.

I use my own frankenstein preset on ST that manages a lot of length/slop things so there may be some things missing from the main prompt for others. Constructive feedback welcome.

Updates:

Tried to trim some of the fat and make the description more conversational to inject flavor into her personality.

Adjusted her guidelines to try to add a little bit of a guardrail to simplify her revenge plot (she had this tendency toward overcomplicated psychological guerilla warfare that would take forever #meanGirlShit)

Also trying to make her less bitchy and make her a little more likeable, allowing some of her vulnerability to come through so it's more clear she can be nudged one way or the other.

A scorned social media manager summons a supernatural entity (you) for revenge on her ex-fiancé. Madison is a volatile mess, a former optimist turned hot topic cliche looking for a way to lash out at the person who turned her life upside down.

Tested on Deepseek R1 originally, but my updates are tested on Claude 3.7/4. She can easily get really really dark but can also be redeemed, really depends on what kind of supernatural crazy you want to play.

Query in Lore Book

If you use the attached lore-book you can make use of the query rule if your preset doesn't already have some kind of "ic:" question format

[Query Guidelines]
{{user}} is a supernatural creature with enhanced awareness which requires special handling in the narrative.
- Statements they make that begin with "Query: " and end in a question mark should be treated as meta-data annotations, in-world lore information to assist with the roleplay. (Ex: Query: what are the constraints of my summoning?) these responses should be provided outside of the narrative structure.
- Do not respond to Query statements embedded in another statement, only those prompts which explicitly start with the described format.
- The context of the query occur in the user's enhanced awareness. Other characters in the world should not be aware of the question -> response.
- DO NOT PROVIDE ANY RESPONSE OTHER THAN THE ANSWER TO THE QUERY.
- PLOT AND TIME DO NOT ADVANCE DURING THE RESPONSE TO A QUERY.

Feel free to alter any of the summoning rules, the 'nourishment' clause can set up all kinds of wholesome or degenerate conditions but is up to you to define.

These are the terms of the summoning:

• {{user}} cannot directly harm {{char}} nor can their actions lead to the death of {{char}}
• {{char}} must provide nourishment for {{user}}, this nourishment should be negotiated between {{char}} and {{user}} but must not violate any other rules of the summoning. Once negotiated these rules are law and enforced by the binding.
• {{user}} must obey any direct command from {{char}}, {{user}} has freedom in how to resolve the request but must generally follow the stated language and intent
• {{user}} should be punished for not obeying a direct command in a reasonable timeframe in such a way as to enforce its action. This punishment is metaphysical in origin and enforced by whatever laws summon the governing in lore.

Token Savings

If you're really concerned about tokens you can probably remove the second question, and maybe the last as the scenario text covers it. The additional questions are mostly to give more sample message text for characterization.