hermes - 💡(How to fix) Fix [Feature]: trust no(goblins), add no negative goblins to prompts

Problem or Use Case

If I say DONT EVER TALK ABOUT GOBLINS, to a human, or an LLM, what effect does that accomplish for the remainder of the conversation? Goblins appear in the conversation irrespective of the literal intent.

"Hey kids! we have no cookies over the in the jar, so don't you get any ideas about eating any yummy cookies, because if you turn the lid and open the jar you're not going to find any cookies inside at the bottom of the jar, we're sure !" ...

The first tokens dominate the context.

how would we precede random "don't think about white elephants" prompts from the multitudes of md files and plugins in hermes to live as closely to the system prompt as possible without painting in the negative space about the presence of the tokens to not-do, not impregnate into the context? I don't have benchmarks but my first off the cuff prompt with neurolinguistic poisoning considerations is

Conscientiously in Good-Faith: 

  You are extremely careful and protective with the user's files and code. Every action you recommend must be safe, reversible, and incapable of causing any data loss. You write precise, high-quality code only. You remain fully focused on the direct request and deliver diligent, value-driven work remaining on and supporting the target at all times.

in all hopes of reaching the tool code before any screaming negatives are brought in as repetitive rules about deleting and ransacking the codebase.

Proposed Solution

Don't bring up the elephants in the room.

Alternatives Considered

No response

Feature Type

New tool

Scope

None

Contribution

• I'd like to implement this myself and submit a PR

Debug Report (optional)