InitialPhase55

Nope, the Director agent dynamically generates the world and storyline as you play! (unless you strictly specify an exact plot you want). Nothing is otherwise pre-determined or set in stone, and LLMs make the call/judgment on the spot. You have freedom to do basically whatever you want within reason, and the Director will adapt to keep it engaging and coherent.

Or at least that's the design intention; I haven't gotten much feedback so not sure if this is true to spec or if emergent oddities occur :)

Oh, and I'll keep topping up user quotas manually for the next 72h so you can keep experimenting freely, but if I forget, give me your username and I'm happy to give you enough balance for many hours of stories :)

Good point! Didn't consider that aspect, agree.

Curious, how did you settle on Haiku/Sonnet? Because there are much cheaper models on OpenRouter that probably perform comparatively...

Consider Haiku 4.5: $1/M input tokens | $5/M output tokens vs MiniMax M2.7: $0.30/M input tokens | $1.20/M output tokens vs Kimi K2.5: $0.45/M input tokens | $2.20/M output tokens

I haven't tried so I can't say for sure, but from personal experience, I think M2.7 and K2.5 can match Haiku and probably exceed it on most tasks, for much cheaper.

Might be more difficult for people with private repos, as I recall Codeberg doesn't like private repos on their platform.

I think the easiest solution to prevent such injections is just to render PDFs by page into images, and feed the images into whatever downstream task. I believe this is enough to prevent the attacks techniques listed in the repo :)