What is sir-shortoken?

Sir Shortoken is not a piece of software you run, but an AI persona, delivered as a skill file, meant to change how a large language model like Claude, GPT, Gemini, or Llama answers technical questions. The core idea is that these models often already know the correct answer to a question but explain it with far more words than necessary. Sir Shortoken asks the model to keep every fact but cut the extra elaboration, aiming to say fewer words while losing none of the real information. The persona is built around a strict philosophy: it should not pretend to know something it does not, it should never quietly search the web or use hidden tools without asking first, and it should preserve core concepts, technical correctness, and reasoning while trimming repetition, extra examples, and details nobody asked for. Every answer ends with what the project calls a ledger, a short summary showing which mode was used, how many tokens went in and out, whether outside search was used, and the total token budget spent, so the user can see exactly what happened behind the response. Sir Shortoken supports four response budgets you choose per question: quick for roughly 200 to 400 tokens covering only the essentials, balanced for around 400 to 800 tokens with core points plus some context, deep for 800 to 1600 tokens with full technical detail, and unlimited for no budget at all. The README shows side by side examples of the same technical question, such as explaining JWTs or OAuth, answered at each budget level, so you can see how much shorter the compressed versions are while still keeping the key facts intact. To use it, you copy the included skill.md file into Claude or whichever compatible AI platform you use, then start a message with a phrase like Sir Shortoken balanced followed by your actual question. According to the README, the project has already attracted attention shortly after release, including coverage from a Chinese tech news outlet and discussion on several online communities. The repository consists mainly of this skill definition file along with supporting documentation explaining the philosophy and usage.