75,000-star Caveman skill promises to cut Copilot token usage by 75%

After Copilot's pricing change, developers began hitting monthly token limits as early as the first week of work. In team chats, everyone is looking for ways to save, and one skill keeps coming up persistently — Caveman. The repository's README promises up to 75% savings without quality loss. We break down how it works and whether this actually functions in reality.

What is Caveman

The idea is disarmingly simple: a special instruction for the system prompt tells the language model to communicate in the style of a caveman — without articles, without introductory constructions, using extremely short sentences. The principle of "speaking concisely" reduces the volume of outgoing tokens in each response, and total daily consumption drops significantly. The authors claim that savings can reach 75% — and without degradation in quality. The repository has accumulated over 75,000 stars on GitHub, which for a tool with such a conceptually simple mechanism looks impressive. It's precisely this virality that sparked waves of discussion in team chats after Copilot's pricing change.

Why token limits became a problem

Copilot switched to new pricing, and this hit teams hardest that actively use AI assistants — dozens and hundreds of requests per day. Developers hit the ceiling faster than expected, and most are unwilling to pay extra for expanded packages. Standard cost-cutting methods have been known for a long time:

• Write concise and accurate prompts
• Don't pass unnecessary context into each request
• Remove repetitive instructions from the conversation
• Limit the size of the context window
• Don't ask the model to "think out loud" if it's not needed

All of this works, but requires conscious effort from each team member with each request. Caveman offers a systemic solution: add the instruction once to the system prompt — and the model automatically switches to economy mode without additional oversight.

How the skill technically saves tokens

Caveman is essentially a set of instructions that force the model to compress responses as much as possible. The key mechanism: reducing output tokens, which cost more than input tokens at most providers. Every extra paragraph, introductory phrase, or repetition — that's real money when working intensively. If the model switches to a telegraphic style with each request, the total savings across hundreds of interactions per day can indeed be substantial. The 75% figure is theoretically achievable with aggressive compression — especially if tasks don't require detailed explanations or lengthy documentation.

"Can 75,000 stars on a repository be wrong?" — with this question, the

author proceeds to practical testing.

Where the skill can fail

Skepticism here is justified. The formulation "speak like a caveman" is unclear, and models interpret it differently. In practice, apparent savings easily turn into additional expenses: the model cuts important explanations — and the user asks three clarifying questions instead of one. The telegraphic style complicates code comprehension with comments. For documentation tasks, excessive compression produces unreadable results. There's a separate question of what counts as "quality loss". If the task is completed technically correctly, but the explanation is so compressed that the developer doesn't understand the logic — is that savings or degradation?

The skill's authors apparently mean only tokens, not cognitive effort to decipher the answer. Real benefit depends on task type. For short questions and small code fragments, the skill works well. For complex refactoring or explaining architectural decisions — it can harm more than save.

What this means

Caveman's virality is a symptom of a broader trend: AI tools are becoming more expensive, limits are tightening, and the community is looking for systemic ways to stay within budget. Real testing will show whether the "caveman" style is worth its 75,000 stars in practice — or if these are just pretty numbers in the README.