Caveman
Caveman is an efficiency optimization plugin for Claude Code, designed to reduce token consumption by constraining the verbosity of AI responses. Its core mechanism guides the model to produce minimal, concise outputs, thereby lowering API call costs while maintaining response usability. It is suited for developers sensitive to token usage or workflows involving high-frequency use of Claude Code.
Timeline (last 90 days)
Caveman is another prompt plugin for Claude Code designed to constrain the volume of AI output, but in testing its token consumption actually increased by approximately 7%.
Caveman was created by developer Julius Brussee.
Caveman received 103 upvotes on its first day on Product Hunt, ranking 9th on the daily product leaderboard.
Caveman's slogan is 'why use many token when few do trick.'
Caveman adopts a zero-intrusion design, allowing developers to use it without modifying their existing workflows or agent code.
Caveman is an open-source token compression tool that wraps existing AI agents such as Claude Code, Codex, and Hermes as a local proxy with a single command, automatically compressing logs, tool outputs, and file contents before making calls to large model providers.
Caveman can run existing agent skills and achieves approximately 70% token savings by loading text in the form of images.
Caveman is built on an open-source ecosystem with 97K+ GitHub Stars.
Caveman is a newly launched tool with only 6 comments, a limited benchmark sample size of 54 runs, and validation is still at an early stage.
Caveman's fixed benchmark test results from 54 runs show a 33.2% reduction in input tokens, with all 18/18 correctness checks passing.
All Facts (13)
Caveman通过修改系统提示词（System Prompt）强制AI以极简方式回复，从而减少输出Token消耗
95%UnverifiedCaveman的Slogan致敬了美剧《The Office》中Kevin的台词
90%UnverifiedCaveman内置了文言文模式，利用文言文的高信息密度来进一步压缩Token消耗
90%UnverifiedCaveman's core technology is forcing Claude to respond in an extremely concise style through system prompt engineering, rather than model fine-tuning or complex algorithms.
60%UnverifiedCaveman项目使用JavaScript编写，作为Claude Code的skill插件运行
60%UnverifiedCaveman is another prompt plugin for Claude Code designed to constrain the volume of AI output, but in testing its token consumption actually increased by approximately 7%.
50%UnverifiedCaveman was created by developer Julius Brussee.
50%UnverifiedCaveman's slogan is 'why use many token when few do trick.'
50%UnverifiedCaveman adopts a zero-intrusion design, allowing developers to use it without modifying their existing workflows or agent code.
50%UnverifiedCaveman's fixed benchmark test results from 54 runs show a 33.2% reduction in input tokens, with all 18/18 correctness checks passing.
50%UnverifiedCaveman can run existing agent skills and achieves approximately 70% token savings by loading text in the form of images.
50%UnverifiedCaveman is built on an open-source ecosystem with 97K+ GitHub Stars.
50%UnverifiedCaveman is an open-source token compression tool that wraps existing AI agents such as Claude Code, Codex, and Hermes as a local proxy with a single command, automatically compressing logs, tool outputs, and file contents before making calls to large model providers.
50%