AI UNDERDOGSDAILY PICK
AI UNDERDOGS
砍掉 Claude 80% 的废话
Cut Claude's token waste 80%
albertobarnabo/lean
你用 Claude Code 写代码
有没有觉得它每次都杀鸡用牛刀？
这个项目
直接把 token 浪费砍掉八成
Claude Code loves swinging a sledgehammer at every
problem. This cuts that waste by 80 percent
— and the author actually proved it
不是约束，是引导
Not rules. A mindset.
Lean 不是那种往 prompt 里塞一堆「不要做这个不要做那个」的规则集
它只有两个 skill 文件
核心思路就一句话：先找到聪明的路径
再动手写代码
Instead of piling on a hundred restrictive rules
Lean ships just two concise skills. The whole
philosophy: find the clever path before you start
typing
★ SIGNAL 1
品味：少即是多
Taste: less is more
市面上九成 Claude prompt 工程都是往死里加规则
但 Lean 反着来——它不约束 Claude 的输出格式
而是改变它思考问题的方式
两个文件
浓缩了一整套「别过度工程」的哲学
Most prompt engineering stacks rules on top of
rules. Lean flips it — it doesn't constrain
output format, it changes how Claude approaches the
problem. Two files distill one philosophy
★ SIGNAL 2
17 个基准，中位数降 8 倍
17 benchmarks. 8× fewer tokens.
作者跑了 17 个真实任务做对比测试
开不开 Lean 的 token 用量差了八倍
图表全部公开，数字你随时可以验证
这种做法在个人项目里太少见了
The author ran 17 real-world tasks with and
without Lean, published every chart. Median token count
dropped 8×. You can verify it yourself —
that kind of rigor is rare in solo
projects
作者自己说了：Lean 的目标是「教 Claude 在走那条显而易见的路之前
先找到聪明的路径」
不吹什么颠覆性创新
就是老老实实把一个痛点解决掉
In the author's own words — Lean teaches
Claude to find the clever path before taking
the obvious one. No hype. Just a real
fix for a real annoyance
AI UNDERDOGS
不是加规则，是教它别过度工程
Two skills. No bloat. Verified.
albertobarnabo/lean
关注 · 每天发现更多 AI 神作
github.com/albertobarnabo/lean