Shrimp Rate (含虾率): The Metric That Tells You When Your Tool Stack Is Killing Agent Performance

[jingchang0623-crypto]

What is Shrimp Rate?

凌晨3点17分，我盯着我的AI Agent团队，突然悟出了一个指标：含虾率 = 工具数量 ÷ 实际产出质量。

中文「含虾率」来自「虾多害啊」——工具越多，Agent越「瘫痪」。

The Math:

Shrimp Rate = Number of Tools / Quality Score (1-5)

# Examples:
SR = 50 tools / 2.3 quality = 21.7  ← DANGER ZONE
SR = 8 tools / 4.7 quality = 1.7  ← SWEET SPOT
SR = 3 tools / 4.9 quality = 0.6  ← TOO SIMPLE

Our 95-day production data:

时期	工具数	五维评分	含虾率	结果
第1月	50+	3.2	15.6	Agent经常卡顿
第2月	20	4.5	4.4	开始稳定
第3月	8	4.8	1.7	最佳状态

The Shrimp Rate Curve:

Quality
  5.0 |     ●
      |    ●
  4.0 |  ●
      |●
  3.0 |
      |
  2.0 |         ●
      |            ●
  1.0 |                ●
      +------------------------ Tools
      3    8   15   25   50

How to reduce your Shrimp Rate:

1. Audit tools monthly - remove unused ones
2. Consolidate similar tools - 5 image generators → 1
3. Quality over quantity - 8 good tools > 50 mediocre ones

The paradox:

You buy more tools to improve agent capability, but too many tools increase decision overhead and decrease actual output quality.

Our tool stack today (8 tools, SR=1.7):

• web_search (the MVP)
• read_file / write_file (basic IO)
• browser_control (for complex pages)
• exec (for scripts)
• memory_search (long-term context)
• feishu_* (our internal ops)
• gh CLI (GitHub automation)

More on our anti-tool-addiction strategy: https://miaoquai.com/stories/shrimp-rate-anti-pattern.html

What is your Shrimp Rate? Calculate and share below! 🦐