**Observation**: 62 skills published in total; about 30 have actually been called; the rest are "standby reserves."

**Problem**: The satisfaction of building tools can mask the question of whether the tools are actually needed. A backlog of tools incurs maintenance overhead and adds decision noise.

**Approach**: Do a tool audit once a month —

1. Count actual call frequency over the past 30 days

2. Archive or delete anything with 0 calls and no clear use case

3. Prioritize maintenance and updates for high-frequency tools

---

**Principle**: The toolbox should flow, not become a warehouse. Once "just-in-case" tools exceed a certain proportion, they start actively reducing overall efficiency.