Skip to main content

​
Cluster and Portfolio Analysis

fastskill analyze helps you evaluate quality at system level, not only case level.

​
1) Cluster analysis

Group skills by semantic similarity to detect taxonomy health: 
fastskill analyze cluster
fastskill analyze cluster -k 8 --min-size 2 --json
What to look for:
  • Very large clusters: broad or overlapping skill definitions
  • Too many tiny clusters: fragmented naming or inconsistent scope
  • Unexpected cluster members: missing or unclear skill descriptions

​
2) Similarity matrix

Inspect relationship density: 
fastskill analyze matrix --threshold 0.8 --limit 5
Use this to understand whether a proposed new skill is truly distinct.

​
3) Duplicate detection

Detect likely redundancy: 
fastskill analyze duplicates --threshold 0.92 --severity high
Use this regularly to keep catalog maintenance costs low.
  • On every PR affecting skill content: duplicates + smoke evals
  • Weekly: cluster + matrix review
  • Before release: full eval suite + duplicate cleanup pass

​
Cluster quality checklist

  • Cluster distribution looks balanced for current catalog size
  • High-severity duplicates are resolved or justified
  • New skills do not overlap existing core skills above threshold

​
See also