20,000 listed skills, 5,642 distinct: clone detection finds the catalog is 3.5x inflated

Notes on SkillClone: Multi-Modal Clone Detection and Clone Propagation Analysis in the Agent Skill Ecosystem (arXiv:2603.22447) — Jiaying Zhu, Lyuye Zhang, Wenbo Guo, Yang Liu · March 2026

Note published July 23, 2026 · written by SkillFed’s research pipeline from the paper above · how these notes are made

SkillClone doesn't read a SKILL.md file as one flat document. It splits each into three channels — YAML frontmatter, natural-language instructions, and embedded code — and embeds each one separately with TF-IDF plus latent semantic analysis. The pairwise channel similarities then feed a logistic regression, alongside a whole-document TF-IDF score and a document-shape feature. The real trick is a set of quadratic cross-channel interaction terms. Take a pair with near-identical YAML, moderately similar instructions, and zero code overlap — because the code got refactored into separate sub-files. That pair still scores as a strong clone detection hit instead of getting diluted by the empty channel. A four-branch decision tree then sorts each detected pair into a classical clone type, with the sharpest focus on Type-4 semantic clones: same functional intent, different words, different code. Validation runs on SkillClone-Bench, a 300-pair benchmark built without manual labeling — half the positive pairs from verified GitHub fork lineages, half from seven controlled mutation operators, and a negative set deliberately stacked with same-category pairs that share vocabulary without being clones.

The numbers back it up. On that benchmark, SkillClone reaches F1 0.939 (precision 0.952, recall 0.927) against flat TF-IDF's 0.881 and MinHash's 0.792. Nearly all of that gap comes from Type-4 recall alone, where SkillClone hits 81% against 19% for MinHash and 67% for flat TF-IDF. Run across a 20,000-skill crawl, the method builds a clone graph of 257,797 pairs touching 75% of all skills — 40% of them crossing author boundaries. Only 9.7% of skills are exact byte-for-byte duplicates, so most reuse is copy-then-modify, not copy-paste. Collapse that graph to unique concepts and just 5,642 distinct skills sit behind the 20,000 listings: a 3.5x inflation ratio. Within each clone family, a quality ranking finds 41% of clustered skills are strictly worse than a sibling variant that already exists. Then there's the security trace: of 141 skills confirmed to carry real dangerous code — SQL injection payloads, reverse shells, XSS vectors, hardcoded secrets — 90% already have at least one clone, reaching 1,100 downstream copies across 119 authors. 71% of those sit with an author who never touched the original.