StackMap
Subscribe

n-skills vs SkillNet

Curated skill marketplace for AI coding agents on the universal SKILL.md/AGENTS.md format — write a skill once, install it into Claude Code, Codex, Copilot, Cursor and friends. — versus — Open skill infrastructure from ZJU NLP: search a 500K+ indexed skill library, install, generate skills from repos/docs/traces, score their quality, and compose/orchestrate them. SDK + CLI + MCP.

The curated verdict

Both distribute reusable skills on the SKILL.md format; n-skills is a small curated marketplace, SkillNet is a 500K+ crawled-and-deduplicated index with quality ranking.

n-skillsSkillNet
Stars1.0k1.1k
Forks105127
LanguageTypeScriptPython
LicenseApache-2.0MIT
Last activitytoday5 days ago
Topicsskillsskills
Curated connections43

n-skills — the curator's take

The 'write once, run everywhere' bet applied to agent skills: one curated marketplace on the SKILL.md + AGENTS.md conventions, installed via openskills into whichever agent you drive. Smaller and more opinionated than the mega-registries — curation IS the pitch, every skill is hand-picked. NOT a discovery engine: if you want breadth across hundreds of sources, a package manager with security scanning covers more ground; and it's one person's taste with ~1k stars — check the skills you'd rely on exist before adopting the workflow.

SkillNet — the curator's take

Use it when you want skills treated like packages, not pasted folders: semantic search over a 500K+ GitHub-skill index is credential-free, and `evaluate` gives a safety/completeness/executability score before you drop a random skill into an agent. Create/evaluate/analyze need an OpenAI-compatible endpoint; `orchestrate` additionally needs a Claude-Agent-SDK-capable gateway and ships exactly one scene (sciatlas) so far. Prefer skillkit or asm when you just want install plus cross-agent format translation — SkillNet's edge is the research layer: skill creation from execution traces, quality scoring, and skill-relationship graphs, with an arXiv report behind it.