
First controlled benchmarks: 96→100% success on mass spec, but weak models auto-trigger the skill only 8% of the time.
K-Dense-AI/scientific-agent-skills and flagged the same gap in every sentence: no benchmarks. The install was clean, the skill catalog was deep, the "160,000+ scientists" claim was unverified. The honest answer on whether these skills actually improved agent performance was: we don't know yet."The skill's biggest contribution may be that it prevents silent, plausible-looking wrong science."
"For smaller or cheaper models, do not rely on auto-trigger. Invoke the skill explicitly."
scientific-agent-skills as a benchmark fixture, which means future version updates will have externally validated performance data attached. That's a structural improvement over the zero-benchmark situation from June 1."Skills are the SOPs of the AI era — codified expertise that any agent can use, regardless of the underlying model or framework."
uv package manager, macOS/Linux/WSL2. 3npx skills add K-Dense-AI/scientific-agent-skillsgh skill install K-Dense-AI/scientific-agent-skills --agent cursor|claude-code|codex|geminiUse the pyopenms skill to analyze this LC-MS/MS data file.
Identify all adducts and generate an annotated spectrum plot./literature-review transformer-based protein structure prediction 2022–2025,
PRISMA-compliant output, Nature citation styleUse the bioinformatics-pipeline skill followed by scientific-visualization
to process this RNA-seq count matrix and produce publication-ready volcano plots."It is not a magic button that automatically does science after installation."
uv, and the right agent client configuration add up to 1–4 hours if you're starting from scratch. The install command is one line; the environment it drops into is not.| Repository | K-Dense-AI/scientific-agent-skills 3 |
| License | MIT 3 |
| Latest version | v2.53.0 (June 23, 2026) 3 |
| Stars / forks / commits | 29,300 stars · 3,000 forks · 537 commits 3 |
| Skills | 147 across 18 scientific domains 3 |
| Supported agents | Claude Code, Cursor, Codex, Pi, Google Antigravity, OpenClaw, NemoClaw, Hermes, and any Agent Skills-standard client 3 |
| Prerequisites | Python 3.13+, uv, macOS/Linux/WSL2 3 |
| Install | npx skills add K-Dense-AI/scientific-agent-skills 3 |
| Benchmark (strong model) | 96% → 100% task success, 92% API error reduction 1 |
| Benchmark (weak model) | 74% → 100% when explicitly invoked; 8% auto-trigger rate without explicit call 1 |
| SkillsBench 1.1 lift | +16.6 pp average across model fleet 2 |
| Total installs | 66,300+ (Claude Marketplaces) 7 |
| Author | Timothy Kassis (@TimothyKassis), K-Dense Inc. CTO 3 |
claudemarketplaces.com
knightli.com
이 콘텐츠는 채널이 자동으로 생성했습니다. 한 문장이면 Neodrop이 당신을 위해 계속 만들어 냅니다.